Ted Underwood

Cody Kommers, Ruth Ahnert, Maria Antoniak et al.

Generative AI systems are increasingly recognized as cultural technologies, yet current evaluation frameworks often treat culture as a variable to be measured rather than fundamental to the system's operation. Drawing on hermeneutic theory from the humanities, we argue that GenAI systems function as "context machines" that must inherently address three interpretive challenges: situatedness (meaning only emerges in context), plurality (multiple valid interpretations coexist), and ambiguity (interpretations naturally conflict). We present computational hermeneutics as an emerging framework offering an interpretive account of what GenAI systems do, and how they might do it better. We offer three principles for hermeneutic evaluation -- that benchmarks should be iterative, not one-off; include people, not just machines; and measure cultural context, not just model output. This perspective offers a nascent paradigm for designing and evaluating contemporary AI systems: shifting from standardized questions about accuracy to contextual ones about meaning.

Sarah Griebel, Ted Underwood

Literary translation requires balancing target-language fluency with faithfulness to the source. Recent large language models (LLMs) often produce fluent translations, but it remains unclear whether fluency corresponds to semantic preservation in literary text. We examine this relationship using 130,486 translated paragraphs from 106 novels in 16 source languages, including human, Google Translate, and TranslateGemma translations. Fluency is measured as original-likeness with a translationese classifier trained on paragraph part-of-speech n-grams, and faithfulness with the automatic translation evaluation metric COMET-KIWI. We control for paragraph length and find a consistent negative correlation between fluency and faithfulness. The pattern appears for both human and Google Translate, but is weaker and often non-significant for TranslateGemma. These results show that segment length matters for automatic evaluation and suggest a tradeoff between fluency and faithfulness in literary translation.

Sarah Griebel, Becca Cohen, Lucian Li et al.

Measures of textual similarity and divergence are increasingly used to study cultural change. But which measures align, in practice, with social evidence about change? We apply three different representations of text (topic models, document embeddings, and word-level perplexity) to three different corpora (literary studies, economics, and fiction). In every case, works by highly-cited authors and younger authors are textually ahead of the curve. We don't find clear evidence that one representation of text is to be preferred over the others. But alignment with social evidence is strongest when texts are represented through the top quartile of passages, suggesting that a text's impact may depend more on its most forward-looking moments than on sustaining a high level of innovation throughout.

Ted Underwood, Laura K. Nelson, Matthew Wilkens

Before researchers can use language models to simulate the past, they need to understand the risk of anachronism. We find that prompting a contemporary model with examples of period prose does not produce output consistent with period style. Fine-tuning produces results that are stylistically convincing enough to fool an automated judge, but human evaluators can still distinguish fine-tuned model outputs from authentic historical text. We tentatively conclude that pretraining on period prose may be required in order to reliably simulate historical perspectives for social research.

Ted Underwood

Measuring similarity is a basic task in information retrieval, and now often a building-block for more complex arguments about cultural change. But do measures of textual similarity and distance really correspond to evidence about cultural proximity and differentiation? To explore that question empirically, this paper compares textual and social measures of the similarities between genres of English-language fiction. Existing measures of textual similarity (cosine similarity on tf-idf vectors or topic vectors) are also compared to new strategies that use supervised learning to anchor textual measurement in a social context.

Ted Underwood, Michael L. Black, Loretta Auvil et al.

To mine large digital libraries in humanistically meaningful ways, scholars need to divide them by genre. This is a task that classification algorithms are well suited to assist, but they need adjustment to address the specific challenges of this domain. Digital libraries pose two problems of scale not usually found in the article datasets used to test these algorithms. 1) Because libraries span several centuries, the genres being identified may change gradually across the time axis. 2) Because volumes are much longer than articles, they tend to be internally heterogeneous, and the classification task needs to begin with segmentation. We describe a multi-layered solution that trains hidden Markov models to segment volumes, and uses ensembles of overlapping classifiers to address historical change. We test this approach on a collection of 469,200 volumes drawn from HathiTrust Digital Library. To demonstrate the humanistic value of these methods, we extract 32,209 volumes of fiction from the digital library, and trace the changing proportions of first- and third-person narration in the corpus. We note that narrative points of view seem to have strong associations with particular themes and genres.