<?xml version="1.0" encoding="utf-8"?><feed xmlns="http://www.w3.org/2005/Atom" ><generator uri="https://jekyllrb.com/" version="4.4.1">Jekyll</generator><link href="https://diyclassics.github.io/legacy/feed.xml" rel="self" type="application/atom+xml" /><link href="https://diyclassics.github.io/legacy/" rel="alternate" type="text/html" /><updated>2026-11-20T16:47:31-05:00</updated><id>https://diyclassics.github.io/legacy/feed.xml</id><title type="html">Patrick J. Burns</title><subtitle>Academic portfolio</subtitle><entry><title type="html">How to Fix Your First Billion Latin Typos: AI &amp;amp; Philology at Astronomical Scale</title><link href="https://diyclassics.github.io/legacy/billion-typos/" rel="alternate" type="text/html" title="How to Fix Your First Billion Latin Typos: AI &amp;amp; Philology at Astronomical Scale" /><published>2026-10-06T00:00:00-04:00</published><updated>2026-10-06T00:00:00-04:00</updated><id>https://diyclassics.github.io/legacy/billion-typos</id><content type="html" xml:base="https://diyclassics.github.io/legacy/billion-typos/"><![CDATA[<h1 id="how-to-fix-your-first-billion-latin-typos-ai--philology-at-astronomical-scale">How to Fix Your First Billion Latin Typos: AI &amp; Philology at Astronomical Scale</h1>

<p>Abstract for invited talk at the University of Massachusetts Boston</p>

<p><em>University of Massachusetts Boston, Boston, MA</em></p>

<h2 class="paper-section-heading" id="keywords">Keywords</h2>
<p>Latin, computational philology, textual criticism, large language models, agentic AI</p>

<h2 class="paper-section-heading" id="abstract">Abstract</h2>

<p>Decades of mass digitization projects have given Latinists access to thousands upon thousands of scanned books online (Marcum and Schonfeld 2021). These scanned books also offer computational Latinists billions of words upon which to train state-of-the-art language models for tasks like question answering, text summarization, and machine translation, among others (Bamman and Burns 2020; Riemenschneider and Frank 2023). That said, due to the extremely variable quality of optical character recognition (OCR), we are as likely to find among these volumes quality Latin stretches like eadem dicta esse repertum sit as we are to find catastrophically corrupt stretches like fim ego exfâ‚¬r$mJkm)iMdem (Bamman n.d.). Correcting texts has long been a central concern of philology through the art of textual criticism (West 1973; Tarrant 2016). As James Zetzel (Zetzel 2015: 54) writes, the work of the philologist can be thought of as “reconstructing what [was] written rather than enshrining or embalming the errors transmitted.” Yet when our scope is in the tens of billions of words—and the errors too number in the billions—the work is clearly beyond human scale. In this talk, I discuss my research over the last year in developing computational methods for discovering and resolving error-prone Latin text in collections that can only be described as astronomical (e.g. LLM training data repositories), including the idea of deploying and learning to “trust” agentic AI in the development of philological “self-correcting systems” (Harari 2024). The key questions—when we see a Latin typo online, how is it that we as Latin experts know how to fix the error, how do we program a computer to do the same, and how do we encourage human-computer collaboration in addressing this task? While examples will be drawn from my Latin-focused research agenda, the talk should be of interest to anyone concerned about the role of AI methods in humanities research (Mollick 2024; Beguš 2025).</p>

<h2 class="paper-section-heading" id="bibliography">Bibliography</h2>

<ul>
  <li>Bamman, D. n.d. “11K Latin Texts.” n.d. http://www.cs.cmu.edu/~dbamman/latin.html.</li>
  <li>Bamman, D., and Burns, P.J. 2020. “Latin BERT: A Contextual Language Model for Classical Philology.” September 21. http://arxiv.org/abs/2009.10053.</li>
  <li>Beguš, N. 2025. Artificial Humanities: A Fictional Perspective on Language in AI. Ann Arbor: University of Michigan Press. doi:10.3998/mpub.12778936.</li>
  <li>Harari, Y.N. 2024. Nexus: A Brief History of Information Networks from the Stone Age to AI. New York: Random House.</li>
  <li>Marcum, D., and Schonfeld, R.C. 2021. Along Came Google: A History of Library Digitization. Princeton: Princeton University Press. doi:10.1515/9780691208039.</li>
  <li>Mollick, E. 2024. Co-Intelligence: Living and Working with AI. New York: Portfolio/Penguin.</li>
  <li>Riemenschneider, F., and Frank, A. 2023. “Exploring Large Language Models for Classical Philology.” In Rogers, A., Boyd-Graber, J., and Okazaki, N. eds. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Toronto, Canada: Association for Computational Linguistics. 15181–99. doi:10.18653/v1/2023.acl-long.846. https://aclanthology.org/2023.acl-long.846/.</li>
  <li>Tarrant, R. 2016. Texts, Editors, and Readers: Methods and Problems in Latin Textual Criticism. Cambridge: Cambridge University Press.</li>
  <li>West, M.L. 1973. Textual Criticism and Editorial Technique Applicable to Greek and Latin Texts. Stuttgart: Teubner.</li>
  <li>Zetzel, J.E.G. 2015. “The Bride of Mercury: Confessions of a ’Pataphilologist.” In Pollock, S. ed. World Philology. Cambridge, MA: Harvard University Press. 45–62.</li>
</ul>]]></content><author><name>patrickburns</name></author><category term="paper" /><summary type="html"><![CDATA[Abstract for invited talk at the University of Massachusetts Boston]]></summary></entry><entry><title type="html">The Flipped Lemma: Repurposing a Latin Lexicographical Resource for Disambiguating Word Senses</title><link href="https://diyclassics.github.io/legacy/flipped-lemma/" rel="alternate" type="text/html" title="The Flipped Lemma: Repurposing a Latin Lexicographical Resource for Disambiguating Word Senses" /><published>2026-10-06T00:00:00-04:00</published><updated>2026-10-06T00:00:00-04:00</updated><id>https://diyclassics.github.io/legacy/flipped-lemma</id><content type="html" xml:base="https://diyclassics.github.io/legacy/flipped-lemma/"><![CDATA[<h1 id="the-flipped-lemma-repurposing-a-latin-lexicographical-resource-for-disambiguating-word-senses">The Flipped Lemma: Repurposing a Latin Lexicographical Resource for Disambiguating Word Senses</h1>

<p>Abstract for invited talk at the Thesaurus Linguae Latinae Workshop</p>

<p><em>Thesaurus Linguae Latinae Workshop, Munich, Germany</em></p>

<h2 class="paper-section-heading" id="keywords">Keywords</h2>
<p>Latin lexicography, natural language processing, word sense disambiguation, dataset construction</p>

<h2 class="paper-section-heading" id="abstract">Abstract</h2>

<p>In a follow up to my talk last year at the TLL, I discuss recent efforts to implement a workable Latin word sense disambiguation (WSD) model. In particular, I report on progress with constructing a dataset for training and evaluating such a model. In the first part of the talk, I introduce the LatinCy WSD Dataset, showing that its main contribution is the novelty of its construction: the dataset derives primarily from repurposing an existing Latin lexicographical resource, namely Perseus Project’s digitized version of the Lewis &amp; Short dictionary. WSD datasets require sense-annotated contextual word usage examples. I have been able to bootstrap such a dataset by inverting the logic of L&amp;S: 1. illustration fragments from L&amp;S are expanded to full sentences based on corpus search; and 2. the senses are separated according to the hierarchical structure of the lemmas (i.e. definition subcategories of I, II, A, B, etc.). Using this data repurposing strategy, I have been able to create a dataset with 120K sense-labelled sentences, covering 17K lemmas and 38K distinct senses (~60% of citation-bearing senses from L&amp;S). This data is now being used to train a provisional WSD model for use with the LatinCy NLP pipelines. The second part of the talk suggests two key directions for future work, namely 1. transforming yet again this dataset such that it is compatible with standard NLP instruments for working with word senses (e.g. Senseval and SemEval datasets; cf. also Latin-specific instruments such as Latin WordNet); and 2. pursuing similar sorts of lexicographical inversions from the TLL slip inventory as a potential way to expand and enrich existing resources. The main takeaway of the talk will be an exhortation toward thinking about lexical entries from both top-down (lemma/sense → example) and bottom-up (example → lemma/sense) data perspectives.</p>]]></content><author><name>patrickburns</name></author><category term="paper" /><summary type="html"><![CDATA[Abstract for invited talk at the Thesaurus Linguae Latinae Workshop]]></summary></entry><entry><title type="html">Automating Vocabulary List Production</title><link href="https://diyclassics.github.io/legacy/automating/" rel="alternate" type="text/html" title="Automating Vocabulary List Production" /><published>2026-10-02T10:30:00-04:00</published><updated>2026-10-02T10:30:00-04:00</updated><id>https://diyclassics.github.io/legacy/automating</id><content type="html" xml:base="https://diyclassics.github.io/legacy/automating/"><![CDATA[<h1 id="automating-vocabulary-list-production">Automating Vocabulary List Production</h1>

<p>Abstract for workshop at Classical Association of the Atlantic States Annual Meeting (CAAS2026)</p>

<p><em>Classical Association of the Atlantic States Annual Meeting (CAAS2026), Wilmington, DE</em></p>

<p>Two-hour workshop, 10:30am.</p>

<h2 class="paper-section-heading" id="abstract">Abstract</h2>

<p>The task: given any Latin passage, can we generate accurate and informative vocabulary lists with a minimum of editorial intervention? The subtasks required to achieve this derive from computer-assisted text analysis and specifically the discipline of natural language processing (Jurafsky and Martin 2025). At the very least, automated vocabulary list production requires a pipeline of text activities: segmenting texts into sentences, splitting sentences into words, identifying the lemma and part of speech of these words, identifying realia such as people and places, taking account of collocations and multi-word expressions (i.e. vocabulary items not reducible to single words, such as <em>res publica</em> or <em>quam ob rem</em>), determining which words occur too frequently to be glossed, determining which words are uniquely relevant to the passage, choosing between homonyms (<em>cum</em>/”with” vs. <em>cum</em>/”when”), choosing between word senses (<em>ago</em>/”drive” vs. <em>ago</em>/”do”), formatting entries according to part of speech, and more.</p>

<p>This two-hour workshop demonstrates a Latin NLP pipeline to accomplish each of the tasks above to produce high-quality vocabulary lists. Recent work in Latin NLP through projects like LatinCy (Burns 2023) have made it more straightforward than ever for teachers and students to generate supporting materials for reading Latin, from which vocabulary lists are just one high-value example. The workshop aims to show both 1. how existing tools work on a practical level, but also 2. the underlying logic of why the tools approach the Latin language as they do. In this respect, the workshop aims to offer a computational alternative to the use of large language models for similar tasks: LLMs can produce vocabulary lists for Latin passages, but the way in which the lists are produced is not transparent or reproducible. Through an introduction to NLP basics, Latinists can develop greater confidence in computational language tasks and have greater agency in customizing their outputs.</p>

<p>Using passages from Virgil’s <em>Aeneid</em> and Pliny’s letters, workshop participants will generate vocabulary lists from their own passages of interest. Participants will then discuss as a group where the resulting vocabulary lists succeed and fail, focusing on the remaining deficiencies and what Latinists can do to improve task performance on tasks that computers still struggle with (e.g. word sense disambiguation as with <em>ago</em> above). Throughout the workshop, participants will be invited to connect cutting-edge work in automating Latin vocabulary lists to existing scholarship on vocabulary development (Beyer and Schulz 2020) and related work on Latin lexical frequencies (Francese 2021) and lexical sources (Mulligan 2024).</p>

<p>No prior coding experience is required of participants. That said, participants will be asked to read and refactor existing, working code during the workshop, so an openness to the technical aspects of the process is encouraged. Attendees should bring a laptop to the workshop in order to participate fully. In addition, nothing needs to be installed prior to the workshop; all work will be done in a standard web browser.</p>

<h2 class="paper-section-heading" id="further-reading">Further Reading</h2>

<ul>
  <li>Beyer, Andrea, and Konstantin Schulz. 2020. “Using NLP to Create Corpus-Based Vocabulary Exercises in Latin Classes.” <em>INTED2020 Proceedings</em> (Valencia, Spain), 1750–57. <a href="https://doi.org/10.21125/inted.2020.0562">https://doi.org/10.21125/inted.2020.0562</a>.</li>
  <li>Burns, Patrick J. 2023. “LatinCy: Synthetic Trained Pipelines for Latin NLP.” <a href="https://arxiv.org/abs/2305.04365v1">https://arxiv.org/abs/2305.04365v1</a>.</li>
  <li>Francese, Christopher. 2021. “Latin Core Vocabulary.” Dickinson College Commentaries. <a href="https://dcc.dickinson.edu/vocab/core-vocabulary">https://dcc.dickinson.edu/vocab/core-vocabulary</a>.</li>
  <li>Jurafsky, Daniel, and James H. Martin. 2025. <em>Speech and Language Processing</em>. 3rd edition, draft. <a href="https://web.stanford.edu/~jurafsky/slp3/">https://web.stanford.edu/~jurafsky/slp3/</a>.</li>
  <li>Mulligan, Bret. 2024. “Bridge: Customizable Vocabulary Lists.” <a href="https://bridge.haverford.edu/">https://bridge.haverford.edu/</a>.</li>
</ul>]]></content><author><name>patrickburns</name></author><category term="paper" /><summary type="html"><![CDATA[Abstract for workshop at Classical Association of the Atlantic States Annual Meeting (CAAS2026)]]></summary></entry><entry><title type="html">Discovering Novel Word Order Patterns in Latin Epic in 21 Centuries of Latin Epic</title><link href="https://diyclassics.github.io/legacy/word-order/" rel="alternate" type="text/html" title="Discovering Novel Word Order Patterns in Latin Epic in 21 Centuries of Latin Epic" /><published>2026-07-14T00:00:00-04:00</published><updated>2026-07-14T00:00:00-04:00</updated><id>https://diyclassics.github.io/legacy/word-order</id><content type="html" xml:base="https://diyclassics.github.io/legacy/word-order/"><![CDATA[<h1 id="discovering-novel-word-order-patterns-in-latin-epic-in-21-centuries-of-latin-epic">Discovering Novel Word Order Patterns in Latin Epic in 21 Centuries of Latin Epic</h1>
<p>Abstract for talk at Epic Through Numbers: Digital and Quantitative Approaches to Greek and Latin Epic, Celtic Conference in Classics. Maynooth University Department of Classics. Maynooth, Ireland.</p>

<h2 class="paper-section-heading" id="abstract">Abstract</h2>

<p>Golden lines—whether a literary critical invention of 17th-century English writers like Burles or Dryden or a genuine feature of Latin versification (Mayer 2020)—have attracted outsized attention due to their classical yet formalist poetics (Wilkinson 1963; Heikkinen 2015). A “golden” arrangement of a two adjectives, a central verb, and two nouns in (usually parallel but also chiastic) agreement with the preceding modifiers within a five-word hexameter can be found in varying proportions in all of the Latin epic poets following Catullus 64 (Conrad 1965): e.g. Virgil, <em>Aeneid</em> 4.139: <em>Aurea</em>-ADJECTIVE/1 <em>purpuream</em>-ADJECTIVE/2 <em>subnectit</em>-VERB <em>fibula</em>-NOUN/1 <em>vestem</em>-NOUN/2.</p>

<p>The novel poetic form makes an attractive test case for Latin-language computational literary criticism since its identification draws not only on textual formalisms like word position and line arrangement but also requires natural language processing (NLP) annotations like part-of-speech tags to identify the nouns, adjectives, and verbs as well as morphological tags to establish agreement between those nouns and adjectives. The landscape for Latin NLP is as robust as ever, and with the ready availability of digitized Latin text we are in an excellent position to identify poetic-grammatical features like golden lines at scale from two millennia of epic verse.</p>

<p>In this paper-as-workshop, I will demonstrate recent NLP models I have developed for Latin, including the LatinCy spaCy pipelines as well as those for Stanza, Flair, and UDPipe (Burns 2023). I will take participants through the whole experimental research design cycle: formulating the golden line-based research goal, building the epic text collection, processing these texts with various NLP workflows with attention to POS and morphological tagging, and most importantly, a reporting of results—especially with respect to later renaissance Latin and neo-Latin epic—on golden line trends throughout the language’s literary history. One novel result in particular will combine the “golden” NLP annotations noted above with named entity recognition to provide us (again, at scale) with a list of hexameter lines displaying poetically marked versification that also calls attention to the names of important people and locations.</p>

<h2 class="paper-section-heading" id="works-cited">Works Cited</h2>

<ul>
  <li>Burns, P.J. 2023. “LatinCy: Synthetic Trained Pipelines for Latin NLP.” <a href="https://arxiv.org/abs/2305.04365v1">https://arxiv.org/abs/2305.04365v1</a>.</li>
  <li>Conrad, C. 1965. “Traditional Patterns of Word-Order in Latin Epic from Ennius to Vergil.” <em>Harvard Studies in Classical Philology</em> 69: 195-258.</li>
  <li>Heikkinen, S. 2015. “From Persius to Wilkinson: The Golden Line Revisited.” <em>Arctos–Acta Philologica Fennica</em> 49: 57–77.</li>
  <li>Mayer, K. 2020. “The Schoolboys’ Revenge: How the Golden Line Entered Classical Scholarship.” <em>Classical Receptions Journal</em> 12(2): 248–78.</li>
  <li>Wilkinson, L.P. 1963. <em>Golden Latin Artistry</em>. Cambridge: Cambridge University Press.</li>
</ul>]]></content><author><name>patrickburns</name></author><category term="paper" /><summary type="html"><![CDATA[Abstract for talk at Epic Through Numbers: Digital and Quantitative Approaches to Greek and Latin Epic, Celtic Conference in Classics, Maynooth University]]></summary></entry><entry><title type="html">Automatic Vocabulary Lists: NLP Experiments toward a Pedagogical Goal</title><link href="https://diyclassics.github.io/legacy/vocab-lists-hunter/" rel="alternate" type="text/html" title="Automatic Vocabulary Lists: NLP Experiments toward a Pedagogical Goal" /><published>2026-06-10T00:00:00-04:00</published><updated>2026-06-10T00:00:00-04:00</updated><id>https://diyclassics.github.io/legacy/vocab-lists-hunter</id><content type="html" xml:base="https://diyclassics.github.io/legacy/vocab-lists-hunter/"><![CDATA[<h1 id="automatic-vocabulary-lists-nlp-experiments-toward-a-pedagogical-goal">Automatic Vocabulary Lists: NLP Experiments toward a Pedagogical Goal</h1>
<p>Abstract for talk at Hunter College High School Classics Club</p>

<p><em>Hunter College High School Classics Club, New York, NY</em></p>

<h2 class="paper-section-heading" id="abstract">Abstract</h2>

<h2 id="abstract-1">Abstract</h2>

<p>Using Pliny the Younger’s letter on Vesuvius as a case study, I discuss take up the following task: given a plaintext passage of Latin, how can leverage natural language processing methods, tools, and dataset to produce publication-ready vocabulary lists, i.e. the kinds of well-formatted word lists that a printed in textbooks to support student learning. With looks at tokenization, lemmatization, part-of-speech tagging, lexical expansion, word frequency analysis, word sense disambiguation, and other NLP techniques, I introduce core NLP methods and frame the vocabulary list as an example of what Jeanette Wing has called “computational thinking,” that is a way of formulating problems as series of smaller formal tasks. I conclude with an honest assessment of the current state of Latin NLP and where the greatest opportunities can be found for computationally curious Latin students.</p>]]></content><author><name>patrickburns</name></author><category term="paper" /><summary type="html"><![CDATA[Abstract for talk at Hunter College High School Classics Club]]></summary></entry><entry><title type="html">Text Analysis for Historical Language Research</title><link href="https://diyclassics.github.io/legacy/isaw-f2026-ga3023/" rel="alternate" type="text/html" title="Text Analysis for Historical Language Research" /><published>2026-05-19T00:00:00-04:00</published><updated>2026-05-19T00:00:00-04:00</updated><id>https://diyclassics.github.io/legacy/isaw-f2026-ga3023</id><content type="html" xml:base="https://diyclassics.github.io/legacy/isaw-f2026-ga3023/"><![CDATA[<h1 id="text-analysis-for-historical-language-research">Text Analysis for Historical Language Research</h1>

<p><em>Taught at the Institute for the Study of the Ancient World (ISAW-GA 3023), Fall 2026. This course meets 1 time a week in a 3-hour block.</em></p>

<h2 id="course-description">Course Description</h2>

<p>This course introduces students to computational research methods helpful for producing data-driven scholarship involving large collections of historical-language text. Drawing on relevant topics in exploratory data science, corpus linguistics, and natural language processing, the course provides a forum for students to develop hands-on skills in computer programming (using Python), focused primarily on managing textual data, string manipulation, text mining and analysis, language modeling, and data visualization. Special attention will be given to the use of word embeddings, transformer models, and large language models and their applicability to historical-language text collections. Demonstrations throughout the course will draw primarily on English-language examples, but because of the philological range and diversity at ISAW, students are encouraged to work with digitized text collections in the languages most relevant to their research. There are no prerequisites, though students are expected to be open to reading, writing, and editing computer programs; students are required to bring laptops to class. Note that historical-language text for the purpose of this course covers texts or collections of texts written before the Early Modern period.</p>]]></content><author><name>patrickburns</name></author><category term="syllabus" /><summary type="html"><![CDATA[Graduate course at ISAW (Fall 2026) introducing computational methods for historical-language research: Python-based text analysis & NLP via word embeddings, transformer models, and large language models.]]></summary></entry><entry><title type="html">RespondeoQA: a Benchmark for Bilingual Latin-English Question Answering</title><link href="https://diyclassics.github.io/legacy/respondeoqa/" rel="alternate" type="text/html" title="RespondeoQA: a Benchmark for Bilingual Latin-English Question Answering" /><published>2026-04-23T00:00:00-04:00</published><updated>2026-04-23T00:00:00-04:00</updated><id>https://diyclassics.github.io/legacy/respondeoqa</id><content type="html" xml:base="https://diyclassics.github.io/legacy/respondeoqa/"><![CDATA[<h1 id="respondeoqa-a-benchmark-for-bilingual-latin-english-question-answering">RespondeoQA: a Benchmark for Bilingual Latin-English Question Answering</h1>
<p>Abstract for RespondeoQA paper on arXiv.<br />
Written with M. Hudspeth and B. O’Connor.</p>

<h2 class="paper-section-heading" id="abstract">Abstract</h2>
<p>We introduce a benchmark dataset for question answering and translation in bilingual Latin and English settings, containing about 7,800 question-answer pairs. The questions are drawn from Latin pedagogical sources, including exams, quizbowl-style trivia, and textbooks ranging from the 1800s to the present. After automated extraction, cleaning, and manual review, the dataset covers a diverse range of question types: knowledge- and skill-based, multihop reasoning, constrained translation, and mixed language pairs. To our knowledge, this is the first QA benchmark centered on Latin. As a case study, we evaluate three large language models – LLaMa 3, Qwen QwQ, and OpenAI’s o3-mini – finding that all perform worse on skill-oriented questions. Although the reasoning models perform better on scansion and literary-device tasks, they offer limited improvement overall. QwQ performs slightly better on questions asked in Latin, but LLaMa3 and o3-mini are more task dependent. This dataset provides a new resource for assessing model capabilities in a specialized linguistic and cultural domain, and the creation process can be easily adapted for other languages. The dataset is available at: https://github.com/slanglab/RespondeoQA</p>]]></content><author><name>patrickburns</name></author><category term="paper" /><summary type="html"><![CDATA[Abstract for RespondeoQA paper on arXiv.]]></summary></entry><entry><title type="html">Automatic Vocabulary Lists: NLP Experiments toward a Pedagogical Goal</title><link href="https://diyclassics.github.io/legacy/vocab-lists-fordham/" rel="alternate" type="text/html" title="Automatic Vocabulary Lists: NLP Experiments toward a Pedagogical Goal" /><published>2026-04-01T00:00:00-04:00</published><updated>2026-04-01T00:00:00-04:00</updated><id>https://diyclassics.github.io/legacy/vocab-lists-fordham</id><content type="html" xml:base="https://diyclassics.github.io/legacy/vocab-lists-fordham/"><![CDATA[<h1 id="automatic-vocabulary-lists-nlp-experiments-toward-a-pedagogical-goal">Automatic Vocabulary Lists: NLP Experiments toward a Pedagogical Goal</h1>
<p>Abstract for talk at Fordham Preparatory School</p>

<p><em>Fordham Preparatory School, Bronx, NY</em></p>

<h2 class="paper-section-heading" id="abstract">Abstract</h2>

<p>Using Pliny the Younger’s letter on Vesuvius as a case study, I discuss take up the following task: given a plaintext passage of Latin, how can leverage natural language processing methods, tools, and dataset to produce publication-ready vocabulary lists, i.e. the kinds of well-formatted word lists that a printed in textbooks to support student learning. With looks at tokenization, lemmatization, part-of-speech tagging, lexical expansion, word frequency analysis, word sense disambiguation, and other NLP techniques, I introduce core NLP methods and frame the vocabulary list as an example of what Jeanette Wing has called “computational thinking,” that is a way of formulating problems as series of smaller formal tasks. I conclude with an honest assessment of the current state of Latin NLP and where the greatest opportunities can be found for computationally curious Latin students.</p>]]></content><author><name>patrickburns</name></author><category term="paper" /><summary type="html"><![CDATA[Abstract for talk at Fordham Preparatory School]]></summary></entry><entry><title type="html">Is Agentic Philology an Oxymoron? Some Thoughts on Error, Control, and Disciplinary Definition</title><link href="https://diyclassics.github.io/legacy/agentic-philology/" rel="alternate" type="text/html" title="Is Agentic Philology an Oxymoron? Some Thoughts on Error, Control, and Disciplinary Definition" /><published>2026-03-13T00:00:00-04:00</published><updated>2026-03-13T00:00:00-04:00</updated><id>https://diyclassics.github.io/legacy/agentic-philology</id><content type="html" xml:base="https://diyclassics.github.io/legacy/agentic-philology/"><![CDATA[<h1 id="is-agentic-philology-an-oxymoron-some-thoughts-on-error-control-and-disciplinary-definition">Is Agentic Philology an Oxymoron? Some Thoughts on Error, Control, and Disciplinary Definition</h1>
<p>Abstract for talk at <em>AI &amp; the Study of Antiquity</em>, Rutgers University Classics Department. New Brunswick, NJ.</p>

<h2 class="paper-section-heading" id="abstract">Abstract</h2>

<p>In this talk, I look at the Latin text content found in large language model (LLM) training data repositories and specifically the high rates of corrupted text resulting from decades of OCR errors and other scanning artifacts. I argue here that the extent of the corruption—measured in the <em>billions</em> of errors—is beyond human scale, perhaps even testing the limits of smaller model-based correction approaches, and that we should therefore consider the possibility of an “agentic philology.” That is, we should look at the possibility of using AI agents to perform computational-scale text critical work on these collections in such a way that we take advantage of what Stuart Russell and Peter Norvig (2021, p. 3) have defined as the five defining characteristics of agents, namely the ability to “operate autonomously, perceive their environment, persist over a prolonged period of time, adapt to change, and create and pursue goals.” I discuss the ways in which such human-out-of-the-loop approaches do not easily align with existing philological expectations of authority and control in the face of error and call for more critical discussion of novel agentic methods in the face of computational-scale error correction challenges.</p>

<h2 class="paper-section-heading" id="references">References</h2>

<ul>
  <li>Russell, S. and Norvig, P. (2021). <em>Artificial Intelligence: A Modern Approach</em> (4th ed.). Pearson.</li>
</ul>]]></content><author><name>patrickburns</name></author><category term="paper" /><summary type="html"><![CDATA[Abstract for talk at AI & the Study of Antiquity, Rutgers University Classics Department]]></summary></entry><entry><title type="html">Knowledge and Know-How in the Humanities (or Turning an AI Challenge into a Higher Ed Advantage)</title><link href="https://diyclassics.github.io/legacy/know-how/" rel="alternate" type="text/html" title="Knowledge and Know-How in the Humanities (or Turning an AI Challenge into a Higher Ed Advantage)" /><published>2026-03-03T00:00:00-05:00</published><updated>2026-03-03T00:00:00-05:00</updated><id>https://diyclassics.github.io/legacy/know-how</id><content type="html" xml:base="https://diyclassics.github.io/legacy/know-how/"><![CDATA[<h1 id="knowledge-and-know-how-in-the-humanities-or-turning-an-ai-challenge-into-a-higher-ed-advantage">Knowledge and Know-How in the Humanities (or Turning an AI Challenge into a Higher Ed Advantage)</h1>

<p>Abstract for invited talk at the Brooklyn College Department of Classics &amp; Teaching and Learning Center</p>

<p><em>Brooklyn College Department of Classics &amp; Teaching and Learning Center, Brooklyn, NY</em></p>

<h2 class="paper-section-heading" id="keywords">Keywords</h2>
<p>generative AI, humanities, pedagogy, large language models</p>

<h2 class="paper-section-heading" id="abstract">Abstract</h2>

<p>A longstanding objection to the possibility of artificial general intelligence is that “our background knowledge consists largely of skills for dealing with things and people”—that is, common sense—“rather than facts about them,” (Dreyfus 1992: xii; cf also Levesque 2017). Under this “commonsense-knowledge problem,” as philosopher Hubert Dreyfus defines it in the introduction to <em>What Computers Still Can’t Do</em>, it can be argued that no amount of training data can replace the embodied, world-lived experience that is at the heart of genuine human intelligence. AI can (perhaps) represent knowledge, i.e. knowing-that; humans recognize in each other both knowledge and knowing-how, that is the kind of metaknowledge that (56) “guides our changing sense of what counts as the relevant facts.” With that in mind, it seems to me that, when an English professor (Rotella 2025) describes components of an “AI-resistant” course in the ChatGPT era such as teaching process over product and nurturing a classroom community dynamic, the suggestions are—though without explicit reference to the AI critical tradition—an appeal to the ability of such courses to cultivate a humanities-informed know-how.</p>

<p>In this talk, I discuss two contemporary aspects of humanities method that speak to the cultivation of know-how in the university classroom: 1. drawing on the work of digital humanist Jentery Sayers, I discuss the idea of humanist-as-tinkerer (Sayers 2011); and 2. drawing on my own experience as a Latinist who develops computational language models, I discuss the idea of humanist-as-craftsman (cf. Mills 1980). Both tinkering and craftsmanship, I argue, demonstrate an honest curiosity about how knowledge structures and technologies work, reinforcing the know-how that characterizes humanities traditions worth preserving (cf. LaPensee and Kaya 2025) like critical thinking and interpretation as well as close reading and deep engagement.</p>

<h2 class="paper-section-heading" id="works-cited">Works Cited</h2>

<ul>
  <li>Dreyfus, H.L. 1992. <em>What Computers Still Can’t Do: A Critique of Artificial Reason</em>. MIT Press.</li>
  <li>LaPensee, B., and Kaya, D. 2025. “The Humanities in the Age of AI: Notes on a Participatory Exploration at MLA 2025.” <em>JSTOR Blog</em>. https://about.jstor.org/blog/mla_2025_highlights/.</li>
  <li>Levesque, H.J. 2017. <em>Common Sense, the Turing Test, and the Quest for Real AI</em>. MIT Press.</li>
  <li>Mills, C.W. 1980. “On Intellectual Craftsmanship (1952).” <em>Society</em> 17(2): 63–70.</li>
  <li>Rotella, C. 2025. “I’m a Professor. A.I. Has Changed My Classroom, but Not for the Worse.” <em>New York Times Magazine</em>, November 25.</li>
  <li>Sayers, J. 2011. “Tinker-Centric Pedagogy in Literature and Language Classrooms.” In McGrath, L. ed. <em>Collaborative Approaches to the Digital in English Studies</em>. Logan, UT: Utah State University Press. 279–300.</li>
</ul>]]></content><author><name>patrickburns</name></author><category term="paper" /><summary type="html"><![CDATA[Abstract for invited talk at the Brooklyn College Department of Classics & Teaching and Learning Center]]></summary></entry></feed>