Department of Linguistics

Comprehenders in many of the world's languages exhibit a preference or a greater ease in comprehending transitive relative clauses (RCs) when associating the dislocated head with a subject position. Some theories relate this preference to an early prediction that an animate head will serve as a subject. We present a picture selection experiment with eye-tracking investigating the role of such predictions in comprehending transitive RCs in Santiago Laxopa Zapotec, an Oto-Manguean language of southern Mexico where RCs are ambiguous unless they contain a resumptive pronoun. Patterns of offline choices and incremental looking suggest that Santiago Laxopa Zapotec comprehenders avoid forming predictive interpretations of the RC head, and do not take its animacy into account in early processing. Instead, comprehenders exhibit a later, animacy-sensitive bias to interpret the RC-internal co-argument as a subject. Overall, we take this pattern as evidence that a comprehender's structural predictions must be somehow dependent on experience-based tuning, while more universal pressures like similarity-based interference remain fixed. The lingering question for theories of cross-linguistic processing is how much of our predictive mechanism is tunable: do we tune just the distribution of expected structures which directs predictions, or also the practice of deploying predictions itself? We discuss how either of these approaches might explain the pattern we observe in Santiago Laxopa Zapotec, and highlight that an answer to this question will depend on continued investigations across a diverse sample of the world's languages.

In Gricean pragmatics, inference during communication is regarded as a form of rational, domain-general reasoning about the intentions of other agents. Studies using the pictorial communication "reference game" task are sometimes used in support of this hypothesis. Yet, measures of pragmatic reasoning in this task sometimes reveal poor performance, with participants requiring many rounds of play before they exhibit patterns which match Gricean inferences, and demonstrating substantial individual differences in behavior. Do these results challenge the idea of widespread inferencing via fundamental social competence? We advance an alternative proposal here, which posits that these patterns emerge as a factor of the way participants perform pragmatic reasoning in a task: namely, they prefer to use simpler interpretation strategies until experience motivates the use of additional resources. Building off of work modeling task adaptation as reinforcement learning, we use the cognitive architecture ACT-R to simulate the expected behavior of individuals with this kind of resource-rational performance algorithm, subject to individualized parameters for reinforcement learning. These simulations provide a proof-of-concept for our adaptation proposal, recreating known patterns and generating new concrete predictions for the particular domain-general sources of individual variance in reference game tasks. We then go on to validate some of these new predictions in a pre-registered experiment, and find that pragmatic response behavior is indeed related to a participant's general persistence in self-directed exploration of strategies for task completion. Our results offer a path to reconcile variable empirical data with models of core pragmatic competence. From a broader perspective, we see this as an important step towards more robust theories of performance factors in pragmatic reasoning, and ultimately, a case study in the value of process-level computational modeling.

In 2025, psycholinguistic research has the benefit of large, high-quality datasets of human behavior, and massively-scalable metrics for variables of interest like frequency and association. This means we have more data than ever before to shed light on classic language processing phenomena like associative priming. But in order to build and test rigorous theories against this data, we also need computational modeling tools that can simulate cognitive mechanisms and generate quantitative predictions at the same scale. In this paper, we assemble one such case, adapting the ACT-R cognitive modeling framework to make use of association metrics derived from language model embeddings, in service of a scalable model of associative priming in the Lexical Decision Task. ACT-R implements a model of memory retrieval that can use itemwise predictors like frequency and association to predict task response times (RTs), via interpretable and meaningfully-parameterized components like spreading activation. But currently, ACT-R's spreading activation calculations rely on manually-coded similarity scores, which are labor-intensive and prone to inaccuracies, particularly for large vocabularies. In this study, we replace these hand-coded associations with cosine similarity scores derived from Word2Vec and BERT embeddings, thereby improving both scalability and predictive accuracy while retaining ACT-R's interpretability. We compare various versions of our model against observed human RTs from the Semantic Priming Project dataset, observing impressive item-wise prediction accuracy, and achieving the strongest alignment with a model where spreading activation is penalized via a scalable approximation of the classic “fan effect.” These findings provide a proof of concept for integrating embedding-based representations into algorithmic-level models of language processing. More than an insight into models of priming, we see this as a first step toward scalable and specific models of more complex phenomena.

Relative clauses can range over degrees, just as they can range over individuals. But this class of degree relatives is rarely studied uniformly across the diverse constructions they help form. As a result, there have been a number of construction-specific proposals to model the semantics of degree relatives: in equatives, amount relatives, and wh -exclamatives. The goals of this article are to review a standard semantics of relativization writ large, to supplement it with standard assumptions from degree semantics, and to explain how the previously strange behavior of a variety of constructions formed from degree relatives comes out as a natural consequence of a handful of straightforward assumptions. Specifically, I argue that degree relatives are just relative clauses that range over degrees, and that these degree readings of relative clauses are available wherever we have ( a ) the appropriate morphology (e.g., a relativizer that can range over degrees), ( b ) a context of utterance that makes salient some informative (monotonic) dimension of measurement, and ( c ) an individual referent that is associated with a single, determinate measure along that dimension in the context of utterance.

In some discourses, a given inference would be natural at one point, but impossible at the next. We focus here on such cases with inferences of temporal and causal order usually described as ambiguities of discourse coherence. The fact that these inferences come and go over the course of a discourse raises challenges for their representation within theories of dynamic semantics: representations must either include coherence-related meaning in a way that can be selectively edited later, or else persistently underspecify it. As theories of humans' actual cognitive states, these approaches make different predictions for patterns of difficulty during real-time comprehension. Borrowing a standard assumption from sentence processing, true incremental representation should be associated with some detectable "Regret" when a comprehender meets input incompatible with a previously-preferred analysis. Previous work on the processing of coherence has failed to demonstrate Regret, leaving underspecification on the table. In a new study, we probe further, by constructing discourses with a strong, verified bias towards an initial inference, and testing the effects of cues which can reverse this inference early in the discourse, or much later. We observe that these discourses provoke Regret in self-paced reading, in particular when the critical cue comes late, in a discourse unit following the initial ambiguity. Among many formal accounts which could explain this pattern, we elaborate in particular on a version of Segmented Discourse Representation Theory where analyses are ranked at the offset of each discourse unit.

This study investigates young school-aged children's knowledge (at 4-7 years) of accurate English word-initial onset clusters. By this age, we expect children to be mostly accurate in producing #CC clusters (rather than repairing them with deletion or epenthesis). We ask how well can they recognize and reject cluster repair errors, in both real and nonce word tasks. The results suggest that these learners' cluster judgment skills lag behind their cluster production abilities, and that asymmetries in error types do not overall align between the two domains. Perceptual errors are made most often when comparing clusters with epenthesis repairs, not deletion, and the cluster's sonority profile does not directly influence error rates. After comparing these findings with similar results from adult L2 English speakers as well, we discuss the ways in which issues like recoverability, salience, and contiguity can account for our findings. We also suggest that more work on phonological knowledge and judgments in older children will provide a broader understanding of sound pattern acquisition across development.

Recent research has shown that 6-month-olds relate novel words suffixed with -s, like babs, that are embedded in passages, with just the stem bab, demonstrating an early sensitivity to morphological relatedness. This study builds on these findings by investigating the role of allomorphy in early morphological acquisition. We tested whether infants relate novel words suffixed with [-z] and [-s] allomorphs of the -s suffix and their stems. We find that English-learning 6-month-olds relate novel words suffixed with the [-z], but not [-s], allomorph with stems, providing evidence for an acquisition trajectory where infants discover morphemes one allomorph at a time.

This paper describes a process of laryngeal reduction in San Mart´ın Peras Mixtec (SMPM; ISO: jmx), an Otomanguean language spoken in Oaxaca and by diasporic communities throughout Mexico and the US. In this process, roots containing a laryngealized vowel often appear in a highly reduced form in fast speech. Laryngeal reduction is gradient, dependent on speech rate, and lacks a phonologically-defined conditioning environment, giving it the characteristics of a phonetic process. However, it is at least sometimes correlated with a phonological process of mora deletion, as evidenced by the fact that some highly reduced laryngealized roots—but no unreduced laryngealized roots—undergo a phonolog-ical tone sandhi alternation that applies only to mono-moraic rising tones. The phonological process of mora deletion is argued to be conditioned by the same phonetic factors that drive laryngeal reduction, constituting an instance of a phonological process triggered by purportedly phonetic factors.

Abstract: Most second language acquisition (SLA) research has documented the processes involved in learning second/foreign languages, with few studies focusing on the durability of attained second language (L2) skills once instructed learners/users are no longer receiving formal instruction. The current study examines the effects of continued exposure and peak instructional attainment on the long‐term evolution of advanced, instructed L2 learners’ skills following a longitudinal mixed‐methods research design. Participants (n = 28) completed an oral proficiency test, an oral interview, and a vocabulary knowledge test at multiple times over an 8‐year period, 6 years of which were postinstruction. Results showed that continued exposure contributes to long‐term retention (and some further development) of oral proficiency and fluency and that peak attainment at the end of formal instruction is also an important variable for some areas of L2 performance. Additionally, even the participants with limited exposure demonstrated little attrition over time.