[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-156753-105":59,"doc-detail-156753-en":130},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":123,"head_meta":125,"extra_data":127,"updated_unix":129},105,"en","learning-interpretable-style-embeddings-via-prompting-llms-a-synthetic-stylometry-dataset-and-interpretable-model","Learning Interpretable Style Embeddings via Prompting LLMs - A Synthetic Stylometry Dataset and Interpretable Model","","Style representation learning focuses on authorial cues that remain consistent across different content, enabling tasks such as stylometry for authorship attribution. Prior work either relies on expensive human stylometric annotation or uses unsupervised neural disentanglement that yields latent, uninterpretable style vectors. This work uses prompting with a large language model to synthesize a stylometry dataset, STYLEGENOME, then trains linguistically interpretable LISA embeddings with one dimension per style attribute for explainability and auditing.",{"@graph":69,"@context":122},[70,84,105],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":40,"@type":76,"position":81},"https://docshare.wps.com/document/research-report/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/learning-interpretable-style-embeddings-via-prompting-llms-a-synthetic-stylometry-dataset-and-interpretable-model/156753/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/learning-interpretable-style-embeddings-via-prompting-llms-a-synthetic-stylometry-dataset-and-interpretable-model/156753.png","ImageObject",300,407,{"name":92,"@type":93},"River Wang","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-11-07","2026-08-29",true,{"@type":102,"interactionType":103,"userInteractionCount":52},"InteractionCounter",{"@type":104},"ViewAction",{"@type":106,"mainEntity":107},"FAQPage",[108,114,118],{"name":109,"@type":110,"acceptedAnswer":111},"Why are existing neural style representations difficult to use in authorship attribution?","Question",{"text":112,"@type":113},"They often produce uninterpretable latent vectors, making auditing and explainability hard. This limits reliability in real-world settings where interpretability is critical.","Answer",{"name":115,"@type":110,"acceptedAnswer":116},"How does the work create the synthetic stylometry dataset?",{"text":117,"@type":113},"It uses a large language model with zero-shot prompts to generate stylometric annotations for many texts, forming STYLEGENOME.",{"name":119,"@type":110,"acceptedAnswer":120},"What are LISA embeddings and how are they made interpretable?",{"text":121,"@type":113},"LISA maps a text to a D-dimensional vector where each dimension corresponds to a specific linguistic style attribute and is constrained to represent attribute presence probability.","https://schema.org",{"og:url":83,"og:type":124,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":126,"canonical":83},"index,follow",{"doc_id":128,"site_id":62},156753,1787965716,{"code":4,"msg":5,"data":131},{"doc_id":128,"user_id":132,"nickname":92,"user_avatar":133,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":52,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":139,"language":140,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":141,"faqs":142,"seo_title":143,"seo_description":67,"update_tm":129,"read_time":144,"preview_page_count":4},1099514067438,"https://ap-avatar.wpscdn.com/avatar/100002539ee87300030?x-image-process=image/resize,m_fixed,w_180,h_180&k=1780474512215547542","Learning Interpretable Style Embeddings via Prompting LLMs  \nAjay Patel  \nUniversity of Pennsylvania [ajayp@upenn.edu](ajayp@upenn.edu)  \nDelip Rao  \nUniversity of Pennsylvania [deliprao@gmail.com](deliprao@gmail.com)  \nAnsh Kothary  \nColumbia University [ank2145@columbia.edu](ank2145@columbia.edu)  \nKathleen McKeown  \nColumbia University [kathy@cs.columbia.edu](kathy@cs.columbia.edu)  \nChris Callison-Burch  \nUniversity of Pennsylvania [ccb@upenn.edu](ccb@upenn.edu)  \nAbstract  \nStyle representation learning builds contentindependent representations of author style in text. To date, no large dataset of texts with stylometric annotations on a wide range of style dimensions has been compiled, perhaps because the linguistic expertise to perform such annotation would be prohibitively expensive. Therefore, current style representation approaches make use of unsupervised neural methods to disentangle style from content to create style vectors. These approaches, however, result in uninterpretable representations, complicating their usage in downstream applications like authorship attribution where auditing and explainability is critical. In this work, we use prompting to perform stylometry on a large number of texts to generate a synthetic stylometry dataset. We use this synthetic data to then train humaninterpretable style representations we call LISA embeddings. We release our synthetic dataset (STYLEGENOME) and our interpretable style embedding model (LISA) as resources.  \n1 Introduction  \nStyle representation learning aims to represent the stylistic attributes of an authored text. Prior work has treated the style of a text as separable from the content. Stylistic attributes have included, but are not limited to, linguistic choices in syntax, grammar, spelling, vocabulary, and punctuation (Jafaritazehjani et al., 2020) . Style representations should represent two texts with similar stylistic attributes more closely than texts with different attributes independent of what content is present in the texts.  \nStylometry, the analysis of style, applies forensic linguistics to tasks like authorship attribution. Stylometry often relies on semi-manual analysis by forensic linguistic experts (Mosteller and Wallace, 1963 ; Holmes, 1994 ; Rosso et al., 2016) . Computational stylometry often uses rule-based approaches utilizing count-based features like the frequencies of function words (Stamatatos, 2009 ; Koppel et al.,  \nFigure 1: An example of a 768-dimensional interpretable style vector produced by LISA, trained using a GPT-3 annotated synthetic stylometery dataset.  \n2009 ; Tausczik and Pennebaker, 2010) . More modern, neural approaches attempt to learn style representations in an unsupervised fashion through a proxy task like style transfer (Shen et al., 2017 ; Fu et al., 2018 ; John et al., 2019 ; Dai et al., 2019 ; Liet al., 2019 ; Yi et al., 2021 ; Zhu et al., 2022) or authorship verification (Boenninghoff et al., 2019 ; Hay et al., 2020 ; Zhu and Jurgens, 2021 ; Wegmannet al., 2022) . These stronger neural approaches, unlike simpler frequency-based techniques, are uninterpretable. This makes it difficult to effectively analyze their representations and their failure modes, and precludes their usage in real-world authorship attribution scenarios because interpretability and verification is critical for legal admissibility (Tiersma and Solan, 2002) .  \nWith this motivation, we propose a humaninterpretable style representation model M which, for a given text t, produces a D-dimensional vector M (t) ∈ [0 .0 , 1.0]D . Each dimension corresponds to one of D style attributes {a0 , a 1 ,..., aD } . Each element at dimension d of this vector is constrained in the range [0 .0 , 1.0] to represent the probability of the corresponding style attribute ad being present  \n15270  \nFindings of the Association for Computational Linguistics: EMNLP 2023 , pages 15270–15290 December 6-10, 2023 ©2023 Association for Computational Linguistics  \nin the text","cbCaitkMdM2bFiUD","https://ap.wps.com/l/cbCaitkMdM2bFiUD","pdf",628092,21,"English","# Abstract\n# Introduction\n# Generating STYLEGENOME\n## Training data and synthetic stylometric annotations\n# LISA embeddings and evaluation\n## Interpretability and transparency","[{\"question\":\"Why are existing neural style representations difficult to use in authorship attribution?\",\"answer\":\"They often produce uninterpretable latent vectors, making auditing and explainability hard. This limits reliability in real-world settings where interpretability is critical.\"},{\"question\":\"How does the work create the synthetic stylometry dataset?\",\"answer\":\"It uses a large language model with zero-shot prompts to generate stylometric annotations for many texts, forming STYLEGENOME.\"},{\"question\":\"What are LISA embeddings and how are they made interpretable?\",\"answer\":\"LISA maps a text to a D-dimensional vector where each dimension corresponds to a specific linguistic style attribute and is constrained to represent attribute presence probability.\"}]","Learning Interpretable Style Embeddings via Prompting LLMs - A Synthetic Stylometry Dataset and Interpretable Model | PDF",53]