Marisa Hudspeth

Marisa Hudspeth, Brendan O'Connor, Laure Thompson · princeton

Existing Latin treebanks draw from Latin's long written tradition, spanning 17 centuries and a variety of cultures. Recent efforts have begun to harmonize these treebanks' annotations to better train and evaluate morphological taggers. However, the heterogeneity of these treebanks must be carefully considered to build effective and reliable data. In this work, we review existing Latin treebanks to identify the texts they draw from, identify their overlap, and document their coverage across time and genre. We additionally design automated conversions of their morphological feature annotations into the conventions of standard Latin grammar. From this, we build new time-period data splits that draw from the existing treebanks which we use to perform a broad cross-time analysis for POS and morphological feature tagging. We find that BERT-based taggers outperform existing taggers while also being more robust to cross-domain shifts.

Marisa Hudspeth, Patrick J. Burns, Brendan O'Connor

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

Marisa Hudspeth, Patrick J. Burns, Brendan O'Connor

Tokenization is a critical component of language model pretraining, yet standard tokenization methods often prioritize information-theoretical goals like high compression and low fertility rather than linguistic goals like morphological alignment. In fact, they have been shown to be suboptimal for morphologically rich languages, where tokenization quality directly impacts downstream performance. In this work, we investigate morphologically-aware tokenization for Latin, a morphologically rich language that is medium-resource in terms of pretraining data, but high-resource in terms of curated lexical resources -- a distinction that is often overlooked but critical in discussions of low-resource language modeling. We find that morphologically-guided tokenization improves overall performance on four downstream tasks. Performance gains are most pronounced for out of domain texts, highlighting our models' improved generalization ability. Our findings demonstrate the utility of linguistic resources to improve language modeling for morphologically complex languages. For low-resource languages that lack large-scale pretraining data, the development and incorporation of linguistic resources can serve as a feasible alternative to improve LM performance.

Catherine Arnett, Marisa Hudspeth, Brendan O'Connor

While tokenization is a key step in language modeling, with effects on model training and performance, it remains unclear how to effectively evaluate tokenizer quality. One proposed dimension of tokenizer quality is the extent to which tokenizers preserve linguistically meaningful subwords, aligning token boundaries with morphological boundaries within a word. We expand MorphScore (Arnett & Bergen, 2025), which previously covered 22 languages, to support a total of 70 languages. The updated MorphScore offers more flexibility in evaluation and addresses some of the limitations of the original version. We then correlate our alignment scores with downstream task performance for five pre-trained languages models on seven tasks, with at least one task in each of the languages in our sample. We find that morphological alignment does not explain very much variance in model performance, suggesting that morphological alignment alone does not measure dimensions of tokenization quality relevant to model performance.

Prashant Jayannavar, Liliang Ren, Marisa Hudspeth et al.

Developing interactive agents that can understand language, perceive their surroundings, and act within the physical world is a long-standing goal of AI research. The Minecraft Collaborative Building Task (MCBT) (Narayan-Chen, Jayannavar, and Hockenmaier 2019), a two-player game in which an Architect (A) instructs a Builder (B) to construct a target structure in a simulated 3D Blocks World environment, offers a rich platform to work towards this goal. In this work, we focus on the Builder Action Prediction (BAP) subtask: predicting B's actions in a multimodal game context (Jayannavar, Narayan-Chen, and Hockenmaier 2020) - a challenging testbed for grounded instruction following, with limited training data. We holistically re-examine this task and introduce BAP v2 to address key challenges in evaluation, training data, and modeling. Specifically, we define an enhanced evaluation benchmark, featuring a cleaner test set and fairer, more insightful metrics that also reveal spatial reasoning as the primary performance bottleneck. To address data scarcity and to teach models basic spatial skills, we generate different types of synthetic MCBT data. We observe that current, LLM-based SOTA models trained on the human BAP dialogues fail on these simpler, synthetic BAP ones, but show that training models on this synthetic data improves their performance across the board. We also introduce a new SOTA model, Llama-CRAFTS, which leverages richer input representations, and achieves an F1 score of 53.0 on the BAP v2 task and strong performance on the synthetic data. While this result marks a notable 6 points improvement over previous work, it also underscores the task's remaining difficulty, establishing BAP v2 as a fertile ground for future research, and providing a useful measure of the spatial capabilities of current text-only LLMs in such embodied tasks.

Marisa Hudspeth, Sogol Balali, Cindy Grimm et al.

In this paper we investigate the influence interfaces and feedback have on human-robot trust levels when operating in a shared physical space. The task we use is specifying a "no-go" region for a robot in an indoor environment. We evaluate three styles of interface (physical, AR, and map-based) and four feedback mechanisms (no feedback, robot drives around the space, an AR "fence", and the region marked on the map). Our evaluation looks at both usability and trust. Specifically, if the participant trusts that the robot "knows" where the no-go region is and their confidence in the robot's ability to avoid that region. We use both self-reported and indirect measures of trust and usability. Our key findings are: 1) interfaces and feedback do influence levels of trust; 2) the participants largely preferred a mixed interface-feedback pair, where the modality for the interface differed from the feedback.