Get local news, offers & more… Cultural Solidarity Fund (CSF), a coalition of cultural institutions administered by the IndieSpace/Indie Theater Fund with leadership from LEIMAY was organized during the pandemic to provide both solidarity and support to the arts community at a time when its very survival was in question. The coalition started with just an idea in spring 2020 but has announced it has now granted $1,005,000 to 2,010 arts workers in need in the form of $500 microgrants for housing, food, medical expenses, and family care. When the COVID19 pandemic hit NYC, 66% of arts workers lost their only source of income. Like many creative fields, they had little to no reserves to support them during a crisis, revealing the lack of any financial safety net comparable to those afforded to people in other professions. A group of artists and institutions, each facing their own challenges, responded to this crisis by forming the Cultural Solidarity Fund (CSF) to raise essential funds for arts workers facing financial insecurities. Members of the CSF understood that these workers represent the backbone of our cultural ecology, and organized for collective action. “We are so thrilled to have been able to put solidarity to action and provide such significant resources to our community during this incredibly difficult time.” said Randi Berry, Executive Director, IndieSpace. “The continued support and desire to uplift our fellow artists during times of need is a true testament to the strength of our community, and what’s possible when we work together.” Support for these artists-in-need remains an important part of our economic revitalization as a whole. According to the State Comptroller’s recent report on the impact of Covid-19, NYC’s creative sector accounts for 13 percent of the city’s total economic output, with 12 percent of all U.S. creative industry jobs located in NYC. While CSF is no longer accepting applications and is focusing on funding as many current applicants as possible right now, both individuals and organizations can still sponsor a grant at culturalsolidarityfund.org. There are 225 artists waiting for support. The Cultural Solidarity Fund, administered by the IndieSpace/Indie Theater Fund with leadership by LEIMAY, provided relief microgrants of $500 to artists and cultural workers – prioritizing Black, Indigenous, People of Color (BIPOC), immigrant, undocumented, disabled/deaf, and trans and gender-nonconforming individuals who have been most severely impacted by the pandemic both epidemiologically and economically. Cultural Solidarity Fund grants were open to artists and cultural workers who resided within the five boroughs of New York City on April 1, 2020. These include: Actors, Administrators, Choreographers, Custodians, Curators, Dancers, Designers, Directors, Drivers, Dramaturgs, Front of House, Guards, Janitors, Installers, Musicians, Playwrights, Production Managers, Receptionists, Technical Crew and Art Handlers, Stage Managers, Teaching Artists, Technicians, Ushers, Visual Artists, and anyone working in any capacity in the arts/cultural non-profit or community-based sector. The Cultural Solidarity Fund aims to direct funds specifically to artists and cultural workers in most urgent need of assistance with rent, groceries, medical expenses, and other essential services. In the spirit of coalition and resource sharing, NYC arts and cultural organizations of all sizes and structures come together to grow a joint Cultural Solidarity Fund. This coalition both envisions and provides an interim model of a sustainable economic system for cultural workers in NYC. In the absence of a safetynet, over 250 organizations and individuals have helped CSF raise $1,069,504, including generous donations from The Andrew W. Mellon Foundation ($550,000), The Howard Gilman Foundation ($100,000) and The Tow Foundation ($50,000). The current Organizing Group of the Cultural Solidarity Fund emerged out of the Culture@3 daily calls and comprises 17 organizations that meet weekly in collective decision making and idea implementation processes. These organizations include: Dancewave, Dance Parade, Caribbean Cultural Center African Diaspora Institute, Elevator Repair Service, HERE, Hi-Arts, LEIMAY, Mark Morris Dance Group, National Dance Institute, NYC & Company, New Yorkers for Culture and Arts, Performance Space New York, Theater for a New Audience, The Bushwick Starr, and IndieSpace/The Indie Theater Fund with additional support from ADVANCE/MORE Opera and Bronx Arts Ensemble. As of April 2022, sponsoring organizations include: 651 ARTS, Apollo Theater, Artists Space, Arts House School, Asian American Arts Alliance, Barre Vida, Broadway Cares / Equity Fights Aids, Bronx Council on the Arts, Brooklyn Arts Exchange, Brooklyn Ballet, Caribbean Cultural Center African Diaspora Institute, Cherry Lane Theatre, Children Museum of Arts, Dance/NYC, Dance Theatre of Harlem, Dancewave, Dance Parade, Danspace Project, Danza Espana, Elevator Repair Service Theater, Flushing Council on Culture & The Arts, Foundation for Contemporary Arts, Ford Foundation, Friends of The Anthropologists, Frigid NY, Fund for the City of New York, Gibney Dance, Harlem Stage, HERE Arts Center, Hi-ARTS, Hispanic Society of America, Ice Theatre of New York, ID Studio Theater, International Arts Relations, Irish Repertory Theatre, Issue Project Room, Jack Arts, JP Morgan – The Renate, Hans and Maria Hofmann Trust, Kinesis Project dance theatre, La MaMa, League of Independent Theater, LEIMAY, Lifetime Arts, Louis Armstrong House Museum, Mabou Mines Development Foundation, Mark Morris Dance Group, MCC Theater, Meredith Monk/The House Foundation for the Arts, Movement Research, Museum of Contemporary African Diasporan Arts, National Black Theatre, National Dance Institute, New York Foundation for the Arts, New York Live Arts, New York Theatre Workshop,New Yorkers for Culture and Arts, NYC & Company, Path to Rhythm, Performance Space New York, Ping Chong and Company, Primary Stages, Queens Theatre, Rattlestick Playwrights Theater, Rooftop Films, Roundabout Theatre Company, Scrap Paper Pictures, Seven Loaves Inc, Snug Harbor Cultural Center & Botanical Garden, Soho Rep, Stavros Niarchos Foundation, Stefanie Nelson Dancegroup, STREB, Symphony Space, The 52nd Street Project, The Andrew W. Mellon Foundation, The Bushwick Starr, The Caribbean Cultural Center African Diaspora Institute, The Chocolate Factory, The Dance Enthusiast, The Field, The Hispanic Museum & Library, The Howard Gilman Foundation, The Indie Theatre Fund/IndieSpace, The José Limón Dance Foundation, The Joyce Theater, The Kitchen, The Lucille Lortel Theatre Foundation, The New Group, The Playwrights Realm, The Tank, The Tow Foundation, TheaterLab, Theatre for a New Audience, Theatre Projects, Thompson Family Foundation, UrbanGlass, Venturous Theater Fund, Wildlife Conservation Society, What We Do. IndieSpace was established in 2016 to disrupt the ongoing displacement of small theaters and to address systemic inequities in NYC real estate. In 2022, it merged with Indie Theater Fund, an organization focused on a new model for equitable funding for the indie theater community. By contributing a nickel per ticket from their shows to a pot of money for funding, the indie theater community could create a method of self-sustainability and could rethink philanthropy and the process of grant making. Through radically transparent and equitable grants, community resources and advocacy, the Fund supported hundreds of indie theater companies and thousands of individual artists. Now as one organization, IndieSpace has an expanded capacity to serve hundreds more artists on an annual basis, helping individual artists, small companies and indie venues plant roots and grow. In addition to continuing its founding mission of supporting indie venues facing a real estate transition, challenge, or opportunity, and helping them create permanent solutions, IndieSpace continues initiatives started by Indie Theater Fund, such as running a Mental Health program that has served over 200 artists; leading several coalition-based funding programs, including the Cultural Solidarity Fund, the AAPI Transportation Fund, and the DCLA City Artist Corps grants; and providing hundreds of hours of professional development to individuals and small companies that cannot afford those services directly, focused on fundraising, budgeting, marketing, anti-racism and anti-oppression trainings. Since its founding, IndieSpace has: consulted with 70+ venues making real estate decisions, including The Tank, Frigid NY, The Chain, the wild project, Wooster Group, and Classical Theater of Harlem; helped 18 organizations sign new leases; saved seven theaters from being closed or repurposed; created four real estate operation partnerships; walked one venue through the purchase of their new home; and co-established a rehearsal space co-op in a 99-year lease. During Covid, IndieSpace supported over 50 venues navigating their leases by helping them stay open, and also collaborated with Indie Theater Fund to provide over $1.5M in relief grants to the indie theater community. For service to the community, IndieSpace received the Ellen Stewart Award and a citation from the City Council of New York. IndieSpace also worked with Community Board 2 to negotiate a permanent space for the cultural community in a new development on Gansevoort Street, which it will co-operate with HERE, the New Ohio and Rattlestick Playwrights Theater. LEIMAY is a grassroots artist-run organization and performing arts ensemble that exists out of the art space called CAVE in Williamsburg, Brooklyn. LEIMAY is an entanglement of art and life. The word LEMAY is a Japanese word that means a moment of light in the darkness or a moment of transition. Initiated by Shige Moriya of Japan in 1996 and joined by Ximena Garnica of Colombia in 2001, LEIMAY is led by artists and holds a regular NYC home season, offers classes to the public, brings in guest artists for workshops, and collaborates with presenting/educational partners across NYC, in upstate NY, and beyond. LEIMAY was incorporated in 2003 by young immigrants (Ximena and Shige) who were passionate about creating a space for art to exist amongst the barriers and challenges they felt as new NYC residents. LEIMAY believes in the circulation of energies, resources, and imagination. For over 25 years, CAVE has been a refuge for immigrants and New Yorkers. LEIMAY, as cited in the book ‘Alternative Histories: New York Art Spaces from 1960-2010’ (published in 2012), is embedded in the aesthetic and social fabric of NYC as a site for experimentation for artists to innovate, perform, and exhibit. From 1996-2006, CAVE opened as a gallery/performance space in Brooklyn, featuring 70 visual art, music, and theatrical performance exhibitions w/300 artists. From 2003-2009, LEIMAY produced four sold-out editions of the New York Butoh Festival with 100 dancers (from Japan, Sweden, Germany, France, Colombia, & the US) for over 8000 audience members. Between 2001-2007, LEIMAY organized 50 butoh workshops attended by 800 dancers, and from 2008-2011, LEIMAY introduced the NY Butoh Kan Training Initiative, which led to the establishment of the LEIMAY LUDUS Training program. Since 2008, LUDUS has served 2500 performers and hosted 65 teaching residencies by theater-makers. In addition to producing original creative performances, LEIMAY currently operates the following community programs: INCUBATOR Program (10-14 artists/year who receive studio space, mentoring, and marketing); LUDUS Training Programs (community classes/intensives taught by LEIMAY Ensemble members and guest artists); OUTSIGHT Presenting Series (performances by local and international theater, dance and performance artists in Community Gardens); LEIMAY spearheads the Cultural Solidarity Fund administered by Indie Space. LEIMAY’s work ranges from sculptural, video, mixed-media, and light installations to photography, training projects, stage performances, and publications, including 22 performances; 9 gallery shows; 6 photograph series; 4 drawing series; and 12 video installation series. Broadway Licensing and Broadway Media to Offer CLUE Scenic Projections Open Jar Studios Acquires Theatrical Printing Service BWAY PRINTING Industry Pro Newsletter: Using the Arts to Fund the Arts? PRETTY WOMAN: THE MUSICAL Tour Recoups Investment Hot stories
Robert LuPone, MCC Theater Founder and Artistic Director, Passes Away at 76 Broadway Grosses: Week Ending 8/28/22 Broadway Grosses: Week Ending 8/21/22 Ryan Vona, Janine DiVita & More to Star in Industry Reading of New Musical CHARLIE HUSTLE BroadwayWorld TV
By Journalist Ms. Jones FRESH MEADOWS – Montgomery’s own David Quinones released his new film, “Love”… Falling Is The Easy Part,” on Saturday, August 27th. Well, he’s actually from Brooklyn (Red Hook Projects), but has been living in Montgomery for the last four years because he likes the “farmland.” David began his filmmaking journey in 2001 creating, alongside Jonathan Hall & IMOK Enterprises, the documentary “Heart Of The Brooklyn Sound.” The documentary shows the shaping of gospel music from the 70s to the early 2000s from its start in Brooklyn. “That’s what started me. The documentary… That’s what sparked my interest into doing film,” said Quinones. Quinones released his new film, “Love… Falling Is The Easy Part” with his partners Clyde Salliey and John Arrington with their production companies Reel Block Pictures, Inc. and Select House Pictures, Inc. It stars Tashan Steven Muldrow as the male lead who plays “Mason.” Amber Miller Moore plays “Brooklyn,” the female lead. “A synopsis of the film [is] two people… fall… in love… That is the easy part… They enjoy one another and then, of course, as time progresses, we start to see everything isn’t as what it seems which is very [much] based in reality. You learn as individuals the things you like, the things you don’t like… Throughout the course of the movie, we have to find out if they make it… if they can get through the obstacles and the things that they’re learning along this journey called love,” said Miller Moore whose character “Brooklyn” is working all of the time and climbing up the corporate ladder and her man “Mason” doesn’t like it as he’s a successor to generational wealth who doesn’t enjoy being alone. No movie is complete without a comedic sidekick. Ashford J. Thomas fills those shoes. He plays “Rich”- another obstacle. “As a bachelor… I think a brother should stay a bachelor. He should be living better than what he is. She’s beautiful and all that stuff. I’m just educating the young. He needs to get out there and spread his wings a little bit,” said Thomas who plays “Rich” who hates on “Mason” and “Brooklyn’s” relationship, encouraging “Mason” to “play the field.” Not only was the film released on Saturday, there was also a red carpet movie premiere at the Bombay Theater in Fresh Meadows. After the premiere there was a Q&A with the cast and crew, a networking party, and an after party. The movie is available in several locations. ““Love… Falling is The Easy Part” is available on Amazon Prime… We’re launching a film app., like Netflix and Hulu… ‘Love…’ is gonna be able to be seen on our app called Reel House Pix,” said Quinones who is also working to get it on Hulu and Tubi. Quinones and his partners have other projects in the works. They are already working on their next project. “We’re going into pre-production on a film called “The Quarantine Hustle’… It’s a comedy… We already have the script… We’re probably going to start casting and doing location scouting,” said Quinones. Actors can look on Actors Access, Backstage, and Casting Networks for casting information. Journalist Ms. Jones
Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript. Advertisement Scientific Reportsvolume 12, Article number: 14743 (2022) Cite this article 2 Altmetric Metrics details This paper highlights the potential of using polarimetric methods for the inspection of plant diseased tissues. We show how depolarizing observables are a suitable tool for the accurate discrimination between healthy and diseased tissues due to the pathogen infection of plant samples. The analysis is conducted on a set of different plant specimens showing various disease symptoms and infection stages. By means of a complete image Mueller polarimeter, we measure the experimental Mueller matrices of the samples, from which we calculate a set of metrics analyzing the depolarization content of the inspected leaves. From calculated metrics, we demonstrate, in a qualitative and quantitative way, how depolarizing information of vegetal tissues leads to the enhancement of image contrast between healthy and diseased tissues, as well as to the revelation of wounded regions which cannot be detected by means of regular visual inspections. Moreover, we also propose a pseudo-colored image method, based on the depolarizing metrics, capable to further enhance the visual image contrast between healthy and diseased regions in plants. The ability of proposed methods to characterize plant diseases (even at early stages of infection) may be of interest for preventing yield losses due to different plant pathogens. Polarimetric instrumentation and methods are of interest in a wide range of applications, as for instance, in astronomy1, atmospheric pollution studies2,3, security and remote sensing4, materials characterization5, biomedicine6, etc. In the case of applications in biophotonics, polarimetric methods have proved to be very useful tools to enhance the image contrast of some organic structures, and/or providing information of certain structures invisible by using regular (non-polarimetric) images. This situation is useful, for instance, for the early detection of some diseases, such as breast cancer7,8, colon cancer9, skin cancer10, or brain cancer11, among others. The above-stated use of polarimetric methods for the study of animal12 or even human8,9,10 tissues is a well-established field of work13 and nowadays continues in constant development. However, the application of polarimetric methods for the study of plant diseases is less common and, in the last decade, there is a growing interest of exploring more complex (and rich in terms of information) polarimetric solutions for applications in plant science. Historically, one of the most widely used polarization-based optical instruments for studying plant structures is the polarimetric microscope14. This instrument allows clear observation of vegetal cells, as centrosome-nuclear complexes in cell division15 or cell suspension culture of plant specimens, as in Picea glauca16. Note that polarimetric features exploited by polarimetric microscopes are dichroism and birefringence, when they are present in samples. For instance, in the case of vegetal structures, dichroism measurements reveal the concentration and spatial organization of some plant organelles such as pigment–protein complexes in plant thylakoid membranes17, chloroplasts18, or quantasomes19. Moreover, the birefringence signature of some structures and macromolecules (for instance, cellulose), allows to study the cell wall composition20, the trichomes structure21 or stomata22. In addition to dichroism and birefringence, depolarization is third polarimetric channel which provide valuable information, although it has been underused in plant characterization. Depolarization is a statistical concept originated by incoherent (temporal or spatial) addition of different light polarizations at the level of the detector, and it can be understood as the degree of polarization disorder (randomness) introduced by a given structure to an input polarization. A suitable tool for studying depolarizing information are Mueller polarimeters23,24, that commonly performs macroscopic analysis of samples. Recent works4,25,26 have shown and discussed the advantages of the depolarization-related observables to enhance image contrast between different plant structures of interest, as midrib, secondary veins, stomata and raphides. This is because their constituent units (cellulose, pectin, water content, among others) present different polarimetric features (retardance or dichroism) and/or different spatial organization, these situations inducing depolarization at macroscopic scale observation. Importantly, depolarization methods have proved to be a very interesting tool for the inspection of plant tissues, independently or in combination to other optical instrumentation, as spectroscopic instrumentation, phase or fluorescent microscopy, among others25,27. A popular figure or metric to study the depolarizing response of plant structures has been the so-called degree of polarization (DoP)26,28, which measures the global change in the polarization degree of a radiation beam after interacting with a given material media, in our case, a plant tissue. This metric has been used, for instance, to estimate the chlorophyll content29, the plant stress30 and for vegetation classification purposes31. The description of the depolarization provided by the DoP, can be refined with the use of the so-called depolarization metric sets as it has been discussed in recent works25,26. A depolarization space is a combination of at least three depolarization-related metrics which are derived from the Mueller matrix, to enhance the image contrast in plant structures4,25,26,32. This is the case of the so-called Indices of Polarimetric Purity (IPPs)33,34, that are connected with the type of polarimetric randomness that a system induces to incident light (they give information of the depolarization anisotropy of the system, i.e., dependence of the input polarization with the resultant depolarization response). When it comes to plant pathogens, virial infections in plant specimens can induce, among others, a decrease in photosynthesis through decreases in chlorophyll efficiency and the disruption of cellular processes. In turn, the damaged cellular metabolism induces the development of abnormal substances which are injurious to the functions of the plant. Furthermore, plant diseases caused by fungi infections are generally described by the pathogen consuming cells or secreting toxins. These may lead to plant tissues containing a mix of both necrotic (dead) and healthy tissues intermingled with the fungal mycelium and may induce to microscopic structural changes35. Those biological modifications in plants caused by virus and fungi may be the origin of depolarization differences between healthy and pathological tissues in plants, but more study must be developed in this research line to connect microscopic biological changes with macroscopic depolarization measures. In the present paper we discuss for the first time, the use of polarimetry and in particular the use of depolarization-related metrics for the visualization and characterization of plant diseases. In particular we show how depolarizing metrics are suitable for the discrimination between healthy tissues and different type and stages of plant infections. The depolarization set of metrics chosen for the present study is the IPPs set because, as stated before, it provides three suitable metrics for plant visualization25,26. In addition to IPPs, we include in the study, for completeness, a second set of depolarizing observables: the so-called Components of Purity (CPs)36. The selection of these two polarimetric spaces is not arbitrary. Whereas the IPPs describe the capability of depolarizing samples to introduce polarimetric randomness to incident light, the CPs provide information of the polarimetric characteristics in samples inducing depolarization (diattenuation, polarizance and retardance). Importantly, these two spaces result in complementary analytical tools, and their combined use completely describes the polarimetric behavior of depolarizing samples37. We further demonstrate the suitability, from a quantitative and qualitative point of view, of such depolarizing spaces for plant diseases visualization enhancement. The analysis is conducted on a set of different botanical specimens showing different disease symptoms and injury stages due to the particular infection of different pathogen agents. Moreover, the potential of the method is strengthened by applying a pseudo-colored approach that helps to magnify the visual contrast between healthy and diseased tissues, or between different stages of the disease. We believe that the use of polarimetric methods, which are non-invasive and non-contact, for the early detection of plant diseases are of interest because they may contribute to prevent large product (and economical) losses in crops38,39. The results presented in this work are related to the study of two leaf specimens suffering different infections: (1) a leaf from a specimen of Medicago sativa (alfalfa), which was found infected with alfalfa mosaic virus (AMV); and (2) a leaf from a specimen of Olea europaea (olive), infected with the fungus Venturia oleaginea (olive leaf spot). We have bounded the study in these two representative samples, for the impact and utility of these specimens in humans, but the main conclusions of this works can be extrapolated to other specimens and infectious agents. In particular, we tested the methods in a set of 73 leaves corresponding to 18 plant diseased species (complete list presented in Supplementary Table S1) and the obtained results agree. In the particular case of study, the M. sativa and the O. europaea, both leaves showed the characteristic lesions of their respective diseases. Medicago sativa showed chlorotic areas surrounding the affected vascular structure (see Fig. 1a and Supplementary Fig. S1a). In turn, O. europaea showed alternating necrotic and chlorotic ring-like lesions surrounding a chlorotic spot (see Fig. 2a and Supplementary Fig. S2a). A description of the plants used on the present study is included in the “Methods” section. Polarimetric images of Medicago sativa leaf used in this study. (a) Picture of the underside part of the Medicago sativa leaf. White square denotes for selected region of interest (ROI) analyzed in remaining images, (b) regular intensity image (M00) of the M. sativa underside ROI and its corresponding polarimetric observables (c) P1, (d) P2, (e) P3, (f) PΔ, (g) P, (h) D and (i) PS for visual comparison. All images correspond to 625 nm illumination wavelength measurements performed at scattering set-up configuration. Yellow arrows correspond to the enhanced vascular structures within the sample, whereas numeric labels (from 1 to 7) indicate the number of chlorotic spots unraveled by means of polarimetric observables. Polarimetric images of Olea europaea leaf used in this study. (a) Picture of the underside part of O. europaea leaf. Black square denotes for selected region of interest (ROI) analyzed in remaining images, (b) regular intensity image (M00) of the O. europaea transmission ROI and its corresponding polarimetric observables (c) P1, (d) P2, (e) P3, (f) PΔ, (g) P, (h) D and (i) PS for visual comparison. All images correspond to 625 nm illumination wavelength measurements performed at transmission set-up configuration. Red and yellow dotted lines correspond to diameter and width measurements for chlorotic spot and necrotic ring, respectively. Yellow arrows indicate the unveiled vascular structures. The potential of polarimetric observables for the characterization of the above-mentioned diseases was studied. The polarimetric and standard intensity images of the M. sativa and O. europaea leaves were taken by means of a complete image Mueller polarimeter working at three different illumination wavelengths (625 nm, 530 nm and 470 nm) in scattering and transmission configurations. In the context of Mueller matrix-derived images, we refer to the standard intensity (i.e., the M00 coefficient) as the non-polarized intensity image due to the fact that it may be interpreted as the image that would have been taken if the sample was illuminated with natural (or unpolarized) light36. A representative example of standard intensity images of samples can be seen in Figs. 1b and 2b. In addition, the detailed description of the polarimeter used in this study can be found in “Methods” section. The results presented here correspond to the underside part of the chlorotic M. sativa and O. europaea leaves measured in scattering and transmission configuration, respectively, at a wavelength of 625 nm because they represent the most relevant and interesting findings in terms of image enhancement using the polarimetric approach. With regards to the illumination channel selection, 625 nm wavelength light penetrates more deeply into the sample than shorter wavelengths (530 nm or 470 nm in our case). In this way, in the case of O. europaea leaf, light is capable to reach the opposite surface of the leaf, providing information about the diseased tissues at all depth levels. Similar behavior occurs for M. sativa leaf, where 625 nm illumination wavelength measurements provide a more accurate description of the lesions. Complementary measurement configurations and their respective obtained polarimetric images are presented in Supplementary Figs. S1 and S2. For the optimal presentation and interpretation of results, we perform a qualitative analysis by comparing the acquired standard intensity images with the polarimetric observables described in “Methods” section. For consistency, the analysis is complemented with quantitative study regarding the polarimetric behavior on different structures of healthy and diseased tissues of samples. We also present a pseudo-colored approach image technique that allows a better visualization of certain healthy and diseased plant structures. By taking advantage of the different depolarizing behavior of biological structures in samples40, we calculated different Mueller matrix-derived polarimetric observables images and we compared them with standard intensity images, for the two specimen leaves analyzed. The selected observables for the analysis of plant samples, which are described in the “Methods” section, provide a complete description of the enpolarizing properties of samples37. In particular, we use the indices of polarimetric purity—P1, P2, P3 -, the components of purity—P, D, PS—and the depolarization index PΔ, for the polarimetric sample description. The obtained results show that the depolarization metrics P1, P2, P3, PΔ and PS clearly manifest an overall enhancement of image contrast and they help to unveil wounded zones or vascular structures that are invisible by using standard non-polarized images. In the following we present two illustrative results. The standard non-polarized transmission images for the M. sativa and O. europaea leaves are shown in Figs. 1b and 2b, respectively. These images were obtained from certain regions of interest (ROIs) in the leaves, indicated with white and black squares in Figs. 1a and 2a, respectively. For comparison purposes, we provide the polarimetric images corresponding to the analyzed polarimetric observables, identifying the ones that provide the largest image enhancement, i.e., the P2 channel for the M. sativa case (Fig. 1d) and the P3 channel for the O. europaea case (Fig. 2e). Importantly, thanks to the contrast enhancement of polarimetric observables in the visualization of the O. europaea image, we can clearly distinguish the necrotic ring and chlorotic spot edges visible in different polarimetric channels (Fig. 2c–i), the necrotic ring being invisible by using non-polarized light intensity images (Fig. 2b). A thorough discussion of the visualization of disease symptoms by using polarimetric means as well as the current sample chlorotic spot and necrotic ring characteristics (diameter and width, red and yellow dotted line in Fig. 2e, respectively) is provided in the “Discussion” section. As a complement to the above-presented images, we quantify the potential of depolarizing metrics to characterize plant pathologies. To do so, we consider the values taken along different cross-sections in the images of the observables shown in Figs. 1 and 2. The particular cross-sections analyzed, highlighted in yellow lines, together with the corresponding values are shown in Fig. 3. Figure 3a and b present a comparison between the values of the classical unpolarized light intensity transmission (M00) metric with the depolarization metrics P2. The values of the cross-section displayed correspond to the transition between a healthy and a chlorotic zone in the M. sativa specimen leaf. The difference between the chlorotic and the heathy areas is hardly visible using the classical M00 observable, while it becomes quite apparent in the P2 metric image. Healthy areas show P2 values around 0.20 ± 0.01 while chlorotic areas show characteristic values around 0.50 ± 0.01. Analogously, Fig. 3c and d show a comparison between the values of the M00 metric and the P3 metric for the O. europaea case. The selected cross-section displays the transition between healthy, necrotic, and chlorotic regions. As in the previous example, the classical M00 observable (Fig. 3c) is not sensitive to all the features in the plant, for instance it does not delimitate the necrotic ring, while the P3 observable (Fig. 3d) clearly discriminates between healthy, necrotic and chlorotic. For instance, the value of P3 for healthy, necrotic and chlorotic areas is around 0.20 ± 0.01, 0.45 ± 0.01 and 0.10 ± 0.01 respectively. Such selectivity results in a clear enhancement of the contrast of P3 images when compared to M00 images. In the example shown in Fig. 3d the necrotic area is clearly visible and appears as a “ring” area delimited between the chlorotic spot and the healthy tissue. The chlorotic area appears as dark spot, whereas the healthy area appears in grey. In contrast, M00 observable shows the chlorotic area as a bright spot with values close to 0.50 ± 0.01, the necrotic ring in dark, typical values of 0.05 ± 0.01, and the healthy area in dark grey with typical values of 0.90 ± 0.01. The difference between healthy and necrotic areas is less net in a M00 image than in a P3 one. Pixel profile comparison for Medicago sativa and Olea europaea. (a) Intensity image at 625 nm of the underside part of M. sativa sample and its corresponding pixel profile, (b) polarimetric purity index P2 and its corresponding pixel profile, (c) intensity image at 625 nm of the underside part of O. europaea sample and the corresponding healthy-necrotic-chlorotic transition pixel profile, and (d) polarimetric purity index P3 and its corresponding pixel profile. The vertical yellow lines on polarimetric images indicate the plotted pixel profile segments. The squared numeric labels for M00, P2 and P3 indicate their respective highest and lowest pixel values within the inspected pixel regions. Red-dotted horizontal lines on plots indicate the diameter of the two chlorotic spots of M. sativa and width measurements for chlorotic spot and necrotic ring of O. europaea. We now further analyze the potential of those observables to discriminate between different typologies (healthy or diseased tissues) of plant structures. To do so, we represent the measured data on different polarimetric spaces41, this leading to a very intuitive visualization of data, and also providing quantitative information of the structures (or tissue types) that may be present in the images of the probed samples. Moreover, the data in each ROI is collected and grouped in what we call a data-cloud which can be used for ulterior statistical data treatment or for graphical representation. For instance, data from a homogeneous region should show close values with little variance, in contrast to data from a heterogeneous region, which should show different values with extended deviation from a mean value. When represented in a graph, data from a homogeneous region may group within a well-defined region, whereas data from an heterogenous region may tend to spread over an irregular volume. Therefore, a graphical representation of data-clouds is a good tool to see at a glimpse the variability and the average values of such data. When a given area in a sample is represented according to various observables, such for instance M00, IPPs, or CPs, then, data clouds corresponding each to a given observable, can be defined. Figure 4 presents the data clouds from selected ROIs (marked in Fig. 4a and d as colored rectangles) of healthy (green), chlorotic (yellow) and necrotic (dark blue) tissue regions of M. sativa (first row in Fig. 4) and O. europaea (second row in Fig. 4) leaves. The size of the corresponding regions of interest were selected depending on the tissue availability of sample to perform a homogeneous selection of healthy and diseased tissue. In particular, healthy and chlorotic tissue regions on M. sativa correspond to a total of 2800 pixels (40 × 70 and 70 × 40 pixels, respectively.) In counterpart, healthy, necrotic and chlorotic ROIs for O. europaea are of 90 × 40 (3600 pixels), 30 × 110 (3300 pixels) and 70 × 40 (2800 pixels), respectively. The data clouds are represented in the IPPs space for the M. sativa (Fig. 4b) and O. europaea (Fig. 4e), as well in the CPs space (Fig. 4c,f, respectively). Whereas the M. sativa presents two differentiated structures (healthy and chlorotic), the O. europaea presents three differentiated structures (healthy, necrotic and chlorotic). We see as in both plant specimens, and for the two studied polarimetric spaces, the different leaf structures (healthy, necrotic and chlorotic) are clearly differentiated and located in different spatial positions without data mixing (the different type of data -different colored dots- practically do not overlap in any region of the spaces). This trend was also observed when selecting other affected regions and the selected examples in Fig. 4 are representative cases that illustrate the discriminatory potential of polarimetric observables for symptom detection and description. Scatter data plots of healthy and diseased tissue regions of Medicago sativa and Olea europaea. (a) Visual indicative for healthy (green) and chlorotic (yellow) tissue selected region of interest (ROI) for M. sativa, (b) corresponding IPPs space (P1, P2, P3) for healthy and chlorotic data clouds representation, (c) components of purity (P, D, PS) space for healthy and chlorotic data clouds representation, (d) visual indicative for selected healthy (green), chlorotic (yellow) and necrotic (dark blue) tissue ROIs for O. europaea, (e) corresponding IPPs space (P1, P2, P3) for healthy, chlorotic and necrotic data clouds representation and (f) components of purity (P, D, PS) space for healthy, chlorotic and necrotic data clouds representation. In this section, we want to go one step further in plant pathology imaging based on polarimetric observables by using the discussed polarimetric spaces into a pseudo-coloration image method42,43,44. The main idea consists of building a “polarimetric triplet” by selecting, between all the used polarimetric observables, the three of them leading to larger image contrast between healthy and diseased tissues within the inspected sample. Afterwards, each one of the selected polarimetric images of the triplet is associated with a primary color (red, green and blue, respectively), and they are properly combined to build pseudo-colored images providing visual contrast between plant structures of interest. Detailed description of the pseudo-colored approach proposed in this study can be found in Supplementary Sect. 3.3. To achieve an optimal use of the pseudo-color approach, the three observables used to represent the RGB base must be the ones that better discriminate between healthy and wounded areas. The discrimination ability of a given set of observables can be quantified by measuring the differences of these observables when applied to image healthy and wounded areas. To estimate these differences, we performed a Boxplot analysis45,46 for the regions of interest (ROIs) shown in Fig. 4. In Fig. 5 we show the split Boxplot analysis of the Indices of Polarimetric Purity (P1, P2, P3) and the Components of Purity (P, D, PS) for healthy, chlorotic and necrotic tissue regions of M. sativa (Fig. 5a,b) and O. europaea (Fig. 5c,d). Boxplot charts for healthy and diseased regions for Medicago sativa and Olea europaea. (a) Indices of polarimetric purity (P1, P2, P3) boxplot for healthy and chlorotic locations on M. sativa, (b) Components of purity (P, D, PS) Boxplot for healthy and chlorotic locations on M. sativa, (c) Indices of polarimetric purity (P1, P2, P3) Boxplot for healthy, necrotic and chlorotic locations on O. europaea and (d) Purity components (P, D, PS) Boxplot for healthy, necrotic and chlorotic locations on O. europaea. The corresponding healthy, chlorotic and necrotic data distributions are labeled and colored as H (green), C (yellow) and N (dark blue), respectively. Red-dashed lines indicate the locations of median values and illustrate they do not fit within the boxes of different tissue conditions (healthy, chlorotic or necrotic), allowing discrimination. Circles and stars correspond to mild and extreme outlier values, respectively. We can see in Fig. 5 that some observables clearly separate the studied necrotic, chlorotic and healthy tissues. To help readers to evaluate such differences, the median values are highlighted with dotted red lines in the case of largest differences. In the case of the M. sativa, which presents chlorotic and healthy tissues, they are clearly separated by all the IPPs (P1, P2, P3) metrics (Fig. 5a) as well as by the sphericity degree observable, PS (Fig. 5b). In the case of the O. europaea, which presents chlorotic, necrotic and healthy tissues, we see how the same observables (IPPs and sphericity degree PS) are those leading to the larger distances between different tissues (Fig. 5c,d). Under these results, the IPPs as well as the sphericity degree, PS, are considered to be good candidates to implement the pseudo-colored images. Moreover, Boxplot unveils the amount of outlier values (mild and extreme, illustrated as small circles and stars in Fig. 5, respectively) of each data distribution so that the homogeneity of the selected tissue region can be quantified. The current distributions of outliers range from 0.44 to 3.67%. Note that the low percentage of data outliers ensures the homogeneous selection of the tissue conditions (healthy, necrotic or chlorotic). The complete description of outlier values for each analyzed tissue region can be found in Supplementary Table S2. In addition to the above-presented Boxplot analysis, we computed, for each observable in Fig. 5, the distance between medians in yellow and green boxes (chlorotic and healthy, respectively) for the M. sativa case (Fig. 5a,b); and between medians in yellow and green boxes (chlorotic and healthy, respectively) as well as between medians in blue and green boxes (necrotic and healthy, respectively) for the O. europaea case (Fig. 5c,d). The values resulting from such comparison are shown in Table 1. The largest values in Table 1 are highlighted in bold. The median values of each polarimetric observable for selected healthy and diseased regions on both plant species are shown in Supplementary Table S3. By considering the largest differences between means according to Table 1, we selected two triplets of polarimetric observables for the pseudo-colored images construction: one mixing observables of the IPPs and CPs spaces (P2, P3, PS) and another based on the IPPs by themselves (P1, P2, P3). At this stage, for each depolarizing observable selected Pi (where i = 1, 2, 3, S) we set different threshold values, which were derived from the Boxplot analysis (see Supplementary Table S5). These thresholds allowed us to numerically discriminate between different tissue conditions: healthy/chlorotic for M. sativa and healthy/chlorotic/necrotic, for O. europaea, and then, each separated condition is assigned to a primary color (red, green and blue, for chlorotic, healthy, and necrotic tissues, respectively). This process is illustrated by considering particular examples. In the case of the M. sativa, and for a particular observable selected Pi, a first binary image (black–red) is constructed according to pixel values above/below a certain threshold, this image carrying the chlorotic information of the plant. Then, a second binary image (black–green) is similarly constructed carrying the healthy information of the plant. The same procedure is conducted for the for O. europaea, but now, as this specimen presents an extra condition (necrosis), three binary-images are obtained: (1) for the chlorotic content (black–red), (2) for the healthy content (black–green), and (3) for necrotic content (black–blue). Finally, for each studied metric (P1, P2, P3, PS), a final pseudo-colored image is obtained by adding all the contributions as following, where the sub-index i denotes for the particular depolarizing observable (i.e., i = 1, 2, 3, S). Note that in the case of the M. sativa we can consider Pi, Blue-Necrotic = 0 because there is not necrotic content. Importantly, each term in Eq. (1) provides a binary-colored image that has been filtered according to the threshold criteria above-explained. Finally, the content of a full polarimetric triplet, [P1, P2, P3] or [P2, P3, PS], is put together by constructing a linear combination of the pseudo-colored observables implemented according to Eq. (1), so generalized final pseudo-colored images, valid for both M. sativa and O. europaea, are obtained as: As shown in Eqs. (2) and (3), we built two general pseudo-colored functions (corresponding to each of the two selected observables triplets) labeled as #1 and #2. Note that both Eqs. (2) and (3) can be applied for the two plant specimens analyzed: The M. sativa and the O. europaea cases. Note as well that the corresponding weights of each term in Eqs. (2) and (3) were chosen, for simplicity, as the unit, this giving the same weight to all the observables in the triplet, but other weights could be selected to enhance plant structures visualization. In fact, the low values of the selected polarimetric observables in the case of the O. europaea (see Supplementary Table S3) lead to a darkened final image. In such a case, to visually improve the pseudo-coloration approach, the weights were pondered by a factor 2 so that the resulting image turned brighter. A more detailed description of the method proposed to build pseudo-colored functions is provided in Supplementary Sect. 3.3. Figure 6 shows three images that illustrate the effect of pseudo-coloration to enhance the contrast of polarimetric images as well as the interest of this technique to discriminate between different tissues. The figure shows images of classical M00, the representative polarimetric observable P2 and the corresponding pseudo-colored image for M. sativa, and P1 for O. europaea. In particular the first row in Fig. 6 presents the classical non-polarizer transmission (M00, Fig. 6a), the polarimetric purity index P2 (Fig. 6b), and pseudo-coloration resulting from Eq. (2), labeled as #1 (Fig. 6c), for M. sativa leaf sample. The second row in Fig. 6 presents the intensity (Fig. 6d), the polarimetric purity index P1 (Fig. 6e), and pseudo-coloration resulting from Eq. (2), labeled as #1 (Fig. 6f), for O. europaea leaf. We found that the pseudo-colored resulting images for purity-mixed space and isolated purity space both shown very similar results. To not to present redundant information, pseudo-colored images regarding isolated Indices of Polarimetric Purity [Eq. (3)] for M. sativa and O. europaea, are presented in Supplementary Fig. S4d and h, respectively. The images resulting from polarimetric based pseudo-colored processing are excellent in terms of enhanced visual discrimination of healthy/diseased tissues. Detailed analysis of enhanced structures is provided in “Discussion” section. Visual comparison of Medicago sativa leaf: (a) 625 nm intensity image (M00), (b) polarimetric purity index P2, (c) processed image by means of #1 pseudo-coloration; Visual comparison of Olea europaea leaf: (d) 625 nm intensity image (M00), (e) polarimetric purity index P1, (f) processed image by means of #1 pseudo-coloration. The present work highlights the suitability of using polarimetric observables, in particular, the Indices of Polarimetric Purity (IPPs) and the Components of Purity (P, D and PS), for the inspection of plant disease symptoms. In addition, we show that the implementation of a pseudo-coloration image processing method, which is based on the above-mentioned polarimetric observables, is a useful tool to enhance the image contrast between different tissue natures (healthy, necrotic and chlorotic) of botanical samples. Although the potential of depolarizing metrics to characterize different plant pathologies was observed in 18 different species affected by different infection agents, we focus our attention in two specific plant species because of their relevance in agricultural production: alfalfa (M. sativa) and olive (O. europaea). The complete list of the specimens used in this study is presented in Supplementary Table S1. Normal levels of chlorophyll in plant cells are the cause of the characteristic green color of leaves and allow the natural performance of photosynthetic processes, i.e., the synthesis of organic substances like carbohydrates, from the sources of inorganic prime matter and water from soil, CO2 from the atmosphere, and light35. Chlorophyll deficiency in plant foliage triggers the appearance of chlorosis: the lack of green pigments, which leads to the yellowing of tissues, thus hampering the production of nutritional substances. The chlorotic symptoms can appear due to different biotic and abiotic causes, e.g., iron-deficient or alkaline soils (high pH levels), and plant pathogens35, such as the alfalfa pathosystem (AMV) studied here. With regards to the necrotic symptomatology caused by a pathogen infection, it is characterized by the premature death of affected plant cells and the darkening of tissues. The release of pathogen toxins into the plant cell or, additionally, the release of residual components from surrounding dead cells into the intracellular space typically induces the necrotic lesion appearance. In this study we selected the case of necrosis in leaves of O. europaea as an illustrating example. All plant samples were inspected at different wavelengths (625 nm, 530 nm and 470 nm) and measuring configurations (scattering–measuring both the beam and the underside part of the samples, and transmission). However, to not to extent the content of the manuscript, we only discuss here the cases providing the most interesting results in terms of disease symptom visualization. In particular, we consider M. sativa leaves measurements at 625 nm in scattering configuration from the underside, and the O. europaea leaves measured in transmission configuration from the outside. Measurements with additional orientations of the leaves were performed and the corresponding results are summarized in the Supplementary document. Although different optical leaf properties may play an important role in leaves spectral response40, as for example the spectral signature of a leaf, in the specimens studied in this work we have observed that longer wavelengths penetrate more into samples than shorter wavelengths, as they are less affected by scattering processes than shorter wavelengths. This result agrees with discussion provided in Ref.47. In fact, longer wavelengths may carry more information about microstructures present in the bulk of the sample than shorter wavelengths, which are more sensitive to features present in surface in a bulk region near the surface of the leaves. Since chlorosis or necrosis affect the whole bulk of the leaf, a relative long wavelength such as 625 nm provides a more complete picture of the leave than a shorter wavelength and it is for this reason why we chose in the results shown in the main text of the present manuscript. A second reason for our choice, is that chlorophyll has an absorption peak close to 625 nm, and although being weak this particular absorption feature will impact the measured depolarization. In transparent scattering media, photons which contribute to increase the depolarization of light are those who followed a higher number of scattering events before reaching the detector. Those photons are the ones which also show longer paths inside the sample because of the multiple scattering events that they experienced. In absorbing media, photons which account a higher number of scattering events will be preferentially removed from the beam because their paths are longer than those of photons which account a fewer number of scattering events. Therefore, and in general, it can be said that an absorbing media is expected to show less depolarization than analogous transparent media. In the cases discussed here, because of chlorotic and necrotic tissues show reduced chlorophyll concentration, they appear more transparent than healthy ones at 625 nm: reason why depolarization measurements at this particular wavelength may be more sensitive to the presence of chlorophyll (and therefore to the presence of a wounded tissue) than measurements at wavelengths for which chlorophyll does not show any absorption. In the case of the M. sativa leaf (Fig. 1), the polarimetric images; in particular, the ones corresponding to the depolarization-related P1, P2, P3, PΔ and PS observables (Fig. 1c–f,i, respectively); show a significant enhancement of image contrast between healthy and diseased regions when compared to the image of the standard M00 unpolarized transmitted or scattered light intensity. For instance, up to seven chlorotic spots (labeled from 1 to 7 in Fig. 1d) as well as some vascular structures (yellow arrows in Fig. 1c), which are barely observable by standard intensity images (Fig. 1b), are well visible by using polarimetric means. The improvement is further highlighted after comparing the line cross-sections corresponding to the yellow segments in Fig. 3a (regular intensity image, M00 channel) and Fig. 3b (P2 channel), which are taken within healthy and diseased regions in M. sativa sample (diseased spots labeled as 4 and 5 in Fig. 1d). In this sense, considering the maximum and minimum values (peak-to-valley) in the Fig. 3a and b pixel profiles (corresponding to the healthy/diseased tissues regions) we demonstrate a significant increase of the chlorotic symptom visibility. For practical reason we quantify visibility with the figure defined by the following expression: (V = left[ {I_{max } – I_{min } } right]/left[ {I_{max } + I_{min } } right]), where Imax and Imin correspond to the maximum and minimum value of the pixel within the selected segment in the cross-section or the region in the image. In the case discussed here, the visibility features a value of V = (0.41 ± 0.05) for the non-polarized transmission/scattering intensity in Fig. 3a and of V = (0.61 ± 0.03) for the P2 channel in Fig. 3b. Moreover, thanks to the visualization and contrast improvement in polarimetric images, it is possible to measure the width of the diseased regions with a pixel precision limited by the density of pixels in the detector. After calibration we found that one pixel of CCD corresponds to approx. 21 µm in sample. Therefore, the size of the chlorotic regions shown in Fig. 1 are estimated to (34 ± 1) pixel, (730 ± 21) µm, and (38 ± 1) pixel, (820 ± 21) µm, for the first and second chlorotic spots, respectively. On the other hand, note that we can physically interpret the depolarizing behavior of healthy/diseased tissues in M. sativa, considering that the higher the value of IPPs within a particular tissue region, the lower depolarization induces to incident light33. Under this scenario, and according to images in Fig. 1, the chlorotic spots induce fewer depolarizing effects on incident light (high IPPs values) than healthy tissues (low IPPs values) as expected in connection with the previous discussion, because of the reduced absorption of light in the sample due to low levels of chlorophyll. In the case of the O. europaea sample (Fig. 2), it presents a necrotic ring (surrounding a chlorotic spot) hardly visible in non-polarized transmission/scattering images (Fig. 2b). In contrast, polarimetric observables show an increased contrast compared to that of non-polarized transmission/scattering, which provide a more accurate visualization and spatial delimitation of the lesions in the leaf. Moreover, as in the case of M. sativa previously discussed, the use of polarimetric observables unveils the presence of vascular structures (see Fig. 2c–i) non-visible in non-polarized transmission/scattering images. Thanks to this visual improvement, the width of the necrotic ring and the diameter of the chlorotic spot are clearly delimited in polarimetric images (yellow and red dotted lines in Fig. 2e, respectively): (120 ± 1) pixel, (2.580 ± 0.021) mm, for the necrotic ring width and (230 ± 1) pixel, (4.940 ± 0.021) mm for the chlorotic spot diameter. The unveiled vascular structures are pointed out with yellow arrows in Fig. 2d. As in the previous case, this visual discrimination between the diseased and healthy regions within the sample can be quantitatively measured analyzing the pixel profile corresponding to the cross-section represented by yellow segments in Fig. 3c (non-polarized transmission/scattering, M00, image) and Fig. 3d (observable P3). As in the previous case, the cross-sections are selected crossing healthy and wounded areas of the tissue. The chlorotic spot appears as a broad bell-like feature in the non-polarized transmission/scattering cross-section (Fig. 3c) hiding the presence of the necrotic area. In contrast, both the chlorotic spot and the necrotic ring can be distinguished in the cross-section corresponding to the P3 observable (indicated with a red dashed segment in Fig. 3d). To measure the contrast improvement in polarimetric images compared to non-polarized transmission/scattering ones, we take for instance the visibility value for the necrotic ring which is estimated to V = (0.84 ± 0.03) and V = (0.00 ± 0.03) respectively. Furthermore, the necrotic ring registers the highest IPPs values (P3 from 0.15 to 0.56) followed, in descending order, by the healthy leaf lamina (P3 values ranging from 0.13 to 0.31) and the chlorotic spot (e.g. P3 from 0.05 to 0.18). This behavior is physically translated as the necrotic ring inducing fewer depolarizing effects on incident light than healthy regions whose response is, in turn, even less depolarizing than the chlorotic spot. Moreover, note that the left side (leaf lamina) of the O. europaea sample presents similar values as the necrotic ring for some polarimetric channels (see Fig. 2), although such region does not present necrotic-like tissue condition. This result, which can lead to errors in the physical interpretation of the studied structure, is originated by the curvature of the leaf at that left-region, which leads to out of focus measurements. Under this scenario, the collected polarimetric information of a plant region is affected by out of focus polarimetric information corresponding to other plant regions, all this information being mixed with the actual polarimetric information of the studied structure (the left-side of the leave, in this case). This situation highlights the importance of analyzing well-focused images of plant structures, for the visual inspection of diseased plant samples through polarimetric methods. If due to the non-planar surface of the studied sample a proper image focusing of plant structures of interest cannot be simultaneously ensured by a single image shot, a scanning-based imaging approach is recommended. In this work, we focus on the structures at the right-side of the image, as all structures of interest (healthy and wounded tissues) are in a non-curved and well-focused region of the leaf. To better illustrate the interest of the use of polarimetric images to characterize disease symptoms in vegetal tissues, we chose to collect the pixels from the original image and to represent them as a whole in a data cloud figure. In a data cloud figure, pixels corresponding to different regions should group in separate clouds. Non-overlapping clouds indicate that the related regions are well discriminated. On the contrary, either fully or partially overlapping clouds indicate a poor discrimination of nominally different zones. For the present study, we used three-dimensional data clouds with a selection of variables corresponding to the so-called space of Indices of Polarimetric Purity, (P1, P2 and P3), or the space of the Components of Purity, (P, D and PS) (see Fig. 4). In the case of the M. sativa leaf, the data clouds, corresponding to healthy and chlorotic regions (green and yellow squares in Fig. 4a, respectively), are well-discriminated as these two tissue conditions are clearly spatially separated (i.e., practically no data mixing between tissues with different health condition is produced) when represented in either the: IPPs or the CPs space (Fig. 4b,c, respectively). According to the previous, healthy tissue (green squares) tends to group close to the point (0,0,0) which corresponds to higher depolarization, while chlorotic regions tend to group to areas related with less depolarization. Importantly, a stronger depolarization response occurs when the leaves contain non-organized spatial structures (i.e., neither not aligned nor homogeneously distributed) or an important number of microstructures which efficiently scatter light. In this context, either the biological or the structural changes caused by the chlorotic symptoms of infected regions are translated into a less depolarizing effect on incident light when compared with healthy tissues, thus increasing the sensitivity of depolarizing channels to chlorosis detection. In the case of the O. europaea leaf, both depolarizing spaces clearly discriminate between tissue conditions: see how data clouds corresponding to healthy, chlorotic, and necrotic tissue regions in O. europaea (green, yellow and blue squares, respectively, in Fig. 4d) are clearly spatially separated in both IPPs and CPs spaces (Fig. 4e,f, respectively). However, in this case, the chlorotic data (yellow squares) are quite close to the point (0,0,0), therefore indicating that, in this case, chlorotic tissue is more depolarizing than healthy tissue (green squares). The differential depolarizing behavior among the two studied plant species might be related to specific biological characteristics of these species. Regarding the visible chlorotic symptoms that showed a differential polarimetric response between the species, several hypotheses could be considered: first, leaves of M. sativa and O. europaea correspond to non-sclerophyllous and sclerophyllous species, respectively, thus suggesting that different leaf tissue structure could be involved in this differential depolarizing response. Secondly, the chlorophyll-a and chlorophyll-b leaf content is different for each plant species48,49,50 so that the chlorotic symptoms development could be depending on the infected specimen chlorophyllic profile. Additionally, the type of pathogen which caused the chlorotic symptoms on both inspected leaves may also play a role in the differential depolarizing behavior: M. sativa was infected with the alfalfa mosaic virus (AMV) and O. europaea was infected with the fungus V. oleaginea. As far as the infection strategy of viruses is different from fungi, the chlorotic symptoms may manifest in different ways and lead to different physical transformations of plant tissues (thickness, stress resistance, turgor, structure, organization, etc.) of the unitary scatters (cells), therefore resulting in different depolarizing behavior. Moreover, Lanza et al.51 described the morphological changes induced by the infection of V. oleaginea on O. europaea, consisting of a progressive loss and degradation of plastids and chloroplasts in the palisade parenchyma cells. This degradation process leads to a release of cytoplasmic contents of palisade cells at advanced infection stages, which may affect the cuticle by reducing the resistance to water loss and causing stress on leaf tissues. Interestingly, the chlorotic ring on O. europaea might be also caused by fungal phytotoxins, as it was previously reported52. On the basis of the above comments, it is not surprising that a same symptom, i.e. chlorosis, is a consequence of different biological processes in different plants specimens, and therefore, may correspond to different polarimetric responses. In terms of the enhancement of the image contrast of disease symptoms, we have showed that differential information of polarimetric channels can well discriminate between different infection status (i.e., chlorosis, necrosis and healthy tissue). This has been demonstrated for M. sativa and O. europaea, and additionally confirmed in other plant specimens (see the complete list of analyzed specimens in Supplementary Table S1). The pseudo-coloration is the second approach that we chose to better use the contrast enhancement provided by polarimetric images for a visual discrimination of features present in complex scattering media. As previously described, we performed a pseudo-coloration image processing based on triplets of polarimetric observables information codified in three color (R, G, B) channels. At this point we would like to emphasize that we use for the first time the pseudo-colored approach to the analysis of plant disease symptoms. Compared with previous references, the pseudo-coloration approach was improved by conducting two main modifications: (1) the depolarizing observables were not restricted to the IPPs space, but extended to an optimized selection of polarimetric observables within the IPPs and CPs spaces (2) an image filtering process, based on data obtained from a Boxplot analysis (see Fig. 5), was included in the process to largely discriminate between different tissue conditions (healthy/diseased). In this section we bring the discussion to the pseudo-colored images resulting from the triplet P2, P3 and PS, as it is the most suitable one to construct the pseudo-colored functions for M. sativa and O. europaea samples (see Table 1). The pseudo-colored images obtained based on the IPPs lead to similar results, therefore, we invite the interested readers to consult results for IPPs in Supplementary Fig. S4. The final pseudo-colored images for M. sativa and O. europaea, presented in Fig. 6c and f, respectively, demonstrate a visual enhancement of disease symptoms: the extreme different coloration of the chlorotic lesions on M. sativa with respect to the healthy tissue of the leaf lamina (red and green regions on Fig. 6c) leads to a more accurate location of the diseased area. Similar behavior occurs for lesions on O. europaea leaf, where performed pseudo-colorations lead to a better delimitation of the different transitions from chlorotic spot to necrotic ring and healthy tissue of leaf lamina (Fig. 6f). Importantly, we remark here the fact that pseudo-colored images enhance the contrast between different tissue conditions, even more than the performance of isolated imaging of polarimetric observables for M. sativa and O. europaea samples (Fig. 6b,e, respectively). This behavior highlights the suitability of using this image-processing method for biological samples analysis. Particularly, the inspection and estimation of direct lesions, and the characterization and early detection of infection processes on plant tissues. The plant samples used in this work were a leaf of Medicago sativa specimen infected with alfalfa mosaic virus (AMV, which causes wilting or white flecks to necrotic wounds and chlorotic mosaics on leaves) and a leaf of Olea europaea specimen infected with Venturia oleaginea (causal agent of the olive leaf spot). This worldwide spread disease of olive may cause severe tree defoliation and a delay in fruit ripening, among other symptoms, thus leading to relevant yield losses. Native from warmer-temperate climates of south-central Asia, M. sativa belongs to the Fabaceae family (legumes), it is cultivated worldwide for livestock feeding purposes. Despite the toxicity of unsprouted alfalfa, it is also suitable for human consumption in sprout stage or dehydrated. Regarding O. europaea, this species belongs to the Oleaceae family. Although the native species were found in eastern land regions around the Mediterranean Sea, his production is not limited to Mediterranean countries: O. europaea is cultivated in several countries such as South Africa, New Zealand, North and South America, and Australia. In addition to olive oil production and fruit consumption (olives), O. europaea trees are also grown for fine wood manufacturing. T. Garnatje and J. Luque undertook the formal identification of the plant material used in this study. An herbarium voucher of both M. sativa and O. europaea are deposited in the Herbarium of the Botanical Institute of Barcelona (BC-983007 and BC-983006, respectively). All methods were performed in accordance with relevant guidelines and regulation. Different mathematical approaches can be used to describe the polarimetric properties of material media36. Among those, in this work we use the Mueller–Stokes (M–S) formalism, because it is especially suited to deal with partially polarized or unpolarized light, as well as with depolarizing samples53,54. In this approach, polarization of light is described by means of Stokes vectors36 which are composed by four real coefficients., S0, S1, S2 and S3, whose physical interpretation is directly related to the irradiance (total intensity of the beam, S0) and the amount of light which is linearly polarized in vertical and horizontal (S1), 45° and 135° (S2) directions, and right or left-handed circularly polarized (S3). This Stokes formalism leads to a tridimensional representation of light states of polarization in the so-called Poincaré sphere36. In the M–S formalism, polarimetric samples are described by 4 × 4 real coefficient matrices, the so-called Mueller matrices M, which can be understood as the polarimetric transfer functions of polarimetric systems. In addition, polarization of light exiting from a sample (represented by the Sout Stokes vector) is linearly related with the input polarization (represented by the Sin vector) through the Mueller matrix describing a sample. The Mueller matrix of a sample can be experimentally measured by using polarimeters23,24, and it encodes rich polarimetric information: dichroism (diattenuation and polarizance), retardance and depolarization. In the following, the polarimetric principle of a Mueller matrix experimental determination is briefly reviewed, but it is thoroughly described in Ref.54. By means of a complete image Mueller polarimeter, the sample is illuminated by a set of n different controlled light beam polarization states. Accordingly, the polarization of detected (imaged) light emerging from the sample is analyzed. To fully determine the experimental Mueller matrix, at least four input (and analyzed) states are required, which corresponds to 16 radiometric measurements. In the current work, a total of 36 measurements are taken: 6 input states of polarization (generators) and the corresponding 6 analyzers proposed in55, which are used for minimization of noise in radiometric measurements. The mathematical relationship between the set of incident states and the detected ones is given by the sample 4 × 4 Mueller matrix, MS, in the following way: where the detected radiometric measurements are given by the n × n matrix I, SPSG is the 4 × n matrix of input set of polarized light beams where the n columns represent the different Stokes vectors used to illuminate the sample, and the n × 4 matrix SPSA which corresponds to the transposed input Stokes vectors. In this way, by computing the pseudoinverse of SPSG and SPSA, matrices in Eq. (4), the corresponding Mueller matrix MS of the sample can be directly deduced. The complete set of polarimetric properties of the sample can be derived from the experimental Mueller matrix36,53,54. While some metrics (as dichroism) can be directly deduced from the Mueller matrix, other information as retardance and depolarization content are entangled in such a way that a straightforward interpretation is not possible and advanced algebraic methods of analysis, known as matrix decomposition methods are needed56,57,58,59,60. In this way, Mueller matrix M can be written as where m00 entails the non-polarized transmission/scattering of the sample, D and P are 3-dimensional vectors encoding diattenuation and polarizance, respectively, and 3 × 3 submatrix m entangles the retardance and depolarization in a complex way. Whereas D describes the dependency of intensity from emergent light from sample with the input state of polarization, P is related to the capability of said sample to polarize light when illuminated with an unpolarized state36. Regarding the m-submatrix, a commonly used formulism to gather entangled polarimetric properties is the so-called Lu-Chipman decomposition55, which describes any Mueller matrix M as the product of three 4 × 4 pure Mueller matrices (pure depolarizers, retarders and diattenuators) that synthesize well-defined polarimetric observables leading to an easier physical interpretation of the medium. Regarding the depolarization behavior of media, a general quantitative indicator of the overall depolarizing power of the sample, the depolarization index PΔ, is commonly used61,62. Despite of the fact that PΔ is suitable to represent homogeneous depolarization, it does not provide enough information regarding the situations where depolarization actually depends on the state of polarization of the illuminating beam. In this way, it is worth defining the covariance matrix H (associated with Mueller matrix, M)61: where mij represent the Mueller matrix coefficients, σ are the Pauli matrices and (otimes) the Kronecker product. Since Mueller matrices are not Hermitian and thus we cannot ensure they are diagonalizable, we define the so-called indices of polarimetric purity (IPP)33 as a set of three real magnitudes, P1, P2 and P3, directly derived from the covariance matrix H eigenvalues: where λ-eigenvalues are taken in decreasing order as λ0 ≥ λ1 ≥ λ2 ≥ λ3 and IPP values are restricted to 0 ≤ P1 ≤ P2 ≤ P3 ≤ 1. Indices of polarimetric purity define a real tridimensional depolarization space whose interpretation, in addition to how much light is depolarized, is related with different depolarizing mechanisms in the sample. Therefore, the depolarization spaces can be potentially used to discriminate among structures which different depolarization signatures due to their properties and structure. Based on the idea of representing depolarization as the incoherent sum of four pure components33, IPPs correspond to the statistical weights of each component: P1 quantifies the relative portion of pure non-depolarizing component, P2–P1 the relative weight of a bidimensional depolarizer, P3–P2 the relative portion of an equiprobable mixture of three pure components (tridimensional depolarizer) and 1 − P3 is associated with an ideal depolarizer. In consequence, different IPPs values lead to the comprehension of the inherent depolarizing mechanisms of samples34. However, we can define depolarization index, PΔ, by means of IPPs but also eventually splitting depolarization information in the commonly used components of purity P, D and PS: where components of purity P and D are the polarizance and diattenuation vector magnitudes, respectively. The sphericity degree, PS, defines the contribution on depolarization which differs from dichroic origin. Therefore, depolarization index builds a common link between both purity spaces. Pure depolarizing systems are those which entail PΔ = P1 = P2 = P3 = 0, meanwhile pure non-depolarizing media is characterized by PΔ = P1 = P2 = P3 = 1. Recently, it has been demonstrated that the combined use of IPPs and components of purity is an ideal framework to describe depolarizing behavior of samples37. The polarimetric images shown in this work (Figs. 1, 2 and Supplementary Figs. S1 and S2) are gathered from the experimental Mueller matrices of the studied samples, which are acquired by means of a complete image Mueller polarimeter. By taking advantage of the wide spectral response of the light source, which actually covers the visible spectrum (from 400 to 700 nm approx.), we use three different illuminating wavelengths (625 nm, 530 nm and 470 nm) for the consequently inspection of the sample at different depths47. The polarimeter consists of two independent optical systems based on Parallel Aligned Liquid Crystals (PA-LC) retarders, mounted into two compact mobile arms respectively. The Polarization State Generator (PSG) optical design leads to generate any fully polarized state. It is composed by a linear polarizer oriented at 0° with respect to the laboratory vertical, followed by two PA-LC at 45° and 0°. Equivalently, the Polarization State Analyzer (PSA) is capable to detect any state of polarization. The PSA uses 6 optimized polarization analysis states55. In our instrument, both the PSG and the PSA consist of a linear polarizer followed by two Parallel-Aligned Liquid Crystal cells externally controlled by sending different voltages. The combination of both PSG and PSA, are used to record 36 intensity images which are used to deduce the Mueller matrix55. Regarding internal optical set-up, PSA has the same optical elements as PSG but arranged in reverse order. For the acquisition of sample intensity, a CCD camera is placed on the PSA system. To perform the Mueller matrix measurements of biological samples, two main optical configurations are used. By tilting by 34° the PSG with respect to the horizontal laboratory reference and maintaining the PSA at 0° to avoid the ballistic reflection, we perform what we call scattering measurements. Complementary, by placing both PSG and PSA at 90° we perform transmission measurements. In both configurations, we selected from the whole sample, a region of interest (ROI) of 512 × 512 pixels, which corresponds to an area of 1.1 × 1.1 cm2. The detailed information about optical components and the visualization of measurement configurations (Figs. S5 and S6) is shown in Supplementary document. The datasets generated during and/or analysed during the current study are not publicly available due to the conduction of different research studies but are available from the corresponding author on reasonable request. Hough, J. New opportunities for astronomical polarimetry. J. Quant. Spec. Radiat. Transf.106(1–3), 122–132. https://doi.org/10.1016/j.jqsrt.2006.12.012 (2007). ADSCASArticle Google Scholar Kong, Z. et al. A polarization-sensitive imaging lidar for atmospheric remote sensing. J. Quant. Radiat. Transf.271, 107747. https://doi.org/10.1016/j.jqsrt.2021.107747 (2021). CASArticle Google Scholar Deuzé, J. L. et al. Remote sensing of aerosols over land surfaces from POLDER-ADEOS-1 polarized measurements. J. Geophys. Res.106(D5), 4913–4926. https://doi.org/10.1029/2000JD900364 (2021). ADSArticle Google Scholar Patty, C. H. L. et al. Imaging linear and circular polarization features in leaves with complete Mueller matrix polarimetry. Biochim. Biophys. Acta Gen. Subj.1862(6), 1350–1363. https://doi.org/10.1016/j.bbagen.2018.03.005 (2018). CASArticlePubMed Google Scholar Hall, S. A., Hoyle, M. A., Post, J. S. & Hore, D. K. Combined Stokes vector and Mueller matrix polarimetry for materials characterization. Anal. Chem.85(15), 7613–7619. https://doi.org/10.1021/ac401864g (2013). CASArticlePubMed Google Scholar Tuchin, V. V. Tissue Optics: Light Scattering Methods and Instruments for Medical Diagnosis (SPIE Press, 2007). Book Google Scholar Yaroslavsky, A. N., Feng, X., Muzikansky, A. & Hamblin, M. R. Fluorescence polarization of methylene blue as a quantitative marker of breast cancer at the cellular level. Sci. Rep.9(1), 940. https://doi.org/10.1038/s41598-018-38265-0 (2019). ADSCASArticlePubMedPubMed Central Google Scholar Patel, A., Khan, A., Quinlan, R. & Yaroslavsky, A. N. Polarization-sensitive multimodal imaging for detecting breast cancer. Cancer Res.74(17), 4685–4693. https://doi.org/10.1158/0008-5472.CAN-13-2411 (2014). CASArticlePubMed Google Scholar Ivanov, D. et al. Colon cancer detection by using Poincaré sphere and 2D polarimetric mapping of ex vivo colon samples. J. Biophotonicshttps://doi.org/10.1002/jbio.202000082 (2020). ArticlePubMed Google Scholar Le, D. L., Nguyen, D. T., Le, T. H., Phan, Q. H. & Pham, T. T. H. Characterization of healthy and cancerous human skin tissue utilizing Stokes–Mueller polarimetry technique. Opt. Commun.480, 126460. https://doi.org/10.1016/j.optcom.2020.126460 (2021). CASArticle Google Scholar Schucht, P. et al. Visualization of white matter fiber tracts of brain tissue sections with wide-field imaging Mueller polarimetry. IEEE Trans. Med. Imaging39(12), 4376–4382. https://doi.org/10.1109/TMI.2020.3018439 (2020). ArticlePubMed Google Scholar Rodríguez, C. et al. Polarimetric data-based model for tissue recognition. Biomed. Opt. Express12, 4852–4872. https://doi.org/10.1364/BOE.426387 (2021). ArticlePubMedPubMed Central Google Scholar Ahmad, I., Khaliq, A., Iqbal, M. & Khan, S. Mueller matrix polarimetry for characterization of skin tissue samples: A review. Photodiagn. Photodyn. Ther.30, 101708. https://doi.org/10.1016/j.pdpdt.2020.101708 (2020). CASArticle Google Scholar Nguyen, T. H., Kandel, M. E., Rubessa, M., Wheeler, M. B. & Popescu, G. Gradient light interference microscopy for 3D imaging of unlabeled specimens. Nat. Commun.8, 210. https://doi.org/10.1038/s41467-017-00190-7 (2017). ADSCASArticlePubMedPubMed Central Google Scholar Hyman, A. A. Centrosome movement in the early divisions of Caenorhabditis elegans: A cortical site determining centrosome position. J. Cell Biol.9(3), 1185–1193. https://doi.org/10.1083/jcb.109.3.1185 (1989). Article Google Scholar Hakman, I. & Fowke, L. C. An embryogenic cell suspension culture of Picea glauca (White spruce). Plant Cell Rep.6(1), 20–22. https://doi.org/10.1007/BF00269730 (1987). CASArticlePubMed Google Scholar Tóth, T. N. et al. Fingerprinting the macro-organisation of pigment–protein complexes in plant thylakoid membranes in vivo by circular-dichroism spectroscopy. Biochim. Biophys. Acta1857(9), 1479–1489. https://doi.org/10.1016/j.bbabio.2016.04.287 (2016). CASArticlePubMed Google Scholar Zeiger, E. & Hepler, P. Production of guard cell protoplasts from onion and tobacco. Plant Physiol.58(4), 492–498. https://doi.org/10.1104/pp.58.4.492 (1976). CASArticlePubMedPubMed Central Google Scholar Sauer, K. & Calvin, M. Molecular orientation in quantasomes. J. Mol. Biol.4, 451–466. https://doi.org/10.1016/s0022-2836(62)80102-8 (1962). CASArticlePubMed Google Scholar Shtein, I. S. et al. Stomatal cell wall composition: Distinctive structural patterns associated with different phylogenetic groups. Ann. Bot.119(6), 1021–1033. https://doi.org/10.1093/aob/mcw275 (2017). CASArticlePubMedPubMed Central Google Scholar Bischoff, V. et al. Trichome birefringence and its homolog AT5G01360 encode plant-specific DUF231 proteins required for cellulose biosynthesis in arabidopsis. Plant Phys.153(2), 590–602. https://doi.org/10.1104/pp.110.153320 (2010). Article Google Scholar Gao, C. J., Xia, X. J., Shi, K., Zhou, Y. H. & Yu, J. Q. Response of stomata to global climate changes and the underlying regulation mechanism of stress responses. Plant Physiol. J.48(1), 19–28 (2012). CAS Google Scholar Tyo, J. S. Review of passive imaging polarimetry for remote sensing applications. App. Opt.45(22), 5453–5469. https://doi.org/10.1364/AO.45.005453 (2006). ADSArticle Google Scholar Spandana, K. U. et al. Polarization-resolved Stokes–Mueller imaging: A review of technology and applications. Lasers Med. Sci.https://doi.org/10.1007/s10103-019-02752-1 (2019). Article Google Scholar Van Eeckhout, A. et al. Polarimetric imaging microscopy for advanced inspection of vegetal tissues. Sci. Rep.11(1), 3913. https://doi.org/10.1038/s41598-021-83421-8 (2021). ADSCASArticlePubMedPubMed Central Google Scholar Van Eeckhout, A. et al. Depolarizing metrics for plant samples imaging. PLoS ONEhttps://doi.org/10.1371/journal.pone.0213909 (2019). ArticlePubMedPubMed Central Google Scholar Arai, K. & Nishmura, Y. Polarization model for discrimination of broad and needle shaped leaves and estimation of LAI using polarization measurements. Adv. Sp. Res.44(4), 510–516. https://doi.org/10.1016/j.asr.2009.04.026 (2009). ADSArticle Google Scholar Vanderbilt, V. C., Grant, L. & Daughtry, C. S. T. Polarization of light scattered by vegetation. Proc. IEEE73(6), 1012–1024. https://doi.org/10.1109/PROC.1985.13232 (1985). Article Google Scholar Yao, C., Lu, S. & Sun, Z. Estimation of leaf chlorophyll content with polarization measurements: Degree of linear polarization. Radiat. Transf.242, 106787. https://doi.org/10.1016/j.jqsrt.2019.106787 (2020). CASArticle Google Scholar Kharuk, V. I. & Yegorov, V. V. Polarimetric indication of plant stress. Remote Sens. Environ.33(1), 35–40. https://doi.org/10.1016/0034-4257(90)90053-O (1990). ADSArticle Google Scholar Curran, P. J. Polarized visible light as an aid to vegetation classification. Remote Sens. Environ.12(6), 491–499. https://doi.org/10.1016/0034-4257(82)90023-2 (1982). ADSArticle Google Scholar Savenkov, S. N. et al. Measurement and interpretation of Mueller matrices of barley leaves. Quantum Electron.50, 55. https://doi.org/10.1070/QEL17178 (2020). ADSCASArticle Google Scholar San José, I. & Gil, J. J. Invariant indices of polarimetric purity: Generalized indices of purity for NXN covariance matrices. Opt. Commun.284(1), 38–47. https://doi.org/10.1016/j.optcom.2010.08077 (2011). ADSArticle Google Scholar Van Eeckhout, A. et al. Synthesis and characterization of depolarizing samples based on the indices of polarimetric purity. Opt. Lett.42(20), 4155–4158. https://doi.org/10.1364/OL.42.004155 (2017). ADSArticlePubMed Google Scholar Agrios, G. N. Plant Pathology 5th edn. (Elsevier Academic Press, 2005). Google Scholar Gil, J. J. & Ossikovski, R. Polarized Light and the Mueller Matrix Approach (CRC Press, 2016). MATH Google Scholar Van Eeckhout, A. et al. Unravelling physical information of depolarizers. Opt. Express29(23), 38811–38823. https://doi.org/10.1364/OE.438673 (2021). ADSArticlePubMed Google Scholar Savary, S., Ficke, A., Aubertot, J. N. & Hollier, C. Crop losses due to diseases and their implications for global food production losses and food security. Food Secur.4, 519–537. https://doi.org/10.1007/s12571-012-0200-5 (2012). Article Google Scholar Oerke, E. C. Crop losses to pests. J. Agric. Sci.144(1), 31–43. https://doi.org/10.1017/S0021859605005708 (2006). Article Google Scholar Jacquemoud, S. & Ustin, S. Leaf Optical Properties (Cambridge University Press, 2019). Book Google Scholar Van Eeckhout, A. et al. Depolarization metric spaces for biological tissues classification. J. Biophotonicshttps://doi.org/10.1002/jbio.202000083 (2020). ArticlePubMed Google Scholar Rodríguez, C. et al. Indices of polarimetric purity: application in biological tissues. IProceedings of the SPIE, 116460P. https://doi.org/10.1117/12.2578254 (2021). Van Eeckhout, A. et al. Indices of polarimetric purity for biological tissues inspection. In Proceedings of the SPIE, 104971V. https://doi.org/10.1117/12.2290283 (2018). Van Eeckhout, A. et al. Polarimetric imaging of biological tissues based on the indices of polarimetric purity. J. Biophotonics11, e201700189. https://doi.org/10.1002/jbio.201700189 (2017). ArticlePubMed Google Scholar Tukey, J. W. Exploratory Data Analysis (Addison-Wesley, 1977). MATH Google Scholar McGill, R., Tuckey, J. W. & Larsen, W. A. Variations of box plots. Am. Stat.32(1), 12–16. https://doi.org/10.2307/2683468 (1978). Article Google Scholar Mustafa, F. H. & Jaafar, M. S. Comparison of wavelength-dependent penetration depths of lasers in different types of skin in photodynamic therapy. Indian J. Phys.87(3), 203–209. https://doi.org/10.1007/s12648-012-0213-0 (2013). ADSCASArticle Google Scholar Rekik, I. et al. Effects of untreated and treated wastewater at the morphological, physiological and biochemical levels on seed germination and development of sorghum (Sorghum bicolor (L.) Moench), alfalfa (Medicago sativa L.) and fescue (Festuca arundinacea Schreb). J. Hazard Mater.326, 165–176. https://doi.org/10.1016/j.jhazmat.2016.12.033 (2017). ADSCASArticlePubMed Google Scholar Bahloul, N., Kechaou, N. & Mihoubi, N. B. Comparative investigation of minerals, chlorophylls contents, fatty acid composition and thermal profiles of olive leaves (Olea europeae L.) as by-product. Grasas Aceiteshttps://doi.org/10.3989/gya.0102141 (2014). Article Google Scholar Kakade, M. D. & Borse, R. D. Effect of gamma radiation on chlorophyll content of lucerne (Medicago sativa L.). Online Int. Interdiscip. Res. J.9, 2249–9598 (2019). Google Scholar Lanza, B., Ragnelli, A. M., Priore, M. & Aimola, P. Morphological and histochemical investigation of the response of Olea europaea leaves to fungal attack by Spilocaea oleaginea. Plant Path.66(8), 1239–1247. https://doi.org/10.1111/ppa.12671 (2017). CASArticle Google Scholar Yahiaoui, R., Guechi, A., Lukasova, E. & Girre, L. Mutagenic and membranal effect of a phytotoxic molecule isolated from olive leaves parasitized by the fungus Cycloconium oleaginum Cast. Mycopathologia126, 121–129. https://doi.org/10.1007/BF01146204 (1994). CASArticlePubMed Google Scholar Goldstein, D. Polarized Light 2nd edn. (Marcel Dekker, 2003). Google Scholar Chipman, R. A. Polarimetry: Handbook of Optics 2nd edn. (McGraw Hill, 1995). Google Scholar Peinado, A., Lizana, A., Vidal, J., Iemmi, C. & Campos, J. Optimization and performance criteria of a Stokes polarimeter based on two variable retarders. Opt. Express18, 9815–9830. https://doi.org/10.1364/OE.18.009815 (2010). ADSCASArticlePubMed Google Scholar Lu, S. Y. & Chipman, R. A. Interpretation of Mueller matrices based on polar decomposition. J. Opt. Soc. Am. A13(5), 1106–1113. https://doi.org/10.1364/JOSAA.13001106 (1996). ADSArticle Google Scholar Ossikovski, R., De Martino, A. & Guyot, S. Forward and reverse product decompositions of depolarizing Mueller matrices. Opt. Lett.32(6), 689–691. https://doi.org/10.1364/OL.32.000689 (2007). ADSArticlePubMed Google Scholar Gil, J. J. Transmittance constraints in serial decompositions of depolarizing Mueller matrices: The arrow form of a Mueller matrix. J. Opt. Soc. Am. A30(4), 701–707. https://doi.org/10.1364/JOSAA.30.000701 (2013). ADSArticle Google Scholar Ossikovski, R. Analysis of depolarizing Mueller matrices through a symmetric decomposition. J. Opt. Soc. Am. A26(5), 1109–1118. https://doi.org/10.1364/JOSAA.26.001109 (2009). ADSMathSciNetArticle Google Scholar Ossikovski, R. Canonical forms of depolarizing Mueller matrices. J. Opt. Soc. Am. A27(1), 123–130. https://doi.org/10.1364/JOSAA.27.000123 (2010). ADSMathSciNetArticle Google Scholar Gil, J. J. Polarimetric characterization of light and media. Eur. Phys. J. Appl. Phys.40, 1–47. https://doi.org/10.1051/epjap:2007153 (2007). ADSCASArticle Google Scholar Gil, J. J. & Bernabéu, E. Polarization and depolarization indices of an optical system. Opt. Acta33(2), 185–189. https://doi.org/10.1080/713821924 (1986). ADSArticle Google Scholar Download references We acknowledge the financial support of Spanish MINECO (RTI2018-097107-B-C31, and Fondos FEDER); Catalan Government (2017-SGR-001500). J. L. was supported by the CERCA Program, Generalitat de Catalunya. Optics Group, Physics Department, Universitat Autònoma de Barcelona, 08193, Bellaterra, Spain Carla Rodríguez, Mireia Serra i Ribas, Juan Campos & Angel Lizana LPICM, CNRS, Ecole Polytechnique, Institut Polytechnique de Paris, 91120, Palaiseau, France Enrique Garcia-Caurel Botanical Institute of Barcelona (IBB, CSIC-Ajuntament de Barcelona), 08038, Barcelona, Spain Teresa Garnatje Institute of Agrifood Research and Technology (IRTA), 08348, Cabrils, Spain Jordi Luque You can also search for this author in PubMedGoogle Scholar You can also search for this author in PubMedGoogle Scholar You can also search for this author in PubMedGoogle Scholar You can also search for this author in PubMedGoogle Scholar You can also search for this author in PubMedGoogle Scholar You can also search for this author in PubMedGoogle Scholar You can also search for this author in PubMedGoogle Scholar C.R., T.G., J.L., E.G.C, A.L. and J.C. were involved in the study design. T.G. and J.L. collected samples. C.R. and M.S.R. conducted measurements. J.C. and C.R. developed software for data analysis. C.R., M.S.R., E.G.C., J.C. and A.L. analyzed data. C.R., E.G.C., and A.L. drafted the manuscript. A.L. and J.C. supervised the project. All authors reviewed the manuscript. Correspondence to Carla Rodríguez. The authors declare no competing interests. Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. Reprints and Permissions Rodríguez, C., Garcia-Caurel, E., Garnatje, T. et al. Polarimetric observables for the enhanced visualization of plant diseases. Sci Rep12, 14743 (2022). https://doi.org/10.1038/s41598-022-19088-6 Download citation Received: Accepted: Published: DOI: https://doi.org/10.1038/s41598-022-19088-6 Anyone you share the following link with will be able to read this content: Sorry, a shareable link is not currently available for this article.
Exclusive: The 23-year-old has opened up about why he didn’t follow in his famous father’s footsteps Love Love Island? Our Factor 50 email brings all the goss direct to your inbox We have more newsletters Former Love Island star Jack Keating has revealed he is considering taking up a career in music production. The 23-year-old opened up about why he didn't follow in his famous father's footsteps, saying that while he thinks he is a great singer “in the shower”, he doesn’t have the same pipes that his father Ronan has. But the Dubliner – who lives in London with the Boyzone singer and his wife Storm – said he is thinking about a career in music production after his short stint on the ITV dating show. Read More:Jack Keating was 'in the dog house' with famous dad after awkward remark on Love Island He told us: “I always thought about it. I was going to do a course in music production, a long time before Covid, but then Covid happened so it put a halt on everything. Yeah, I was always thinking about it. I love music. I wouldn’t be the best singer in the world but I’m pretty decent in the shower. “I have always loved music. It’s a big part of my family’s life and I would never write it out,” he added. It comes after Jack recently revealed Love Island bosses made his constantly repeat his chat with Gemma Owen about their famous fathers by the pool. He told MailOnline: “I think the biggest, most surprising thing for me is just like how many times they have to build things if they want to get different shots, different angles, different conversations. “The [producers] will ask you again to sit down and have the same conversation again and bring up the same sort of things. That's what I thought was the most surprising thing. The reality star – whose mum is Irish model Yvonne Connolly – continued: '”My chat with Gemma we were having about our dads, they asked us to go sit down and have the same chat about our dads… I was like, "Oh god. When quizzed about reasons behind the request, he replied: “They just wanted to get some more footage pretty much, I'm not really sure, more angles or… I'm not really too sure to be fair. “They were just like, ‘Can you go sit down by the pool again and have a chat about your dads.’ “I was like, "How are we gonna get this conversation back going again?" That was definitely the most surprising thing I think.” READ NEXT: Thousands face energy bill hikes as deadline for meter readings just weeks away Met Eireann pinpoint best day next week with summer to have one last hurrah Irish Game of Thrones star Jack Gleeson marries long-term girlfriend in 'very simple' ceremony in Kerry How dangerous are the little spiders in Irish homes? New study tells where to find them and if you need worry Man, 20s, dies following horror hit and run on Irish road Get breaking news to your inbox bysigning up to our newsletter
BRIDGEPORT, W.Va (WDTV) – West Virginia State Police are investigating one of their own after a complaint was filed against a Bridgeport-based trooper. Jennifer Bruce and her husband, Alex, own a convenience store in Lost Creek. During a WVSP compliance check last week, one of their employees sold tobacco to an underaged customer. That sale resulted in the employee being cited, Bruce told 5 News. Bruce says after the initial citation was issued, she got a call from Trooper Trupo with WVSP’s Bridgeport detachment. She says Trupo told her, as the business owner, she was also due a citation. During the call, Bruce says she asked Trupo if he could mail her the citation, which she claims he agreed to do. She then asked if he could include a copy of the law under which she was being cited. At that point, Bruce says, Trupo told her he would not mail the citation and that she could either go to the Bridgeport detachment office and get it in person or he could issue a warrant for her arrest. Bruce went to the office to get the citation and recorded her interaction with Trupo which she shared with 5 News. Below is an excerpt from the audio recording: Bruce: “By chance can we talk first?” Trooper: “Yeah.” Bruce: “Like, your threat to get a warrant for my arrest.. wait, let me speak first. Please don’t interrupt.” Trooper: “Sit down. Sit down. Be quiet. You’re going to take the citation and you’re going to walk out of here with a citation. If you want to run your mouth go ahead and talk do you understand me?” Bruce: “You offered to mail it.” Trooper: “Sit down. I told you I would not mail you a ticket. Sit down.” Bruce: “Why are you speaking to me like this?” Trooper: “Do you want to go to jail right Godd— now? Sit down, shut your f—ing mouth and you will take a ticket and you will leave this f—ing office. Do you understand? You have two options. One, take a ticket, or you go with charges to North Central Regional Jail. Shut up. Sit down. Shut up. I will not hear any more out of your f—ing mouth.” “I was just in shock that he had exploded like that,” Bruce said. “It wasn’t just being cursed at, it was the pure anger, the pure fear that he put into me.” 5 News reached out to the Bridgeport detachment as well as state police officials in Charleston, including the professional standards unit. WVSP officials said they are aware of the recording involving Trooper Trupo and are investigating. They declined to answer further questions, citing the ongoing investigation. Bruce and her husband say they’ve filed a formal complaint against Trupo. She says she wants to make other people aware of how she was treated. “I’m not saying strip this man from his rights to do anything,” she said. “I just want him to learn from this.” Copyright 2022 WDTV. All rights reserved.
Introducing a list of the Kenyan and Ugandan changemakers that are channeling a future-facing sound into the East African music scene using global inspiration and new technologies. There is a silent renaissance happening in East Africa among communities of artists that are creating space for alternative and experimental musical aesthetics that have historically been desperately lacking support in the region. In Kampala and Nairobi, a new scene of experimental musicians has emerged, inspired by the success of movements like alté, NuNariobi, and Kumasi drill and empowered by digital tools that allow them to create their own platforms and connections, evading the hand of gatekeepers and other institutional bottlenecks. Here is a list of eight producers from Kenya and Uganda that are helping usher in this new dawn in East African music. Chxf Barry is part of a community of creatives that are working to bring Ugandan music into the future. In a musical landscape that at times feels stagnant and unwelcome to experimentation, Barry works untapped gems in the country to develop their styles to innovative, sonically distinct, world-class sounds. One such example, is the new life that he breathed into the remix of Ill Gee’s Arresting Food by elevating the groove, adding a spacy sample, and introducing mwami’s brooding vocals into the mix with an energized verse. It took the already locally popular tune to the farther heights. “My approach is very groove driven. I love a good groove, but also while leaving space for an artist to do their thing. Space is extremely important to me.” He works out of a studio called Villagio nestled in Kampala’s Kiwatule neighborhood where many of Uganda’s new wave scene record; Cxnrvd, Joshua Baraka, Niwe Akeine, to name a few. For many in East Africa, SOULCHYLD (aka MAUIMØON) seemed to come out of nowhere. The previously faceless young Ugandan producer and songwriter has lived everywhere from Zambia and Eritrea to the UK and Toronto. He has amassed a large and loyal soundcloud following (16k strong) for his songs, beats, and flips that fuse influences from across the black diaspora like baile funk, soul, and afrobeat while maintaining a sensual r&b essence. Since moving back to his hometown Kampala during the pandemic, Soulchyld has dived into the scene with intention securing collaboration with rising stars like Joshua Baraka, mau from nowhere, and Cxnrvd with more on the way. Inspired by the melodics of The Neptunes and the percussion of Timbaland, SOULCHYLD aspires “to create unique grooves that others can identify as [his] sound”. Coco Em is one of Nairobi’s most prolific DJs, having made her name for championing East African house music at some of the region’s most prominent stages in the region and beyond; Uganda’s Nyege Nyege Festival, France’s Transmusicales, Boiler Room’s True Music stage in Nairobi to name a few. In February of this year, fueled by the same desire to unite people on the dancefloor, she released her first wholly self-produced EP titled Kilumi. The 7-song EP is an experiment in self-discovery, sampling and drawing from traditional vocals and rhythms of Coco’s Kamba heritage and fusing them with entrancing electronic synths and percussion. She is joined by guest vocalist ilke Sisian, Wuod Baba, Kisiva, and Labdi who deliver dissonant chants and spitfire raps in various local languages. Nairobi’s prodigal son of connoisseur rap. This young polymath came up in the ranks of the Xprso. family as a frequent guest rapper, most notably on 2020’s Gweztapes, an avant-garde opus of swahili-sheng rap with stoned atmospheric production from MR. LU*. Luko has since embarked on his own production journey and though only having started a year and a half-ago, his sound is distinct and respected among his community; lo-fi, atmospheric goodness that is somehow also danceable. After a year of dropping lo-fi flips of popular Kenyan hits on Soundcloud, he made his official debut on streaming in March with Maya Amolo’s infectious r&b earworm “Can’t Get Enough”, a single off her upcoming album which Luko produced most of. mwami is an experimental dance musician from Kampala whose moody electronic sound draws from influences as divergent as Sampha, Chanel Tres, and Sam Fan Thomas. He takes producing back to its roots, before the ubiquity of digital technology conflated the art of production with beatmaking in the minds of many. While he is proficient at the various DAWs, his is a more hands-on collaborative approach that often utilizes the artists he is in community with to create a musical world that is distinctly his own. “In my experience, [production] starts when the raw idea or sketch inspires what’s missing or what is unnecessary and in turn, carves a way forward, until all the missing pieces and feelings exist in the work” he says. This “is almost always an instinctive thing and not a technical one”. When audio-visual production team Xprso. released their EP Nairobass in March of 2020, it reinvigorated listeners and musicians alike – to hear such a masterfully executed contemporary afro-fusion project with an unmistakable Kenyan essence from its production to its references and lyrical content. Though it was their official debut, the collective, which consists of Mars Maasai, Mr. LU*, RVMP, and BBYY, has been dropping remixes, edits, and instrumentals on SoundCloud since 2016. They have also been an essential arm in the musical journeys of emcees Chevy Kev, Lukorito, and tg.blk; whom they frequently collaborate with on singles and joint albums. Their lush and playful records are based on the main pillars of hip-hop, reggae, and dancehall music, bass, dirty drums, dark vocals, and sampling. KNVL is a champion of African dance music in all its forms. The handful of songs and flips he has released thus far, range from afro-house, to trap and baile. Though he has yet to release a full body of work, he is an extremely prolific DJ, regularly playing shows and festivals across Kenya, curating guest mixes for the likes of Soulection and Mercury Loading, and now hosting monthly radio show for Balamii and Ghana’s Oroko Radio where he features his original production alongside favorites from across the African diaspora. “My radio work is all centered around my love for African music and the belief that we as Africans need to take control of the narrative that surrounds it.” Sharon Onyango-Obbo is a Ugandan, Kenyan-based record producer and mix engineer. A formidable trailblazer of audio production in Nairobi, she describes her sonic palette as a blend of safari trap and r&b. Hers is a holistic approach to production, almost like project management; guiding and holding the artist accountable to the goals of a project beyond contributing musically to the track. As an engineer first and foremost, mixing is also central to her compositional process; making sure that her sounds are clean, perfect, and in correct balance with each other. She is perhaps best known for her engineering and project management work on Sol Generation hits like “Nairobi” and “Birthday Song”. Sharon’s mission is to “shape Africa’s socio-economic landscape using music production, technology and innovation as agents of development”.
Located in Glendale, Ariz., The Lotus features 286 apartments, a swimming pool, hammock garden, business center and secured bike storage. GLENDALE, ARIZ. — American Landmark Apartments has purchased Cabana 99th, a multifamily community located at 10000 W. Missouri Ave. in Glendale. The buyer has renamed the 286-unit property as The Lotus. Built in 2022, The Lotus features studio, one- and two-bedroom floor plans ranging from 468 square feet to 828 square feet. Apartments offer dishwashers, full-size washers/dryers, large energy-efficient windows, stainless steel appliances, nine-foot ceiling heights, wood-style vinyl plank flooring and workstation desks in select units. Community amenities include a swimming pool, electric vehicle charging stations, a hammock garden, keyless entry locks, monthly resident events, a business center, multi-purpose lawn, outdoor fitness space, secured bike storage and outdoor barbecue area. Terms of the transaction were not released.
A – A + Updated At: Aug 30, 2022 02:48 PM (IST) Sparsh Shah. Photo credit: Twitter PTI Surat, August 30 Wheelchair-bound singing prodigy Sparsh Shah, who shot to fame for singing the Indian national anthem at a ‘Howdy Modi’ event in the United States three years back, has now taken admission in Boston’s Berklee College of Music, known for producing Grammy Award winners. “Obstacles cannot hold me back,” Sparsh, 19, who was born with nearly 30 fractured bones because of osteogenesis imperfecta, a genetic disorder also known as brittle bone disease, told PTI. A child born with this disorder has soft bones that would break easily. “Due to the medical condition, my lung capacity is less than normal people. Hence, I have to work hard towards breath control,” Sparsh said. “But, such obstacles can never hold me back. My aim is to be self-sufficient and also collaborate with other musicians. That is why I am learning various skills such as singing, songwriting and music production,” he said. He is now pursuing a four-year bachelor course in interdisciplinary music from Berklee College of Music in Boston, Massachusetts. Sparsh, who was fond of singing since childhood, started learning Hindustani classical music when he was six. He is learning from music maestros of the Mewati Gharana, associated with the Pandit Jasraj School of Music. Because of his medical condition, Sparsh, who lives with his parents in New Jersey, has opted for an online course and has already completed one year, his father Hiren Shah said. “We are glad that Sparsh has secured admission in the Berklee College of Music, which has produced nearly 300 Grammy Award winners so far. Due to Sparsh’s medical condition, he has opted for online classes,” his father said. Sparsh has undergone nearly eight surgeries. He now has eight titanium rods and 22 screws inside his body, he said. “He is in a wheelchair because he can neither stand nor walk. His bones would break if he lifts something heavy. Despite such odds, he excelled in his life and secured admission in such a prestigious college,” his proud father said. A – A + SUBSCRIBE TO OUR NEWSLETTER Gross domestic product had expanded by 20.1 per cent in corr… Legal action will also be taken against Jasmine Shah who is … Went to meet agency director Subodh Kumar Jaiswal but not al…
Gurlal Singh has now been posted as an investigating officer… View All
The Tribune, now published from Chandigarh, started publication on February 2, 1881, in Lahore (now in Pakistan). It was started by Sardar Dyal Singh Majithia, a public-spirited philanthropist, and is run by a trust comprising four eminent persons as trustees. The Tribune, the largest selling English daily in North India, publishes news and views without any bias or prejudice of any kind. Restraint and moderation, rather than agitational language and partisanship, are the hallmarks of the paper. It is an independent newspaper in the real sense of the term. The Tribune has two sister publications, Punjabi Tribune (in Punjabi) and Dainik Tribune (in Hindi). Remembering Sardar Dyal Singh Majithia Designed and Developed by: Grazitti Interactive
Episode 4 of Dream Home Makeover Season 3 introduced us to Michelle and Ross, who were looking for a home layout suited to comfortably accommodate their three-generation household. This webisode takes a closer look at how we remodeled their top floor to serve as a primary suite with a bathroom, desk, walk-in closet, and lounge area — allowing Michelle’s parents to move into the home and help with their new little one! Source List from our Partners: Countertops: Ann Sacks, Countertops: Caesarstone, Paint: Clare Paint, Closet: California Closets, Baseboards / T&G: Metrie, Studio McGee Decor Items: Target, Window Treatments: The Shade Store, Plumbing: Kohler / Kallista, Hardware: Rejuvenation, Lighting: Visual Comfort & Co. Before, the upstairs area had two separate guest bedrooms and a bathroom. By combining the two rooms into one, and stealing a few feet from the hallway, we were able to fashion bigger, more functional spaces. The aesthetic style is light, airy, and peppered with greenery, drawing inspiration from their Tulum wedding. Layout choices, like the use of the bay window, and design touches, like hanging pendant lights, helped blend that feel with the Spanish craftsman-style architecture of the home. The floor plan proved tricky, but we were able to separate the space into multiple zones: the sleeping area, the desk area, and the lounge area (which doubled as an attractive background for video calls!). To create a spa-style bathroom, we brought in a custom double vanity with quartz counters, limestone floors, and an extra-large mirror. Then for added luxury, we converted the standard tub to an elegant shower that tapped into Tulum-inspired textures. Next week brings us yet another inside look at a recent episode of Dream Home Makeover, so check back soon! Your email address will not be published.Required fields are marked *
Thanks for visiting ! The use of software that blocks ads hinders our ability to serve you the content you came here to enjoy. We ask that you consider turning off your ad blocker so we can deliver you the best experience possible while you are here. Thank you for your support!
Matthew Gardiner directs the production at the Arlington, Virginia, venue. No Place to Go, the 2022-23 season opener at the Signature Theatre in Arlington, Virginia, begins performances August 30 in the ARK Theatre. Written by Ethan Lipton with music composed by Lipton, Eben Levy, Ian M. Riggs, and Vito Dieterle, No Place to Go is directed by Signature Artistic Director Matthew Gardiner. Bobby Smith (Signature’s She Loves Me) stars with Tom Lagana (Signature’s Chess) as Jonah/guitar, Riggs (Mean Girls) as Duke/bass and guitar, and Grant Langford (Signature’s Ain’t Misbehavin’) as Sal/saxophone. Riggs is also the production’s music director and arranger. In No Place to Go, a dedicated corporate “information refiner” has learned that his company is moving to a remote location. Backed by a jazz band, this permanent part-timer must decide whether to uproot his family’s life or embark on an unknown venture. “Signature is excited to open our season with No Place to Go, an irreverent and moving musical about a man faced with losing his job after 10 years of corporate loyalty,” said Artistic Director Gardiner in an earlier statement. “This show is for anyone who has been unable to fall asleep at night worried that the bottom is going to fall out of the safe cocoon they’ve built. Alongside witty and impassioned songs, our central character narrates the events that led to his unemployment. Hilarious and heartfelt, Ethan Lipton has crafted a show that resonates powerfully today.” “Ten years after my bandmates and I premiered No Place to Go at The Public Theater, I’m thrilled that Signature is producing the show with DC-area performers (including our bassist Ian M. Riggs),” added writer and composer Lipton. “This show tells the story of someone who is struggling to adapt to a changing world. It’s about the dignity of work and the cultural cost of treating people as if they’re disposable. It’s also a musical that puts music and off-kilter comedy at its center, and I couldn’t be more excited to see Bobby, Matthew, and company give it new life at Signature Theatre.” The production also has scenic design by Paige Hathaway, costume design by Frederick P. Deeben, lighting design by Max Doolittle, sound design by Matt Rowe, and casting by Jorge Acevedo. Joey Blakely is the production stage manager, Erica Feidelseit is the production assistant, and Chris Royal is the copyist. Performances continue through October 16. Tickets are available at SigTheatre.org. The Howard Ashman-Alan Menken musical will begin performances at the Massachusetts venue in September. Based on the 2000 film of the same name, the musical features a score by Tony and Pulitzer winner Tom Kitt and a book and lyrics by screenwriter Cameron Crowe. The coming-of-age musical set in 1982 Dublin is based on the 2016 indie film by John Carney. Diane Paulus’ production of the Stephen Schwartz musical reopens August 30 at the Tokyu Theatre Orb. See what’s coming soon, what’s premiering out of town, and what’s in development. Genao takes on the role of Gloria Estefan at the New Jersey theatre after making her Broadway debut in the musical.
The Broadway and West End performer-director will pay tribute to late songwriters Stephen Sondheim, Marvin Hamlisch, and Michel Legrand. The 2nd annual new works festival will present two new musicals and one new play, as well as cabaret, reading, and conversation events. The production will play the Pennsylvania theatre through September 11. Sign up for reopening news, announcements, and exclusive discounts on tickets to your favorite shows! Blocking belongs on the stage, not on websites. Our website is made possible by displaying online advertisements to our visitors. Please consider supporting us by whitelisting playbill.com with your ad blocker. Thank you!
One could only dream about succeeding over night and getting approval from some of the biggest musicians in the world, but that’s what ‘Si Es Trucho Es Trucho’ by Axel Rulay and his friends Farruko and ElAlfa experienced, which leads us to today, where Tiësto released his remix of the track. In summer of 2021, Axel got inspired by phrases repeated by his friends in their group chat, which incentivized him to take it to music production and make a song out of it. In the middle of a lockdown, he decided to get his friends to go to a resort in Punta Cana in Dominican Republic, and invite his Instagram followers to show up. Little did he know, many of his 5000 followers back then showed up, and his song being already finished, it was shared to the group chat of people in the resort, and as he joined his fans, the song was already playing and everyone was dancing to it. Soon enough the track picked up traction, and was shared by many celebrities, including Cardi B and Bad Bunny. A year later the song sits at more than 7 million plays on Spotify alone, and got Tiësto to make his own club twist to already beloved track. As we approach the end of summer season, clubs seek to mark the season with a spectacular finish, and there’s no better way than with new club-ready releases. This remix is another in line of Tiësto’s endeavors to take hits to the club scene, coming two weeks after his mix of ‘Hot In It’ by Charli XCX. With the extra beat, drums and snares, the latin hit is elevated to be played everywhere. The journey to fame for Axel has already begun, and we’re all here for it. https://www.instagram.com/p/ChvCX1Pvfzf/?hl=en Tiësto Remix of ‘Si Es Trucho Es Trucho’ by Axel Rulay is out now by Musical Freedom Label on all platforms and you can check it out here down below:
Image Credit: Ramona Rosales / Provided by Warner Music Group 23-year-old Economics student with a camera and a suitcase in her hands. Fixating on different EDM sub-genres every month. Find me on train stations around Europe as I collect festival tickets and postcards. [email protected]
Black Rooster Audio has released its RO-GOLD vintage plate reverb plug-in which they say is authentically modelled after a hardware classic upgraded with a 24k gold plate for the most luxurious vintage reverberation experience. Here’s the story in their own words… As an effect, reverb remains the quintessential element of modern-day music and sound design production, adding instant depth and atmosphere to any mix. It is perfectly possible to mimic (almost) any space in today’s age of on-tap reverb — all the more so in an ever-widening world of contemporarily convenient effect plug-ins, yet recording with the desired type and amount of reverberation has not always been so straightforward. Step back in time, then, to a different world, when a plate reverb’s original purpose was to emulate reverberant rooms — somewhat crudely, some might say. Saying that, though, when heard through today’s ears, some such solutions have historically become somewhat legendary themselves. The difference between a classic piece of hardware such as the late-Fifties-vintage (mono) EMT 140 Reverberation Unit — a somewhat bulky beast measuring around eight feet by four feet and weighing in at just over quarter of a ton as the world’s first plate reverb (using an electromechanical transducer, similar to that typically found in a loudspeaker, to create vibrations through the plate made from thin steel suspended between tubular steel frames, while pickups, like those found on an electric guitar, would capture those vibrations as they moved across the plate and output them to the console) — or its early-Sixties-vintage stereo successor and most modern, purely impulse response-based plug-ins is immediately apparent to any discerning ear! Enter Black Rooster Audio’s aptly-named RO-GOLD vintage plate reverb plug-in into that modern-day music and sound design production fray; it is inspired by the aforesaid EMT 140 Reverberation Unit and other classic plate reverbs for maximum modelled authenticity, while still remaining an original Black Rooster Audio plug-in at its creative core. Indeed, its plate simulation approach allowed the German company concerned — best known for its beautifully-designed vintage gear emulation plug-ins — to authentically capture the analogue unit’s sound and feel as closely as possible, while still including a few unexpected twists as a beautifully-designed, easy-to-use plug-in package. Put it this way: while engineers have experimented with foil and metal plates to reproduce expensive reverb rooms since the Fifties, installing and controlling a real metal plate reverb has never been easily attainable, nor has it been readily available to every producer or small studio — all the more so when it comes to real gold plates. It is hardly surprising, therefore, that this type of hardware remained exclusive to a select few elite studios — such as California’s classic Capitol Studios and Gold Star Studios, as well as London’s legendary EMI Recording Studios (subsequently renamed Abbey Road Studios) — for a long time. That said, plate reverberation remains audible in many hit songs recorded throughout the Fifties and Sixties, where engineers extensively used this technique. Fast-forward through to today, though, and not only does RO-GOLD authentically take the sound of those sought-after hardware classics to all major DAWs, but it also captures the true tone of a solid gold plate, which would cost a sizeable seven-figure sum to recreate in the contemporary real world. While this 24k gem allows anyone to set up their own retro room sound that is both warm and mellow, it has never been easier to achieve an authentic vintage reverb sound, thanks to the following few, simple-to-use parameters: complete control over the reverb decay time comes courtesy of a physically modelled version of a classic damper control, so users can choose between 10 DAMPER positions; purposefully placed after the plate reverb emulation to enable users to modulate the reverb tail with DAW automation, the classic PRE/DELAY (0ms to 500ms) control determines the amount of delay between the dry and wet signals; following the classic approach of send effects that run in parallel with the mix channel, the DRY/WET (0% to 100%) control keeps the effect in-chain within the plug-in, without having to set up additional send/return loops; and the BASS CUT control applies a high- pass filter that cuts the bass at any desired frequency between 10Hz and 1kHz. All are assisted by a lovely-looking VU meter that provides visual feedback for the reverb settings and signals. DSP (Digital Signal Processing) operations are pipelined using the latest Apple Silicon and SSE2 (Streaming SIMD Extensions 2) instruction sets to ensure that the RO-GOLD vintage plate reverb plug-in benefits from a high-performance operation — despite its very complex computations. Clearly, the GUI (Graphical User Interface) supports high-pixel density on both macOS and Windows systems, providing users of whatever persuasion with the most enjoyable experience, visually, on high DPI (Dots Per Inch) displays. But best of all, RO-GOLD is free for everyone, so it is really worth its weight in gold! Achieving an authentic vintage reverberation is simply a case of signing up for — or into — a Black Rooster Audio account (here: https://blackroosteraudio.com/en/landing/ro-gold) and activating the plug-in via e-mail. Entering Abbey Road Studios in the UK or Capitol Studios in the USA is entirely optional, but the most luxurious and exclusive vintage room effect meeting modern-day features for the highest-quality sound and optimal usability with RO-GOLD is a given! (Gold Star Studios, sadly, is no longer with us, having closed its doors back in 1984, a fire destroying the building several months after the studios themselves were vacated.) Pricing and Availability:
RO-GOLD is available for free as a 64-bit AAX-, AU-, and VST-compatible plug-in for macOS (10.9 or later) and AAX- and VST-compatible plug-in for Windows (7 or later) by signing up for — or into — a Black Rooster Audio account (and activating it via e-mail) from here. Anyone interested in exploring other plate materials and sizes is invited to check out Black Rooster Audio’s best-selling RO-140 vintage plate reverb plug-in, available to purchase — as a 64-bit AAX-, AU-, and VST-compatible plug-in for macOS (10.9 or later) and AAX- and VST-compatible plug-in for Windows (7 or later) — for $129.00 USD via its dedicated webpage, which also includes in-depth information (including some superb-sounding audio demos), here. More information:
Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript. Advertisement Nature Communicationsvolume 13, Article number: 5024 (2022) Cite this article Metrics details Automated, data-driven construction and evaluation of scientific models and theories is a long-standing challenge in artificial intelligence. We present a framework for algorithmically synthesizing models of a basic part of human language: morpho-phonology, the system that builds word forms from sounds. We integrate Bayesian inference with program synthesis and representations inspired by linguistic theory and cognitive models of learning and discovery. Across 70 datasets from 58 diverse languages, our system synthesizes human-interpretable models for core aspects of each language’s morpho-phonology, sometimes approaching models posited by human linguists. Joint inference across all 70 data sets automatically synthesizes a meta-model encoding interpretable cross-language typological tendencies. Finally, the same algorithm captures few-shot learning dynamics, acquiring new morphophonological rules from just one or a few examples. These results suggest routes to more powerful machine-enabled discovery of interpretable models in linguistics and other scientific domains. A key aspect of human intelligence is our ability to build theories about the world. This faculty is most clearly manifested in the historical development of science1 but also occurs in miniature in everyday cognition2 and during childhood development3. The similarities between the process of developing scientific theories and the way that children construct an understanding of the world around them have led to the child-as-scientist metaphor in developmental psychology, which views conceptual changes during development as a form of scientific theory discovery4,5. Thus, a key goal for both artificial intelligence and computational cognitive science is to develop methods to understand—and perhaps even automate—the process of theory discovery6,7,8,9,10,11,12,13. In this paper, we study the problem of AI-driven theory discovery, using human language as a testbed. We primarily focus on the linguist’s construction of language-specific theories, and the linguist’s synthesis of abstract cross-language meta-theories, but we also propose connections to child language acquisition. The cognitive sciences of language have long drawn an explicit analogy between the working scientist constructing grammars of particular languages and the child learning their languages14,15. Language-specific grammar must be formulated within a common theoretical framework, sometimes called universal grammar. For the linguist, this is the target of empirical inquiry, for the child, this includes those linguistic resources that they bring to the table for language acquisition. Natural language is an ideal domain to study theory discovery for several reasons. First, on a practical level, decades of work in linguistics, psycholinguistics, and other cognitive sciences of language provide diverse raw material to develop and test models of automated theory discovery. There exist corpora, data sets, and grammars from a large variety of typologically distinct languages, giving a rich and varied testbed for benchmarking theory induction algorithms. Second, children easily acquire language from quantities of data that are modest by the standards of modern artificial intelligence16,17,18. Similarly, working field linguists also develop grammars based on very small amounts of elicited data. These facts suggest that the child-as-linguist analogy is a productive one and that inducing theories of language is tractable from sparse data with the right inductive biases. Third, theories of language representation and learning are formulated in computational terms, exposing a suite of formalisms ready to be deployed by AI researchers. These three features of human language—the availability of a large number of highly diverse empirical targets, the interfaces with cognitive development, and the computational formalisms within linguistics—conspire to single out language as an especially suitable target for research in automated theory induction. Ultimately, the goal of the language sciences is to understand the general representations, processes, and mechanisms that allow people to learn and use language, not merely to catalog and describe particular languages. To capture this framework-level aspect of the problem of theory induction, we adopt the paradigm of Bayesian Program Learning (BPL: see ref. 19). A BPL model of an inductive inference problem, such as theory and grammar induction, works by inferring a generative procedure represented as a symbolic program. Conditioned on the output of that program, the model uses Bayes’ rule to work backward from data (program outputs) to the procedure that generated it (a program). We embed classic linguistic formalisms within a programming language provided to a BPL learner. Only with this inductive bias can a BPL model then learn programs capturing a wide diversity of natural language phenomena. By systematically varying this inductive bias, we can study elements of the induction problem that span multiple languages. By doing hierarchical Bayesian inference on the programming language itself, we can also automatically discover some of these universal trends. But BPL comes at a steep computational cost, and so we develop new BPL algorithms which combine techniques from program synthesis with intuitions drawn from how scientists build theories and how children learn languages. We focus on theories of natural language morpho-phonology—the domain of language governing the interaction of word formation and sound structure. For example, the English plurals for dogs, horses, and cats are pronounced /dagz/, /hɔrsәz/, and /kæts/, respectively (plural suffixes underlined; we follow the convention of writing phoneme sequences between slashes). Making sense of this data involves realizing that the plural suffix is actually /z/ (part of English morphology), but this suffix transforms depending on the sounds in the stem (English phonology). The suffix becomes /әz/ for horses (/hɔrsәz/) and other words ending in stridents such as /s/ or /z/; otherwise, the suffix becomes /s/ for cats (/kæts/) and other words ending in unvoiced consonants. Full English morphophonology explains other phenomena such as syllable stress and verb inflections. Figure 1a–c shows similar phenomena in Serbo-Croatian: just as English morphology builds the plural by adding /z/, Serbo-Croatian builds feminine forms by adding /a/. Just as English phonology inserts /ә/ at the end of /hɔrsәz/, Serbo-Croatian modifies a stem such as /yasn/ by inserting /a/ to get /yasan/. Discovering a language’s morphophonology means inferring its stems, prefixes, and suffixes (its morphemes), and also the phonological rules that predict how concatenations of these morphemes are actually pronounced. Thus acquiring the morpho-phonology of a language involves solving a basic problem confronting both linguists and children: to build theories of the relationships between form and meaning given a collection of utterances, together with aspects of their meanings. a Serbo-Croatian data (simplified). This language’s morphology is illustrated for masculine and feminine forms. The data motivate a morphological rule which forms the feminine form by appending /a/. b illustrates a counterexample to this analysis: the masculine, feminine forms of clear are /yasan/, /yasna/. These pronunciations are explained by Serbo-Croatian phonology: the sound /a/ is inserted between pairs of consonants at the end of words, notated (varnothing to )a / C_C#. This rule requires that the true stem for /yasan/, /yasna/ is /yasn/. c shows further stems inferred for this data. These stems are stored in the lexicon. We evaluate our BPL approach on 70 data sets spanning the morphophonology of 58 languages. These data sets come from phonology textbooks: they have high linguistic diversity, but are much simpler than full language learning, with tens to hundreds of words at most, and typically isolate just a handful of grammatical phenomena. We will then shift our focus from linguists to children, and show that the same approach for finding grammatical structure in natural language also captures classic findings in the infant artificial grammar learning literature. Finally, by performing hierarchical Bayesian inference across these linguistic data sets, we show that the model can distill universal cross-language patterns, and express those patterns in a compact, human understandable form. Collectively, these findings point the way toward more human-like AI systems for learning theories, and for systems that learn to learn those theories more effectively over time by refining their inductive biases. One central problem of natural language learning is to acquire a grammar that describes some of the relationships between form (perception, articulation, etc.) and meaning (concepts, intentions, thoughts, etc.; Supplementary Discussion 1). We think of grammars as generating form-meaning pairs, 〈f, m〉, where each form corresponds to a sequence of phonemes and each meaning is a set of meaning features. For example, in English, the word opened has the form/meaning (leftlangle /{{{rm{op}}}}{upvarepsilon}{{{rm{nd}}}}/,,[{{{{{bf{stem}}}}}}:{{{{{rm{OPEN}}}}}};{{{{{bf{tense}}}}}}:{{{{{rm{PAST}}}}}}]rightrangle ), which the grammar builds from the form/meaning for open, namely (leftlangle /{{{rm{op}}}}{upvarepsilon}{{{rm{n}}}}/,,[{{{{{bf{stem}}}}}}:{{{{{rm{OPEN}}}}}}]rightrangle ), and the past-tense form/meaning, namely (leftlangle /{{{{{rm{d}}}}}}/,[{{{{{bf{tense}}}}}}:{{{{{rm{PAST}}}}}}]rightrangle ). Such form-meaning pairs (stems, prefixes, suffixes) live in a part of the grammar called the lexicon (Fig. 1c). Together, morpho-phonology explains how word pronunciation varies systematically across inflections, and allows the speaker of a language to hear just a single example of a new word and immediately generate and comprehend all its inflected forms. Our model explains a set X of form-meaning pairs 〈f, m〉 by inferring a theory (grammatical rules) T and lexicon L. For now, we consider maximum aposteriori (MAP) inference–which estimates a single 〈T, L〉–but later consider Bayesian uncertainty estimates over 〈T, L〉, and hierarchical modeling. This MAP inference seeks to maximize P(T, L∣UG)∏〈f, m〉∈XP(f, m∣T, L), where UG (for universal grammar) encapsulates higher-level abstract knowledge across different languages. We decompose each language-specific theory into separate modules for morphology and for phonology (Fig. 2). We handle inflectional classes (e.g. declensions) by exposing this information in the observed meanings, which follows the standard textbook problem structure but simplifies the full problem faced by children learning the language. In principle, our framing could be extended to learn these classes by introducing an extra latent variable for each stem corresponding to its inflectional class. We also restrict ourselves to concatenative morphology, which builds words by concatenating stems, prefixes, and suffixes. Nonconcatenative morphologies20—such as Tagalog’s reduplication, which copies syllables—are not handled. We assume that each morpheme is paired with a morphological category: either a prefix (pfx), suffix (sfx), or stem. We model the lexicon as a function from pairs of meanings and morphological categories to phonological forms. We model phonology as K ordered rules, written ({left{{r}_{k}right}}_{k=1}^{K}), each of which is a function mapping sequences of phonemes to sequences of phonemes. Given these definitions, we express the theory-induction objective as: where [stem: σ; i] is a meaning with stem σ, and i are the remaining aspects of meaning that exclude the stem (e.g., i could be [tense:PAST; gender:FEMALE]). The expression ({mathbb{1}}left[cdot right]) equals 1 if its argument is true and 0 otherwise. In words, Eq. (1) seeks the highest probability theory that exactly reproduces the data, like classic MDL learners21. This equation forces the model to explain every word in terms of rules operating over concatenations of morphemes, and does not allow wholesale memorization of words in the lexicon. Eq. (1) assumes fusional morphology: every distinct combination of inflections fuses into a new prefix/suffix. This fusional assumption can emulate arbitrary concatenative morphology: although each inflection seems to have a single prefix/suffix, the lexicon can implicitly cache concatenations of morphemes. For instance, if the morpheme marking tense precedes the morpheme marking gender, then L([tense:PAST; gender:FEMALE], pfx) could equal L([tense:PAST], pfx) ⋅L([gender:FEMALE], pfx). We use a description-length prior for P(T, L∣UG) favoring compact lexica and fewer, less complex rules (Supplementary Methods 3.4). We infer grammars (teal) for a range of languages, given only form/meaning pairs (orange) and a space of programs (purple). Form/meaning pairs are typically arranged in a stem × inflection matrix. For example, the lower right matrix entry for Catalan means we observe the form/meaning pair ⟨/grizə/,[stem:GREY; gender:FEM]⟩. Grammars include phonology, which transforms concatenations of stems and affixes into the observed surface forms using a sequence of ordered rules, labeled r1, r2, etc. The grammar’s lexicon contains stems, prefixes, and suffixes, and morphology concatenates different suffixes/prefixes to each stem for each inflection. ϵ refers to the empty string. Each rule is written as a context-dependent rewrite, and beneath it, an English description. In the lower black boxes, we show the inferred derivation of the observed data, i.e. the execution trace of the synthesized program. Grammars are expressed as programs drawn from a universal grammar, or space of allowed programs. Makonde and Catalan are illustrated here. Other examples are in Fig. 4 and Supplementary Figs. 1–3. The data X typically come from a paradigm matrix, whose columns range over inflections and whose rows range over stems (Supplementary Methods 3.1). In this setting, an equivalent Bayesian framing (“Methods”) permits probabilistic scoring of new stems by treating the rules and affixes as a generative model over paradigm rows. Phonemes (atomic sounds) are represented as vectors of binary features. For example, one such feature is nasal, for which e.g. /m/, /n/, are +nasal. Phonological rules operate over this feature space. To represent the space of such rules we adopt the classical formulation in terms of context-dependent rewrites22. These are sometimes called SPE-style rules since they were used extensively in the SoundPattern ofEnglish22. Rules are written (focus) → (structural change)/(left trigger)_(right trigger), meaning that the focus phoneme(s) are transformed according to the structural change whenever the left/right triggering environments occur immediately to the left/right of the focus (Supplementary Fig. 5). Triggering environments specify conjunctions of features (characterizing sets of phonemes sometimes called natural classes). For example, in English, phonemes which are [−sonorant] (such as /d/) become [-voice] (e.g., /d/ becomes /t/) at the end of a word (written #) whenever the phoneme to the left is an unvoiced nonsonorant ([− voice − sonorant], such as /k/), written [-sonorant] → [-voice]/[-voice -sonorant]_#. This specific rule transforms the past tense walked from /wɔkd/ into its pronounced form /wɔkt/. The subscript 0 denotes zero or more repetitions of a feature matrix, called the “Kleene star” operator (i.e., [+ voice]0 means zero or more repetitions of [+ voice] phonemes). When such rules are restricted to not be able to cyclically apply to their own output, the rules and morphology correspond to 2-way rational functions, which in turn correspond to finite-state transducers23. It has been argued that the space of finite-state transductions has sufficient representational power to cover known empirical phenomenon in morpho-phonology and represents a limit on the descriptive power actually used by phonological theories, even those that are formally more powerful, including Optimality Theory24. To learn such grammars, we adopt the approach of Bayesian Program Learning (BPL). In this setting, we model each T as a program in a programming language that captures domain-specific constraints on the problem space. The linguistic architecture common to all languages is often referred to as universal grammar. Our approach can be seen as a modern instantiation of a long-standing approach in linguistics that adopts human-understandable generative representations to formalize universal grammar22. We have defined the problem a BPL theory inductor needs to solve, but have not given any guidance on how to solve it. In particular, the space of all programs is infinitely large and lacks the local smoothness exploited by local optimization algorithms like gradient descent or Markov Chain Monte Carlo. We adopt a strategy based on constraint-based program synthesis, where the optimization problem is translated into a combinatorial constraint satisfaction problem and solved using a Boolean Satisfiability (SAT) solver25. These solvers implement an exhaustive but relatively efficient search and guarantee that, given enough time, an optimal solution will be found. We use the Sketch26 program synthesizer, which can solve for the smallest grammar consistent with some data, subject to an upper bound on the grammar size (see “Methods”). In practice, the clever exhaustive search techniques employed by SAT solvers fail to scale to the many rules needed to explain large corpora. To scale these solvers to large and complex theories, we take inspiration from a basic feature of how children acquire language and how scientists build theories. Children do not learn a language in one fell swoop, instead progressing through intermediate stages of linguistic development, gradually enriching their mastery of both grammar and lexicon. Similarly, a sophisticated scientific theory might start with a simple conceptual kernel, and then gradually grow to encompass more and more phenomena. Motivated by these observations, we engineered a program synthesis algorithm that starts with a small program, and then repeatedly uses a SAT solver to search for small modifications that allow it to explain more and more data. Concretely, we find a counterexample to our current theory, and then use the solver to exhaustively explore the space of all small modifications to the theory which can accommodate this counterexample. This combines ideas from counter-example guided inductive synthesis26 (which alternates synthesis with a verifier that feeds new counterexamples to the synthesizer) with test-driven synthesis27 (which synthesizes new conditional branches for each such counterexample); it also exposes opportunities for parallelism (Supplementary Methods 3.3). Figure 3 illustrates this incremental, solver-aided synthesis algorithm, while Supplementary Methods 3.3 gives a concrete walk-through of the first few iterations. To scale to large programs explaining large corpora, we repeatedly search for small modifications to our current theory. Such modifications are driven by counterexamples to the current theory. Blue:grammars. Red: search radius. This heuristic approach lacks the completeness guarantee of SAT solving: it does not provably find an optimal solution, despite repeatedly invoking a complete, exact SAT solver. However, each such repeated invocation is much more tractable than direct optimization over the entirety of the data. This is because constraining each new theory to be close in theory-space to its preceding theory leads to polynomially smaller constraint satisfaction problems and therefore exponentially faster search times, because SAT solvers scale, in the worst case, exponentially with problem size. We apply our model to 70 problems from linguistics textbooks28,29,30. Each textbook problem requires synthesizing a theory of a number of forms drawn from some natural language. These problems span a wide range of difficulties and cover a diverse set of natural language phenomena. This includes tonal languages, for example, in Kerewe, to count is /kubala/, but to count it is /kukíbála/, where accents mark high tones; languages with vowel harmony, for example Turkish has /el/, /t∫an/ meaning hand, bell, respectively, and /el-ler/, /t∫an-lar/ for the plurals hands, bells, respectively (dashes inserted at suffix boundaries for clarity); and many other linguistic phenomena such as assimilation and epenthesis (Fig. 4 and Supplementary Figs. 1–3). The model observes form/meaning pairs (orange) and jointly infers both a language-specific theory (teal; phonological rules labeled r1, r2, …) and a data set-specific lexicon (teal) containing stems and affixes. Together the theory and lexicon explain the orange data via a derivation where the morphology output (prefix+stem+suffix) is transformed according to the ordered rules. Notice interacting nonlocal rules in Kerewe, a language with tones. Notice multiple vowel harmony rules in Sakha. Supplementary Figs. 1–3 provide analogous illustrations of grammars with epenthesis (Yowlumne), stress (Serbo-Croatian), vowel harmony (Turkish, Hungarian, Yowlumne), assimilation (Lumasaaba), and representative partial failure cases on Yowlumne and Somali (where it recovers a partly correct rule set that fails to explain 20% of the data, while also illustrating spirantization). We first measure the model’s ability to discover the correct lexicon. Compared to ground-truth lexica, our model finds grammars correctly matching the entirety of the problem’s lexicon for 60% of the benchmarks, and correctly explains the majority of the lexicon for 79% of the problems (Fig. 5a). Typically, the correct lexicon for each problem is less ambiguous than the correct rules, and any rules which generate the full data from the correct lexicon must be observationally equivalent to any ground truth rules we might posit. Thus, agreement with ground-truth lexica should act as a proxy for whether the synthesized rules have the correct behavior on the data, which should correlate with rule quality. To test this hypothesis we randomly sample 15 problems and grade the discovered rules, in consultation with a professional linguist (the second author). We measure both recall (the fraction of actual phonological rules correctly recovered) and precision (the fraction of recovered rules which actually occur). Rule accuracy, under both precision and recall, positively correlates with lexicon accuracy (Fig. 5c): when the system gets all the lexicon correct, it rarely introduces extraneous rules (high precision), and virtually always gets all the correct rules (high recall). a Measuring % lexicon solved, which is the percentage of stems that match gold ground-truth annotations. Problems marked with an asterisk are allophony problems and are typically easier. For allophony problems, we count % solved as 0% when no rule explaining an alternation is found and 100% otherwise. For allophony problems, full/CEGIS models are equivalent, because we batch the full problem at once (Supplementary Methods 3). b Convergence rate of models evaluated on the 54 non-allophony problems. All models are run with a 24-h timeout on 40 cores. Only our full model can best tap this parallelism (Supplementary Methods 3.3). Our models typically converge within a half-day. SyPhon36 solves fewer problems but, of those it does solve, it takes minutes rather than hours. Curves show means over problems. Error bars show the standard error of the mean. c Rule accuracy was assessed by manually grading 15 random problems. Both precision and recall correlate with lexicon accuracy, and all three metrics are higher for easier problems requiring fewer phonological rules (red, easier; blue, harder). Requiring an exact match with a ground-truth stem occasionally allows solving some rules despite not matching any stems, as in the outlier problem marked with **. Pearson’s r confidence intervals (CI) were calculated with two-tailed test. Points were randomly jittered ±0.05 for visibility. Source data are provided as a Source data file. Prior approaches to morphophonological process learning either abandon theory induction by learning black-box probabilistic models31, or induce interpretable models but do not scale to a wide range of challenging and realistic data sets. These interpretable alternatives include unsupervised distributional learners, such as the MDL genetic algorithm in Rasin et al.32, which learns from raw word frequencies. Other interpretable models leverage strong supervision: Albright et al.33 learns rules from input–outputs, while ref. 34 learns finite state transducers in the same setting. Other works attain strong theoretical learning guarantees by restricting the class of rules: e.g., ref. 35 considers 2-input strictly local functions. These interpretable approaches typically consider 2–3 simple rules at most. In contrast, Goldwater et al.34 scales to tens of rules on thousands of words by restricting itself to non-interacting local orthographic rules. Our results hinge on several factors. A key ingredient is a correct set of constraints on the space of hypotheses, i.e. a universal grammar. We can systematically vary this factor: switching from phonological articulatory features to simpler acoustic features degrades performance (simple features in Fig. 5a, b). Our simpler acoustic features come from the first half of a standard phonology text28, while the articulatory features come from the latter half, so this comparison loosely models a contrast between novice and expert phonology students (Supplementary Methods 3.5). We can further remove two essential sources of representational power–Kleene star, which allows arbitrarily long-range dependencies, and phonological features, which allow analogizing and generalizing across phonemes. Removing these renders only the simplest problems solvable (-representation in Fig. 5a, b). Basic algorithmic details also matter. Building a large theory at once is harder for human learners, and also for our model (CEGIS in Fig. 5a, b). The recent SyPhon36 algorithm strikes a different and important point on the accuracy/coverage tradeoff: it aims to solve problems in seconds or minutes so that linguists can interactively use it. In contrast, our system’s average solution time is 3.6 h (Fig. 5b). SyPhon’s speed comes from strong independence assumptions between lexica and individual rules, and from disallowing non-local rules. These assumptions degrade coverage: SyPhon fails to solve 76% of our data set. We hope that their work and ours sets the stage for future systems that run interactively while also more fully modeling the richness and diversity of human language. If our model captures aspects of linguistic analysis from naturalistic data, and assuming linguists and children confront similar problems, then our approach should extend to model at least some aspects of the child’s linguistic generalization. Studying children (and adult’s) learning of carefully constructed artificial grammars has a long tradition in psycholinguistics and language acquisition37,38,39, because it permits controlled and careful study of the generalization of language-like patterns. We present our model with the artificial stimuli used in a number of AGL experiments38,39,40 (Fig. 6a), systematically varying the quantity of data given to the model (Fig. 6b). The model demonstrates few-shot inference of the same language patterns probed in classic infant studies of AGL. a Children can few-shot learn many qualitatively different grammars, as studied in controlled conditions in AGL experiments. Our model learns these as well. Grammar names ABB/ABA/AAx/AxA refer to syllable structure: A/B are variable syllables, and x is a constant syllable. For example, ABB words have three syllables, with the last two syllables being identical. NB: Actual reduplication is subtler than syllable-copying20. b Model learns to discriminate between different artificial grammars by training on examples of grammar (e.g., AAB) and then testing on either unseen examples of words drawn from the same grammar (consistent condition, e.g., new words following the AAB pattern); or testing on unseen examples of words from a different grammar (inconsistent condition, e.g. new words following the ABA pattern), following the paradigm of ref. 39. We plot log-odds ratios of consistent and inconsistent conditions: (log P({{{{{rm{consistent}}}}}}|{{{{{rm{train}}}}}})/P({{{{{rm{inconsistent}}}}}}|{{{{{rm{train}}}}}})) (“Methods”), over n = 15 random independent (in)consistent word pairs. Bars show mean log odds ratio over these 15 samples, individually shown as black points, with error bars showing stddev. We contrast models using program spaces both with and without syllabic representations, which were not used for textbook problems. Syllabic representation proves important for few-shot learning, but a model without syllables can still discriminate successfully given enough examples by learning rules that copy individual phonemes. See Supplementary Fig. 4 for more examples. Source data are provided as a Source data file. These AGL stimuli contain very little data, and thus these few-shot learning problems admit a broad range of possible generalizations. Children select from this space of possible generalizations to select the linguistically plausible ones. Thus, rather than producing a single grammar, we use the model to search a massive space of possible grammars and then visualize all those grammars that are Pareto-optimal solutions41 to the trade-off between parsimony and fit to data. Here parsimony means size of rules and affixes (the prior in Eq. (10)); fit to data means average stem size (the likelihood in Eq. (10)); and a Pareto-optimal solution is one which is not worse than any other along both these competing axes. Figure 7 visualizes Pareto fronts for two classic artificial grammars while varying the number of example words provided to the learner, illustrating both the set of grammars entertained by the learner and how the learner weighs these grammars against each other. These figures show the exact contours of the Pareto frontier: these problems are small enough that exact SAT solving is tractable over the entire search space, so our heuristic incremental synthesizer is unneeded. With more examples the shape of the Pareto frontier develops a sharp kink around the correct generalization; with fewer examples, the frontier is smoother and more diffuse. By explaining both natural language data and AGL studies, we see our model as delivering on a basic hypothesis underpinning AGL research: that artificial grammar learning must engage some cognitive resource shared with first language acquisition. To the extent that this hypothesis holds, we should expect an overlap between models capable of learning real linguistic phenomena, like ours, and models of AGL phenomena. Few-shot learning of language patterns can be highly ambiguous as to the correct grammar. Here we visualize the geometry of generalization for several natural and artificial grammar learning problems. These visualizations are Pareto frontiers: the set of solutions consistent with the data that optimally trade-off between parsimony and fit to data. We show Pareto fronts for ABB (ref. 39; top two) & AAX (Gerken53; bottom right, data drawn from isomorphic phenomena in Mandarin) AGL problems for either one example word (upper left) or three example words (right column). In the bottom left we show the Pareto frontier for a textbook Polish morpho-phonology problem. Rightward on x-axis corresponds to more parsimonious grammars (smaller rule size + affix size) and upward on y-axis corresponds to grammars that best fit the data (smaller stem size), so the best grammars live in the upper right corners of these graphs. N.B.: Because the grammars and lexica vary in size across panels, the x and y axes have different scales in each panel. Pink shade: correct grammar. As the number of examples increases, the Pareto fronts develop a sharp kink around the correct grammar, which indicates a stronger preference for the correct grammar. With one example the kinks can still exist but are less pronounced. The blue lines provably show the exact contour of the Pareto frontier, up to the bound on the number of rules. This precision is owed to our use of exact constraint solvers. We show the Polish problem because the textbook author accidentally chose data with an unintended extra pattern: all stems vowels are /o/ or /u/, which the upper left solution encodes via an insertion rule. Although the Polish MAP solution is correct, the Pareto frontier can reveal other possible analyses such as this one, thereby serving as a kind of linguistic debugging. Source data are provided as a Source data file. No theory is built from scratch: Instead, researchers borrow concepts from existing frameworks, make analogies with other successful theories, and adapt general principles to specific cases. Through analysis and modeling of many different languages, phonologists (and linguists more generally) develop overarching meta-models that restrict and bias the space of allowed grammars. They also develop the phonological common sense that allows them to infer grammars from sparse data, knowing which rule systems are plausible based on their prior knowledge of human language, and which systems are implausible or simply unattested. For example, many languages devoice word-final obstruents, but almost no language voices word-final obstruents (cf. Lezgian42). This cross-theory common-sense is found in other sciences. For example, physicists know which potential energy functions tend to occur in practice (radially symmetric, pairwise, etc.). Thus a key objective for our work is the automatic discovery of a cross-language metamodel capable of imparting phonological common sense. Conceptually, this meta-theorizing corresponds to estimating a prior, M, over language-specific theories, and performing hierarchical Bayesian inference across many languages. Concretely, we think of the meta-theory M as being a set of schematic, highly reusable phonological-rule templates, encoded as a probabilistic grammar over the structure of phonological rules, and we will estimate both the structure and the parameters of this grammar jointly with the solutions to textbook phonology problems. To formalize a set of meta-theories and define a prior over that set, we use the Fragment Grammars formalism43, a probabilistic grammar learning setup that caches and reuses fragments of commonly used rule subparts. Assuming we have a collection of D data sets (e.g., from different languages), notated ({{{{{bf{X}}}}}^{d}}_{d=1}^{D}), our model constructs D grammars, ({{langle {{{{{{bf{T}}}}}}}^{d},{{{{{{bf{L}}}}}}}^{d}rangle }}_{d=1}^{D}), along with a meta-theory M, seeking to maximize where P(M) is a prior on fragment grammars over SPE-style rules. In practice, jointly optimizing over the space of Ms and grammars is intractable, and so we instead alternate between finding high-probability grammars under our current M, and then shifting our inductive bias, M, to more closely match the current grammars. We estimate M by applying this procedure to a training subset comprising 30 problems, chosen to exemplify a range of distinct phenomena, and then applied this M to all 70 problems. Critically this unsupervised procedure is not given access to any ground-truth solutions to the training subset. This machine-discovered higher-level knowledge serves two functions. First, it is a form of human understandable knowledge: manually inspecting the contents of the fragment grammar reveals cross-language motifs previously discovered by linguists (Fig. 8c). Second, it can be critical to actually getting these problems correct (Fig. 8a, b and middle column of Fig. 8c). This occurs because a better inductive bias steers the incremental synthesizer toward more promising avenues, which decreases its chances of getting stuck in a neighborhood of the search space where no incremental modification offers improvement. a Re-solving the hardest textbook problems using the learned fragment grammar metatheory leads to an average of 31% more of the problem being solved. b illustrates a case where these discovered tendencies allow the model to find a set of six interacting rules solving the entirety of an unusually complex problem. c The metatheory comprises rule schemas that are human understandable and often correspond to motifs previously identified within linguistics. Left column shows four out of 21 induced rule schemas (Supplementary Fig. 6), which encode cross-language tendencies. These learned schemas include vowel harmony and spirantization (a process where stops become fricatives near vowels). The symbol FM means a slot that can hold any feature matrix, and trigger means a slot that can hold any rule triggering context. Middle column shows model output when solving each language in isolation: these solutions can be overly specific (Koasati, Bukusu), overly general (Kerewe, Turkish), or even essentially unrelated to the correct generalization (Tibetan). Right column shows model output when solving problems jointly with inferring a metatheory. Source data are provided as a Source Data file. To be clear, our mechanized meta-theorizing is not an attempt to learn universal grammar (cf. ref. 44). Rather than capture a learning process, our meta-theorizing is analogous to a discovery process that distills knowledge of typological tendencies, thereby aiding future model synthesis. However, we believe that children possess implicit knowledge of these and other tendencies, which contributes to their skills as language learners. Similarly, we believe the linguist’s skill in analysis draws on an explicit understanding of these and other cross-linguistic trends. Our high-level goal was to engineer methods for synthesizing interpretable theories, using morphophonology as a testbed and linguistic analysis as inspiration. Our results give a working demonstration that it is possible to automatically discover human-understandable knowledge about the structure of natural language. Like linguists, optimal inference hinges on higher-level biases and constraints; but the toolkit developed here permits systematic probing of these abstract assumptions and data-driven discovery of cross-language trends. Our work speaks to a long-standing analogy between the problems confronting children and linguists, and computationally cashes out the basic assumptions that underpin infant and child studies of artificial grammar learning. Within phonology, our work offers a computational tool that can be used to study different grammatical hypotheses: mapping and scoring analyses under different objective functions, and studying the implications of different inductive biases and representations across a suite of languages. This toolkit can spur quantitative studies of classic phonological problems, such as probing extensionally-equivalent analyses (e.g., distinguishing deletion from epenthesis). More broadly, the tools and approaches developed here suggest routes for machines that learn the causal structure of the world, while representing their knowledge in a format that can be reused and communicated to other agents, both natural and artificial. While this goal remains far off, it is worth taking stock of where this work leaves us on the path toward a theory induction machine: what are the prospects for scaling an approach like ours to other domains of language, or other domains of science more broadly? Scaling to the full linguistic hierarchy—acoustics, phonotactics, syntax, semantics, pragmatics—requires more powerful programming languages for expressing symbolic rules, and more scalable inference procedures, because although the textbook problems we solve are harder than prior work tackles, full morpho-phonology remains larger and more intricate than the problems considered here. More fundamentally, however, we advocate for hybrid neuro-symbolic models45,46,47 to capture crisp systematic productivity alongside more graded linguistic generalizations, such as that embodied by distributional models of language structure48. Scaling to real scientific discovery demands fundamental innovations, but holds promise. Unlike language acquisition, genuinely new scientific theories are hard-won, developing over timescales that can span a decade or more. They involve the development of new formal substrates and new vocabularies of concepts, such as force in physics and allele in biology. We suggest three lines of attack. Drawing inspiration from conceptual role semantics49, future automated theory builders could introduce and define new theoretical objects in terms of their interrelations to other elements of the theory’s conceptual repertoire, only at the end grounding out in testable predictions. Drawing on the findings of our work here, the most promising domains are those which are solvable, in some version, by both child learners and adult scientists. This means first investigating sciences with counterparts in intuitive theories, such as classical mechanics (and intuitive physics), or cognitive science (and folk psychology). Building on the findings here and in ref. 11, a crucial element of theory induction will be the joint solving of many interrelated model building problems, followed by the synthesis of abstract over-hypotheses that encapsulate the core theoretical principles while simultaneously accelerating future induction through shared statistical strength. Theory induction is a grand challenge for AI, and our work here captures only small slices of the theory-building process. Like our model, human theorists do craft models by examining experimental data, but they also propose new theories by unifying existing theoretical frameworks, performing thought experiments, and inventing new formalisms. Humans also deploy their theories more richly than our model: proposing new experiments to test theoretical predictions, engineering new tools based on the conclusions of a theory, and distilling higher-level knowledge that goes far beyond what our Fragment-Grammar approximation can represent. Continuing to push theory induction along these many dimensions remains a prime target for future research. We use the Sketch26 program synthesizer. Sketch can solve the following constrained optimization problem, which is equivalent to our goal of maximizing P(X∣T, L)P(T, L∣UG): given observations X and bound on the number of rules K. Sketch offers an exhaustive search strategy, but we use incremental solving in order to scale to large grammars. Mathematically this works as follows: we iteratively construct a sequence of theories T0, T1, … alongside successively larger data sets X0, X1, … converging to the full data set X, such that the tth theory Tt explains data set Xt, and successive theories are close to one another as measured by edit distance: where d( ⋅ , ⋅ ) measures edit distance between theories, Dt+1 is the edit distance between the theory at iteration t + 1 and t, and we use the t = 0 base cases ({{{{{{bf{X}}}}}}}_{0}=varnothing ) and T0 is an empty theory containing no rules. We “minibatch” counterexamples to the current theory (({{{{{bf{X}}}}}}^{prime} ) in Eq. (4)) grouped by lexeme, and ordered by their occurrence in the data (e.g., if the theory fails to explain walk/walks/walked, and this is the next example in the data, then the surface forms of walk/walks/walked will be added to Xt+1). See Supplementary Methods 3.3. We implement all models as Python 2.7 scripts that invoke Sketch 1.7.5, and also use Python 2.7 for all data analysis. Allophony problems comprise the observed form-meaning set X, as well as a substitution, which is a partial function mapping phonemes to phonemes (see Supplementary Methods 3.1). This mapping operates over phonemes called ‘allophones.’ The goal of the model is to recover rule(s) which predicts which element of each allophone pair is an underlying form, and which is merely an allophone. The underlying phonemes are allowed in the lexicon, while the other allophones are not allowed in the lexicon and surface only due phonological rules. For example, an allophony substitution could be (left{b,mapsto, p,d,mapsto, t,g,mapsto, kright}). We extend such substitutions to total functions on phoneme sequences by applying the substitution to phonemes in its domain, and not applying it otherwise. We call this total function s(⋅). For instance, using the previous example substitution, s(abkpg) = apkpk. Solving an allophone problem means finding rules that either map the domain of s(⋅) to its range (T1 below), or vice versa (T2 below): Our few-shot artificial grammar learning simulations require probabilistically scoring held-out unobserved words corresponding to unobserved stems. We now present a refactoring of our Bayesian learning setup that permits these calculations. Given rules T and lexicon L, we define a likelihood PLik over a paradigm matrix X when the data X contain stems disjoint from those in L: where ({{{{{bf{L}}}}}}^{prime} ) ranges over lexica which assign forms to the stems present in X, i.e. ({{{{{bf{L}}}}}}^{prime} ni langle f^{prime},{mathtt{stem}},sigma rangle ) iff X∋ 〈f, [stem: σ; i]〉 for some surface form f and some underlying form (f^{prime} ). The term PLik can be lower bounded by taking the most likely underlying form for each stem: This lower bound will be tightest when each paradigm row admits very few possible stems. Typically only one stem per row is consistent with the rules and affixes, which justifies this bound. The connection between the Bayesian likelihood PLik and the MAP objective (Eq. (1)) can be seen by partitioning the lexicon into affixes (in L) and stems (in ({{{{{bf{L}}}}}}^{prime} )), which also decomposes the objective into a parsimony-favoring prior and a fit-to-data favoring likelihood term: We present our system with a training set Xtrain of words from a target language, such as the ABA language (e.g., /wofewo/, /mikami/, …). We model this training set as a paradigm matrix with a single column (single inflection), with each word corresponding to a different stem (a different row in the matrix). Then we compute the likelihood assigned to a held-out word Xtest either consistent with the target grammar (e.g., following the ABA pattern) or inconsistent with the target grammar (e.g., following the ABB pattern, such as /wofefe/, /mikaka/, …). The probability assigned to a held-out test word, conditioned on the training set, is approximated by marginalizing over the Pareto-optimal grammars for the train set, rather than marginalizing over all possible grammars: which relies on the fact that Sketch has out-of-the-box support for finding Pareto-optimal solutions to multiobjective optimization problems26. We approximate the likelihood PLik(Xtest∣T, L) using the lower bound in Eq. (9), equivalently finding the shortest stem which will generate the test word Xtest, given the affixes in L and the rules in T. At a high level, inference of the cross-language fragment grammar works by maximizing a variational-particle50 lower bound on the joint probability of the metatheory M and the D data sets, ({{{{{{{{bf{X}}}}}}}^{d}}}_{d=1}^{D}): where this bound is written in terms of a set of variational approximate posteriors, ({left{{Q}_{d}right}}_{d=1}^{D}), whose support we constrain to be small, which ensures that the above objective is tractable. We alternate maximization with respect to M (i.e., inferring a fragment grammar from the theories in the supports of ({left{{Q}_{d}right}}_{d=1}^{D})), and maximization with respect to ({left{{Q}_{d}right}}_{d=1}^{D}) (i.e., finding a small set of theories for each data set that are likely under the current M). Our lower bound most increases when the support of each ({left{{Q}_{d}right}}_{d=1}^{D}) coincides with the top-k most likely theories, so at each round of optimization, we ask the program synthesizer to find the top k theories maximizing P(Xd∣Td, Ld)P(Td, Ld∣M). In practice, we find the top k = 100 theories for each data set. We represent M by adapting the Fragment Grammars formalism43. Concretely, M is a probabilistic context free grammar (PCFG) that stochastically generates phonological rules. More precisely, M generates the syntax tree of a program which implements a phonological rule. In the Fragment Grammars formalism, one first defines a base grammar, which is a context-free grammar. Our base grammar is a context-free grammar over SPE rules (Supplementary Fig. 6). Learning the fragment grammar consists of adding new productions to this base grammar (the “fragments”), while also assigning probabilities to each production rule. Formally, each fragment is a subtree of a derivation of a tree generated from a non-terminal symbol in the base grammar; informally, each fragment is a template for a piece of a tree, and thus acts as a schema for a piece of a phonological rule. Learning a fragment grammar never changes the set of trees (i.e., programs and rules) that can be generated from the grammar. Instead, through a combination of estimating probabilities and defining new productions, it adjusts the probability of different trees. See Supplementary Fig. 6, which shows the symbolic structure of the learned fragment grammar. This fragment grammar gives us a learned prior over single phonological rules. We define P(T, L∣M) by assuming that rules are generated independently and that M does not affect the prior probability of L: Our prior over fragment grammars, P(M), works by following the original work in this space43 by assuming that fragments are generated sequentially, with new fragments generated from the current fragment grammar by stochastically sampling them from the current fragment grammar. This encourages shorter fragments, as well as reuse across fragments. We depart from ref. 43 in our inference algorithm: while ref. 43 uses Markov Chain Monte Carlo methods to stochastically sample from the posterior over fragment grammars, we instead perform hillclimbing upon the objective in Eq. (13). Each round of hillclimbing proposes new fragments by antiunifying subtrees of phonological rules in ({left{{{{{{{bf{T}}}}}}}_{d}right}}_{d=1}^{D}), and re-estimates the continuous parameters of the resulting PCFG using the classic Inside–Outside algorithm51. When running Inside-Outside we place a symmetric Dirichlet prior over the continuous parameters of the PCFG with pseudocounts equal to 1. Further information on research design is available in the Nature Research Reporting Summary linked to this article. All data used and generated in this study have been publicly deposited in GitHub52 at https://github.com/ellisk42/bpl_phonology along with the accompanying source code (DOI 10.5281/zenodo.6578329) under the GPLv3 license. Source data are provided with this paper. The code used and developed in this study has been deposited in GitHub52 at https://github.com/ellisk42/bpl_phonology (DOI 10.5281/zenodo.6578329) under the GPLv3 license. Godfrey-Smith, P. Theory and Reality: An Introduction to the Philosophy of Science (University of Chicago Press, 2009). Gerstenberg, T. & Tenenbaum, J. B. In Intuitive Theories. Oxford Handbook of Causal Reasoning 515–548 (Oxford University Press, 2017). Carey, S. Conceptual Change in Childhood (MIT Press, 1985). Schulz, L. The origins of inquiry: inductive inference and exploration in early childhood. Trends Cogn. Sci.16, 382–389 (2012). Article Google Scholar Gopnik, A., Meltzoff, A. N. & Kuhl, P. K. The Scientist in the Crib: Minds, Brains, and How Children Learn (William Morrow & Co, 1999). Langley, P., Simon, H., Bradshaw, G. & Zytkow, J. Scientific Discovery: Computational Explorations of the Creative Processes (MIT Press, 1987). Schmidt, M. & Lipson, H. Distilling free-form natural laws from experimental data. Science324, 81–85 (2009). Kemp, C. & Tenenbaum, J. B. The discovery of structural form. Proc. Natl Acad. Sci. USA105, 10687–10693 (2008). ADSCASArticle Google Scholar Tenenbaum, J. B., Kemp, C., Griffiths, T. L. & Goodman, N. D. How to grow a mind: statistics, structure, and abstraction. Science331, 1279–1285 (2011). ADSMathSciNetCASArticle Google Scholar Bradshaw, G. F., Langley, P. W. & Simon, H. A. Studying scientific discovery by computer simulation. Science222, 971–975 (1983). ADSCASArticle Google Scholar Kemp, C. & Regier, T. Kinship categories across languages reflect general communicative principles. Science336, 1049–1054 (2012). ADSCASArticle Google Scholar Pearl, J. Causality: Models, Reasoning and Inference, Vol. 29 (Springer, 2000). King, R. D. et al. The automation of science. Science324, 85–89 (2009). ADSCASArticle Google Scholar Chomsky, N. Current Issues in Linguistic Theory (De Gruyter, 1964). Chomsky, N. Aspects of the Theory of Syntax, Vol. 11 (MIT Press, 2014). Dupoux, E. Cognitive science in the era of artificial intelligence: a roadmap for reverse-engineering the infant language-learner. Cognition173, 43–59 (2018). Article Google Scholar Frank, M. C., Tenenbaum, J. B. & Gibson, E. Learning and long-term retention of large-scale artificial languages. PLoS ONE8, e52500 (2013). Brown, T. B. et al. Language models are few-shot learners. Adv. Neural Inf. Process. Syst.33, 1877–1901 (2020). Lake, B. M., Salakhutdinov, R. & Tenenbaum, J. B. Human-level concept learning through probabilistic program induction. Science350, 1332–1338 (2015). ADSMathSciNetCASArticle Google Scholar Raimy, E. S. Representing Reduplication. Ph.D. thesis, University of Delaware (1999). Brent, M. R. An efficient, probabilistically sound algorithm for segmentation and word discovery. Mach. Learn.34, 71–105 (1999). Article Google Scholar Chomsky, N. & Halle, M. The Sound Pattern of English. Studies in Language (Harper & Row, 1968). Kaplan, R. M. & Kay, M. Regular models of phonological rule systems. Comput. Linguist.20, 331–378 (1994). Google Scholar Heinz, J. Computational phonology–part i: foundations. Lang. Linguist. Compass5, 140–152 (2011). Article Google Scholar Een, N. Minisat: a sat solver with conflict-clause minimization. Int. Conf. Theory Appl. Satisfiability Testing8, 502–518 (2005). Google Scholar Solar Lezama, A. Program Synthesis By Sketching. Ph.D. thesis, University of California Berkeley (2008). Perelman, D., Gulwani, S., Grossman, D. & Provost, P. Test-driven synthesis. Program. Lang. Des. Implement.49, 408–418 (2014). Google Scholar Odden, D. Introducing Phonology (Cambridge University Press, 2005). Roca, I. & Johnson, W. A Workbook in Phonology (Blackwell, 1991). Halle, M. & Clements, G. N. Problem Book in Phonology: A Workbook for Introductory Courses in Linguistics and in Modern Phonology (MIT Press, 1983). Cotterell, R., Peng, N. & Eisner, J. Modeling word forms using latent underlying morphs and phonology. Trans. Assoc. Comput. Linguist.3, 433–447 (2015). Article Google Scholar Rasin, E., Berger, I., Lan, N. & Katzir, R. Learning phonological optionality and opacity from distributional evidence. North East Linguist. Soc.48, 269–282 (2018). Google Scholar Albright, A. & Hayes, B. Rules vs. analogy in english past tenses: a computational/experimental study. Cognition90, 119–161 (2003). Article Google Scholar Goldwater, S. & Johnson, M. Priors in bayesian learning of phonological rules. Proc. ACL Spl. Interest Group Comput. Phonol.7, 35–42 (2004). Google Scholar Hua, W. & Jardine, A. Learning input strictly local functions from their composition. Int. Conf Grammatical Inference15, 47–65 (2021). MathSciNet Google Scholar Barke, S., Kunkel, R., Polikarpova, N. & Bergen, L. Constraint-based learning of phonological processes. In Empirical Methods in Natural Language Processing 6176–6186 (EMNLP, 2019). Aslin, R. N. & Newport, E. L. Statistical learning from acquiring specific items to forming general rules. Curr. Directions Psychol. Sci.21, 170–176 (2012). Article Google Scholar Gerken, L. Infants use rational decision criteria for choosing among models of their input. Cognition115, 362–366 (2010). Article Google Scholar Marcus, G. F., Vijayan, S., Rao, S. B. & Vishton, P. M. Rule learning by seven-month-old infants. Science283, 77–80 (1999). ADSCASArticle Google Scholar Frank, M. C. & Tenenbaum, J. B. Three ideal observer models for rule learning in simple languages. Cognition120, 360–371 (2011). Article Google Scholar Mattson, C. A. & Messac, A. Pareto frontier based concept selection under uncertainty, with visualization. Optimiz. Eng.6, 85–115 (2005). MathSciNetArticle Google Scholar Yu, A. C. L. Explaining final obstruent voicing in lezgian: phonetics and history. Language80, 73–97 (2004). Article Google Scholar O’Donnell, T. J. Productivity and Reuse in Language: A Theory of Linguistic Computation and Storage (The MIT Press, 2015). Perfors, A., Tenenbaum, J. B. & Regier, T. The learnability of abstract syntactic principles. Cognition118, 306–338 (2011). Article Google Scholar Andreas, J., Rohrbach, M., Darrell, T. & Klein, D. Neural module networks. In Computer Vision and Pattern Recognition 39–48 (IEEE, 2016). Grefenstette, E., Hermann, K. M., Suleyman, M. & Blunsom, P. Learning to transduce with unbounded memory. Adv. Neural Inf. Process. Syst.28, 1828–1836 (2015). Google Scholar Feinman, R. & Lake, B. M. Learning Task-General Representations with Generative Neuro-Symbolic Modeling. In International Conference on Learning Representations (ICLR, 2021). Devlin, J., Chang, M.-W., Lee, K. & Toutanova, K. Bert: pre-training of deep bidirectional transformers for language understanding. Assoc. Comput. Linguistics57, 4171–4186 (2019). Google Scholar Block, N. In The Routledge Encyclopedia of Philosophy. Conceptual Role Semantics 242–256 (Routledge, 1998). Saeedi, A., Kulkarni, T. D., Mansinghka, V. & Gershman, S. Variational particle approximations. J. Mach. Learn. Res.18, 2328–2356 (2014). MathSciNetMATH Google Scholar Lari, K. & Young, S. J. The estimation of stochastic context-free grammars using the inside-outside algorithm. Computer Speech Lang.4, 35–56 (1990). Article Google Scholar Ellis, K. Github, ellisk42/bpl_phonology: v1. https://doi.org/10.5281/zenodo.6578329 (2022). Gerken, L. Decisions decisions: infant language learning when multiple generalizations are possible. Cognition98, B67–B74 (2006). Article Google Scholar Download references We thank Andrea Jang and Francie Freedman for their assistance in transcribing textbook phonology problems. We thank Edward Gibson, Roger Levy, and our anonymous reviewers for comments and suggestions that greatly improved the manuscript. Supported by grants from the Air Force Office of Scientific Research Grant no. FA9550-19-1-0269 (to J.B.T.), the National Science Foundation-funded Center for Brains, Minds, and Machines under Grant No. 1231216 (J.B.T.), the MIT-IBM Watson AI Lab (to J.B.T.), the Natural Sciences and Engineering Research Council of Canada Grant no. RGPIN-2018-06385 (to T.J.O.), the Fonds de Recherche du Québec Société et Culture Grant no. 254753 (to T.J.O.), the Canada CIFAR AI Chairs Program (to T.J.O.), the National Science Foundation under Grant No. 1918839 (to A.S.-L.), and an NSF graduate fellowship (to K.E.). Department of Computer Science, Cornell University, Ithaca, NY, 14850, USA Kevin Ellis Department of Linguistics, Massachusetts Institute of Technology, Cambridge, MA, 02139, USA Adam Albright Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, MA, 02139, USA Armando Solar-Lezama Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology, Cambridge, MA, 02139, USA Joshua B. Tenenbaum Department of Linguistics, McGill University, Montréal, QC, H3A0G4, Canada Timothy J. O’Donnell Canada CIFAR AI Chair, Montréal, QC, H2S3H1, Canada Timothy J. O’Donnell Quebec Artificial Intelligence Institute (Mila), Montréal, QC, H2S3H1, Canada Timothy J. O’Donnell You can also search for this author in PubMedGoogle Scholar You can also search for this author in PubMedGoogle Scholar You can also search for this author in PubMedGoogle Scholar You can also search for this author in PubMedGoogle Scholar You can also search for this author in PubMedGoogle Scholar K.E. and T.J.O. conceived of the project and model. K.E. implemented the model and ran simulations. K.E. and T.J.O. wrote the paper. T.J.O., A.A., J.B.T., and A.S.-L. advised the project. Correspondence to Kevin Ellis. The authors declare no competing interests. Nature Communications thanks Kyle Gorman, Tim Florian Jaeger, and the other anonymous reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available. Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/. Reprints and Permissions Ellis, K., Albright, A., Solar-Lezama, A. et al. Synthesizing theories of human language with Bayesian program induction. Nat Commun13, 5024 (2022). https://doi.org/10.1038/s41467-022-32012-w Download citation Received: Accepted: Published: DOI: https://doi.org/10.1038/s41467-022-32012-w Anyone you share the following link with will be able to read this content: Sorry, a shareable link is not currently available for this article.