Kanishka Misra

Freda Shi, Xinyun Chen, Kanishka Misra et al. · deepmind

Large language models have achieved impressive performance on various natural language processing tasks. However, so far they have been evaluated primarily on benchmarks where all information in the input context is relevant for solving the task. In this work, we investigate the distractibility of large language models, i.e., how the model problem-solving accuracy can be influenced by irrelevant context. In particular, we introduce Grade-School Math with Irrelevant Context (GSM-IC), an arithmetic reasoning dataset with irrelevant information in the problem description. We use this benchmark to measure the distractibility of cutting-edge prompting techniques for large language models, and find that the model performance is dramatically decreased when irrelevant information is included. We also identify several approaches for mitigating this deficiency, such as decoding with self-consistency and adding to the prompt an instruction that tells the language model to ignore the irrelevant information.

Koustuv Sinha, Jon Gauthier, Aaron Mueller et al. · meta-ai, mit

Targeted syntactic evaluations of language models ask whether models show stable preferences for syntactically acceptable content over minimal-pair unacceptable inputs. Most targeted syntactic evaluation datasets ask models to make these judgements with just a single context-free sentence as input. This does not match language models' training regime, in which input sentences are always highly contextualized by the surrounding corpus. This mismatch raises an important question: how robust are models' syntactic judgements in different contexts? In this paper, we investigate the stability of language models' performance on targeted syntactic evaluations as we vary properties of the input context: the length of the context, the types of syntactic phenomena it contains, and whether or not there are violations of grammaticality. We find that model judgements are generally robust when placed in randomly sampled linguistic contexts. However, they are substantially unstable for contexts containing syntactic structures matching those in the critical test content. Among all tested models (GPT-2 and five variants of OPT), we significantly improve models' judgements by providing contexts with matching syntactic structures, and conversely significantly worsen them using unacceptable contexts with matching but violated syntactic structures. This effect is amplified by the length of the context, except for unrelated inputs. We show that these changes in model performance are not explainable by simple features matching the context and the test inputs, such as lexical overlap and dependency overlap. This sensitivity to highly specific syntactic features of the context can only be explained by the models' implicit in-context learning abilities.

Wesley Scivetti, Ethan Wilcox, Nathan Schneider et al.

Grasping the semantics of rare constructions (form-meaning pairings) has been shown to be a challenging problem that has currently only been solved by the largest LLMs. It remains an open question if open-source models have robust constructional understanding, and if so, what learning dynamics underlie the acquisition of this knowledge. Focusing on a set of rare Paired-Focus constructions in English (e.g. "let alone", "much less"), we construct a novel dataset to test their meanings using both scalar adjectival semantics and general world knowledge. Testing a wide range of models differing in parameter count, architecture, and pretraining dataset size, we find that several modestly sized models are sensitive to both the forms and the meanings of Paired-Focus constructions, though models trained on human-scale data fail at all meaning evaluations. Turning to training dynamics for a set of open-checkpoint models, we find that Paired-Focus understanding emerges later in training than Paired-Focus syntactic knowledge, and that learning of Paired-Focus semantics is correlated with gains in some domains of world knowledge. Overall, our empirical results support the conclusion that modestly sized open-source models can grasp the rare Paired-Focus constructions, and demonstrate a connection between knowledge of Paired-Focus constructions and other meaning domains.

Kaijie Mo, Thomas Yang, Chantal Shaib et al.

The morphological form of a word can often give cues to its meaning, but purely relying on these mappings can lead to overgeneralization in high-stakes domains. In the medical domain, for instance, LLMs can confidently reason about fictitious drugs from their affixes alone (e.g., wugcillin) and generate plausible-looking clinical content. We present a behavioral and mechanistic study of LLM "affix heuristics" in pharmacology. Using fictitious drug names built from real affixes, we show that affix signals alone elicit class-level pharmacological responses. We introduce a framework for identifying whether a model's drug semantics are driven mainly by the affix, the stem, or the drug name as a whole. Applied across 653 drugs, our framework reveals that models often induce drug meaning primarily through affix cues, yet rarely explicitly indicate this reliance, and sometimes incorrectly conflate properties among affix-sharing drugs. Activation patching across models further localizes this behavior to early-mid layers. These findings show that morphological shortcuts pose a subtle but measurable risk to safety.

Kanishka Misra

We present minicons, an open source library that provides a standard API for researchers interested in conducting behavioral and representational analyses of transformer-based language models (LMs). Specifically, minicons enables researchers to apply analysis methods at two levels: (1) at the prediction level -- by providing functions to efficiently extract word/sentence level probabilities; and (2) at the representational level -- by also facilitating efficient extraction of word/phrase level vectors from one or more layers. In this paper, we describe the library and apply it to two motivating case studies: One focusing on the learning dynamics of the BERT architecture on relative grammatical judgments, and the other on benchmarking 23 different LMs on zero-shot abductive reasoning. minicons is available at https://github.com/kanishkamisra/minicons

Anshun Asher Zheng, Kanishka Misra, David I. Beaver et al.

We introduce HERO'S JOURNEY, a benchmark for rule induction in goal-directed episodic tasks, where agents must infer hidden rules from demonstrations and act on them through multi-step execution. HERO'S JOURNEY covers eight tasks across attribute and procedural induction families, each with four structural rule forms, controllable lexical grounding, and identifiability conditions. Evaluating state-of-the-art LLMs, we find that models show evidence of rule induction, but the ability is limited and uneven across tasks. Meanwhile, process execution adds an execution bottleneck for models, whereas surface semantics has minimal effect. Induction-specific steering methods improve performance on attribute tasks but show no reliable gains on procedural tasks, suggesting the gap in procedural induction remains an open challenge.

Kanishka Misra, Julia Taylor Rayz, Allyson Ettinger

A characteristic feature of human semantic cognition is its ability to not only store and retrieve the properties of concepts observed through experience, but to also facilitate the inheritance of properties (can breathe) from superordinate concepts (animal) to their subordinates (dog) -- i.e. demonstrate property inheritance. In this paper, we present COMPS, a collection of minimal pair sentences that jointly tests pre-trained language models (PLMs) on their ability to attribute properties to concepts and their ability to demonstrate property inheritance behavior. Analyses of 22 different PLMs on COMPS reveal that they can easily distinguish between concepts on the basis of a property when they are trivially different, but find it relatively difficult when concepts are related on the basis of nuanced knowledge representations. Furthermore, we find that PLMs can demonstrate behavior consistent with property inheritance to a great extent, but fail in the presence of distracting information, which decreases the performance of many models, sometimes even below chance. This lack of robustness in demonstrating simple reasoning raises important questions about PLMs' capacity to make correct inferences even when they appear to possess the prerequisite knowledge.

Kanishka Misra, Najoung Kim

Neural network language models (LMs) have been shown to successfully capture complex linguistic knowledge. However, their utility for understanding language acquisition is still debated. We contribute to this debate by presenting a case study where we use LMs as simulated learners to derive novel experimental hypotheses to be tested with humans. We apply this paradigm to study cross-dative generalization (CDG): productive generalization of novel verbs across dative constructions (she pilked me the ball/she pilked the ball to me)--acquisition of which is known to involve a large space of contextual features--using LMs trained on child-directed speech. We specifically ask: "what properties of the training exposure facilitate a novel verb's generalization to the (unmodeled) alternate construction?" To answer this, we systematically vary the exposure context in which a novel dative verb occurs in terms of the properties of the theme and recipient, and then analyze the LMs' usage of the novel verb in the unmodeled dative construction. We find LMs to replicate known patterns of children's CDG, as a precondition to exploring novel hypotheses. Subsequent simulations reveal a nuanced role of the features of the novel verbs' exposure context on the LMs' CDG. We find CDG to be facilitated when the first postverbal argument of the exposure context is pronominal, definite, short, and conforms to the prototypical animacy expectations of the exposure dative. These patterns are characteristic of harmonic alignment in datives, where the argument with features ranking higher on the discourse prominence scale tends to precede the other. This gives rise to a novel hypothesis that CDG is facilitated insofar as the features of the exposure context--in particular, its first postverbal argument--are harmonically aligned. We conclude by proposing future experiments that can test this hypothesis in children.

Kanishka Misra, Julia Taylor Rayz, Allyson Ettinger

To what extent can experience from language contribute to our conceptual knowledge? Computational explorations of this question have shed light on the ability of powerful neural language models (LMs) -- informed solely through text input -- to encode and elicit information about concepts and properties. To extend this line of research, we present a framework that uses neural-network language models (LMs) to perform property induction -- a task in which humans generalize novel property knowledge (has sesamoid bones) from one or more concepts (robins) to others (sparrows, canaries). Patterns of property induction observed in humans have shed considerable light on the nature and organization of human conceptual knowledge. Inspired by this insight, we use our framework to explore the property inductions of LMs, and find that they show an inductive preference to generalize novel properties on the basis of category membership, suggesting the presence of a taxonomic bias in their representations.

Kanishka Misra, Cicero Nogueira dos Santos, Siamak Shakeri

Despite readily memorizing world knowledge about entities, pre-trained language models (LMs) struggle to compose together two or more facts to perform multi-hop reasoning in question-answering tasks. In this work, we propose techniques that improve upon this limitation by relying on random walks over structured knowledge graphs. Specifically, we use soft prompts to guide LMs to chain together their encoded knowledge by learning to map multi-hop questions to random walk paths that lead to the answer. Applying our methods on two T5 LMs shows substantial improvements over standard tuning approaches in answering questions that require 2-hop reasoning.

Kanishka Misra, Najoung Kim

Exemplar based accounts are often considered to be in direct opposition to pure linguistic abstraction in explaining language learners' ability to generalize to novel expressions. However, the recent success of neural network language models on linguistically sensitive tasks suggests that perhaps abstractions can arise via the encoding of exemplars. We provide empirical evidence for this claim by adapting an existing experiment that studies how an LM (BERT) generalizes the usage of novel tokens that belong to lexical categories such as Noun/Verb/Adjective/Adverb from exposure to only a single instance of their usage. We analyze the representational behavior of the novel tokens in these experiments, and find that BERT's capacity to generalize to unseen expressions involving the use of these novel tokens constitutes the movement of novel token representations towards regions of known category exemplars in two-dimensional space. Our results suggest that learners' encoding of exemplars can indeed give rise to abstraction like behavior.

Kanishka Misra, Kyle Mahowald

Language models learn rare syntactic phenomena, but the extent to which this is attributable to generalization vs. memorization is a major open question. To that end, we iteratively trained transformer language models on systematically manipulated corpora which were human-scale in size, and then evaluated their learning of a rare grammatical phenomenon: the English Article+Adjective+Numeral+Noun (AANN) construction (``a beautiful five days''). We compared how well this construction was learned on the default corpus relative to a counterfactual corpus in which AANN sentences were removed. We found that AANNs were still learned better than systematically perturbed variants of the construction. Using additional counterfactual corpora, we suggest that this learning occurs through generalization from related constructions (e.g., ``a few days''). An additional experiment showed that this learning is enhanced when there is more variability in the input. Taken together, our results provide an existence proof that LMs can learn rare grammatical phenomena by generalization from less rare phenomena. Data and code: https://github.com/kanishkamisra/aannalysis.

Daniel Brubaker, William Sheffield, Junyi Jessy Li et al.

The role of world knowledge has been particularly crucial to predict the discourse connective that marks the discourse relation between two arguments, with language models (LMs) being generally successful at this task. We flip this premise in our work, and instead study the inverse problem of understanding whether discourse connectives can inform LMs about the world. To this end, we present WUGNECTIVES, a dataset of 8,880 stimuli that evaluates LMs' inferences about novel entities in contexts where connectives link the entities to particular attributes. On investigating 17 different LMs at various scales, and training regimens, we found that tuning an LM to show reasoning behavior yields noteworthy improvements on most connectives. At the same time, there was a large variation in LMs' overall performance across connective type, with all models systematically struggling on connectives that express a concessive meaning. Our findings pave the way for more nuanced investigations into the functional role of language cues as captured by LMs. We release WUGNECTIVES at https://github.com/sheffwb/wugnectives.

Serpil Karabüklü, Kanishka Misra, Shester Gueuwou et al.

Models of sign language have historically lagged behind those for spoken language (text and speech). Recent work has greatly improved their performance on tasks like sign language translation and isolated sign recognition. However, it remains unclear to what extent existing models capture various linguistic phenomena of sign language, and how well they use cues from the multiple articulators used in sign language (hands, upper body, face). We introduce a new benchmark dataset for American Sign Language, ASL Minimal Translation Pairs (ASL-MTP), divided into multiple types of sign language phenomena and corresponding minimal pairs of translations, for performing such linguistic analyses. As a case study, we use ASL-MTP to analyze a state-of-the-art ASL-to-English translation model. We conduct a targeted analysis of the model by ablating various input cues during training and inference and evaluating on the phenomena in ASL-MTP. Our results show that, while the model performs above chance level on most of the phenomena, it relies strongly on manual cues while often missing crucial non-manual cues.

Juan Diego Rodriguez, Aaron Mueller, Kanishka Misra

Property inheritance -- a phenomenon where novel properties are projected from higher level categories (e.g., birds) to lower level ones (e.g., sparrows) -- provides a unique window into how humans organize and deploy conceptual knowledge. It is debated whether this ability arises due to explicitly stored taxonomic knowledge vs. simple computations of similarity between mental representations. How are these mechanistic hypotheses manifested in contemporary language models? In this work, we investigate how LMs perform property inheritance with behavioral and causal representational analysis experiments. We find that taxonomy and categorical similarities are not mutually exclusive in LMs' property inheritance behavior. That is, LMs are more likely to project novel properties from one category to the other when they are taxonomically related and at the same time, highly similar. Our findings provide insight into the conceptual structure of language models and may suggest new psycholinguistic experiments for human subjects.

Kanishka Misra, Allyson Ettinger, Kyle Mahowald

Recent zero-shot evaluations have highlighted important limitations in the abilities of language models (LMs) to perform meaning extraction. However, it is now well known that LMs can demonstrate radical improvements in the presence of experimental contexts such as in-context examples and instructions. How well does this translate to previously studied meaning-sensitive tasks? We present a case-study on the extent to which experimental contexts can improve LMs' robustness in performing property inheritance -- predicting semantic properties of novel concepts, a task that they have been previously shown to fail on. Upon carefully controlling the nature of the in-context examples and the instructions, our work reveals that they can indeed lead to non-trivial property inheritance behavior in LMs. However, this ability is inconsistent: with a minimal reformulation of the task, some LMs were found to pick up on shallow, non-semantic heuristics from their inputs, suggesting that the computational principles of semantic property inference are yet to be mastered by LMs.

Yulu Qin, Dheeraj Varghese, Adam Dahlgren Lindström et al.

Does vision-and-language (VL) training change the linguistic representations of language models in meaningful ways? Most results in the literature have shown inconsistent or marginal differences, both behaviorally and representationally. In this work, we start from the hypothesis that the domain in which VL training could have a significant effect is lexical-conceptual knowledge, in particular its taxonomic organization. Through comparing minimal pairs of text-only LMs and their VL-trained counterparts, we first show that the VL models often outperform their text-only counterparts on a text-only question-answering task that requires taxonomic understanding of concepts mentioned in the questions. Using an array of targeted behavioral and representational analyses, we show that the LMs and VLMs do not differ significantly in terms of their taxonomic knowledge itself, but they differ in how they represent questions that contain concepts in a taxonomic relation vs. a non-taxonomic relation. This implies that the taxonomic knowledge itself does not change substantially through additional VL training, but VL training does improve the deployment of this knowledge in the context of a specific task, even when the presentation of the task is purely linguistic.

William Sheffield, Kanishka Misra, Valentina Pyatkin et al. · allen-ai

Discourse particles are crucial elements that subtly shape the meaning of text. These words, often polyfunctional, give rise to nuanced and often quite disparate semantic/discourse effects, as exemplified by the diverse uses of the particle "just" (e.g., exclusive, temporal, emphatic). This work investigates the capacity of LLMs to distinguish the fine-grained senses of English "just", a well-studied example in formal semantics, using data meticulously created and labeled by expert linguists. Our findings reveal that while LLMs exhibit some ability to differentiate between broader categories, they struggle to fully capture more subtle nuances, highlighting a gap in their understanding of discourse particles.

Qing Yao, Kanishka Misra, Leonie Weissweiler et al.

Language models (LMs) tend to show human-like preferences on a number of syntactic phenomena, but the extent to which these are attributable to direct exposure to the phenomena or more general properties of language is unclear. We explore this with the English dative alternation (DO: "gave Y the X" vs. PO: "gave the X to Y"), using a controlled rearing paradigm wherein we iteratively train small LMs on systematically manipulated input. We focus on two properties that affect the choice of alternant: length and animacy. Both properties are directly present in datives but also reflect more global tendencies for shorter elements to precede longer ones and animates to precede inanimates. First, by manipulating and ablating datives for these biases in the input, we show that direct evidence of length and animacy matters, but easy-first preferences persist even without such evidence. Then, using LMs trained on systematically perturbed datasets to manipulate global length effects (re-linearizing sentences globally while preserving dependency structure), we find that dative preferences can emerge from indirect evidence. We conclude that LMs' emergent syntactic preferences come from a mix of direct and indirect sources.

Sriram Padmanabhan, Siyuan Song, Kanishka Misra

Language places subtle constraints on how we make inductive inferences. Developmental evidence by Gelman et al. (2002) has shown children (4 years and older) to differentiate among generic statements ("Bears are daxable"), universally quantified NPs ("all bears are daxable") and indefinite plural NPs ("some bears are daxable") in extending novel properties to a specific member (all > generics > some), suggesting that they represent these types of propositions differently. We test if these subtle differences arise in general purpose statistical learners like Vision Language Models, by replicating the original experiment. On tasking them through a series of precondition tests (robust identification of categories in images and sensitivities to all and some), followed by the original experiment, we find behavioral alignment between models and humans. Post-hoc analyses on their representations revealed that these differences are organized based on inductive constraints and not surface-form differences.

Tianyang Xu, Marcelo Sandoval-Castaneda, Karen Livescu et al.

What is the interplay between semantic representations learned by language models (LM) from surface form alone to those learned from more grounded evidence? We study this question for a scenario where part of the input comes from a different modality -- in our case, in a vision-language model (VLM), where a pretrained LM is aligned with a pretrained image encoder. As a case study, we focus on the task of predicting hypernyms of objects represented in images. We do so in a VLM setup where the image encoder and LM are kept frozen, and only the intermediate mappings are learned. We progressively deprive the VLM of explicit evidence for hypernyms, and test whether knowledge of hypernyms is recoverable from the LM. We find that the LMs we study can recover this knowledge and generalize even in the most extreme version of this experiment (when the model receives no evidence of a hypernym during training). Additional experiments suggest that this cross-modal taxonomic generalization persists under counterfactual image-label mappings only when the counterfactual data have high visual similarity within each category. Taken together, these findings suggest that cross-modal generalization in LMs arises as a result of both coherence in the extralinguistic input and knowledge derived from language cues.

Junyi Jessy Li, Yang Janet Liu, Kanishka Misra et al.

The field of NLP has undergone vast, continuous transformations over the past few years, sparking debates going beyond discipline boundaries. This begs important questions in education: how do we design courses that bridge sub-disciplines in this shifting landscape? This paper explores this question from the angle of discourse processing, an area with rich linguistic insights and computational models for the intentional, attentional, and coherence structure of language. Discourse is highly relevant for open-ended or long-form text generation, yet this connection is under-explored in existing undergraduate curricula. We present a new course, "Computational Discourse and Natural Language Generation". The course is collaboratively designed by a team with complementary expertise and was offered for the first time in Fall 2025 as an upper-level undergraduate course, cross-listed between Linguistics and Computer Science. Our philosophy is to deeply integrate the theoretical and empirical aspects, and create an exploratory mindset inside the classroom and in the assignments. This paper describes the course in detail and concludes with takeaways from an independent survey as well as our vision for future directions.

Sanghee J. Kim, Kanishka Misra

Evaluating the naturalness of dialogue in language models (LMs) is not trivial: notions of 'naturalness' vary, and scalable quantitative metrics remain limited. This study leverages the linguistic notion of 'at-issueness' to assess dialogue naturalness and introduces a new method: Divide, Generate, Recombine, and Compare (DGRC). DGRC (i) divides a dialogue as a prompt, (ii) generates continuations for subparts using LMs, (iii) recombines the dialogue and continuations, and (iv) compares the likelihoods of the recombined sequences. This approach mitigates bias in linguistic analyses of LMs and enables systematic testing of discourse-sensitive behavior. Applying DGRC, we find that LMs prefer to continue dialogue on at-issue content, with this effect enhanced in instruct-tuned models. They also reduce their at-issue preference when relevant cues (e.g., "Hey, wait a minute") are present. Although instruct-tuning does not further amplify this modulation, the pattern reflects a hallmark of successful dialogue dynamics.

Sriram Padmanabhan, Kanishka Misra, Kyle Mahowald et al.

Humans are sensitive to suspicious coincidences when generalizing inductively over data, as they make assumptions as to how the data was sampled. This results in smaller, more specific hypotheses being favored over more general ones. For instance, when provided the set {Austin, Dallas, Houston}, one is more likely to think that this is sampled from "Texas Cities" over "US Cities" even though both are compatible. Suspicious coincidence is strongly connected to pragmatic reasoning, and can serve as a testbed to analyze systems on their sensitivity towards the communicative goals of the task (i.e., figuring out the true category underlying the data). In this paper, we analyze whether suspicious coincidence effects are reflected in language models' (LMs) behavior. We do so in the context of two domains: 1) the number game, where humans made judgments of whether a number (e.g., 4) fits a list of given numbers (e.g., 16, 32, 2); and 2) by extending the number game setup to prominent cities. For both domains, the data is compatible with multiple hypotheses and we study which hypothesis is most consistent with the models' behavior. On analyzing five models, we do not find strong evidence for suspicious coincidences in LMs' zero-shot behavior. However, when provided access to the hypotheses space via chain-of-thought or explicit prompting, LMs start to show an effect resembling suspicious coincidences, sometimes even showing effects consistent with humans. Our study suggests that inductive reasoning behavior in LMs can be enhanced with explicit access to the hypothesis landscape.

Kanishka Misra

My doctoral research focuses on understanding semantic knowledge in neural network models trained solely to predict natural language (referred to as language models, or LMs), by drawing on insights from the study of concepts and categories grounded in cognitive science. I propose a framework inspired by 'inductive reasoning,' a phenomenon that sheds light on how humans utilize background knowledge to make inductive leaps and generalize from new pieces of information about concepts and their properties. Drawing from experiments that study inductive reasoning, I propose to analyze semantic inductive generalization in LMs using phenomena observed in human-induction literature, investigate inductive behavior on tasks such as implicit reasoning and emergent feature recognition, and analyze and relate induction dynamics to the learned conceptual representation space.

Kanishka Misra, Allyson Ettinger, Julia Taylor Rayz

Building on research arguing for the possibility of conceptual and categorical knowledge acquisition through statistics contained in language, we evaluate predictive language models (LMs) -- informed solely by textual input -- on a prevalent phenomenon in cognitive science: typicality. Inspired by experiments that involve language processing and show robust typicality effects in humans, we propose two tests for LMs. Our first test targets whether typicality modulates LM probabilities in assigning taxonomic category memberships to items. The second test investigates sensitivities to typicality in LMs' probabilities when extending new information about items to their categories. Both tests show modest -- but not completely absent -- correspondence between LMs and humans, suggesting that text-based exposure alone is insufficient to acquire typicality knowledge.

Kanishka Misra, Julia Taylor Rayz

Humans often communicate by using imprecise language, suggesting that fuzzy concepts with unclear boundaries are prevalent in language use. In this paper, we test the extent to which models trained to capture the distributional statistics of language show correspondence to fuzzy-membership patterns. Using the task of natural language inference, we test a recent state of the art model on the classical case of temperature, by examining its mapping of temperature data to fuzzy-perceptions such as "cool", "hot", etc. We find the model to show patterns that are similar to classical fuzzy-set theoretic formulations of linguistic hedges, albeit with a substantial amount of noise, suggesting that models trained solely on language show promise in encoding fuzziness.

Qingyuan Hu, Yi Zhang, Kanishka Misra et al.

Natural Language Inference (NLI) or Recognizing Textual Entailment (RTE) is the task of predicting the entailment relation between a pair of sentences (premise and hypothesis). This task has been described as a valuable testing ground for the development of semantic representations, and is a key component in natural language understanding evaluation benchmarks. Models that understand entailment should encode both, the premise and the hypothesis. However, experiments by Poliak et al. revealed a strong preference of these models towards patterns observed only in the hypothesis, based on a 10 dataset comparison. Their results indicated the existence of statistical irregularities present in the hypothesis that bias the model into performing competitively with the state of the art. While recast datasets provide large scale generation of NLI instances due to minimal human intervention, the papers that generate them do not provide fine-grained analysis of the potential statistical patterns that can bias NLI models. In this work, we analyze hypothesis-only models trained on one of the recast datasets provided in Poliak et al. for word-level patterns. Our results indicate the existence of potential lexical biases that could contribute to inflating the model performance.

Kanishka Misra, Allyson Ettinger, Julia Taylor Rayz

Models trained to estimate word probabilities in context have become ubiquitous in natural language processing. How do these models use lexical cues in context to inform their word probabilities? To answer this question, we present a case study analyzing the pre-trained BERT model with tests informed by semantic priming. Using English lexical stimuli that show priming in humans, we find that BERT too shows "priming," predicting a word with greater probability when the context includes a related word versus an unrelated one. This effect decreases as the amount of information provided by the context increases. Follow-up analysis shows BERT to be increasingly distracted by related prime words as context becomes more informative, assigning lower probabilities to related words. Our findings highlight the importance of considering contextual constraint effects when studying word prediction in these models, and highlight possible parallels with human processing.