Search for: All records

We propose an interactive approach to language learning that utilizes linguistic acceptability judgments from an informant (a competent lan- guage user) to learn a grammar. Given a gram- mar formalism and a framework for synthesiz- ing data, our model iteratively selects or synthe- sizes a data-point according to one of a range of information-theoretic policies, asks the in- formant for a binary judgment, and updates its own parameters in preparation for the next query. We demonstrate the effectiveness of our model in the domain of phonotactics, the rules governing what kinds of sound-sequences are acceptable in a language, and carry out two experiments, one with typologically-natural linguistic data and another with a range of procedurally-generated languages. We find that the information-theoretic policies that our model uses to select items to query the infor- mant achieve sample efficiency comparable to, and sometimes greater than, fully supervised approaches. 

Abstract Previous work has shown that English native speakers interpret sentences as predicted by a noisy‐channel model: They integrate both the real‐world plausibility of the meaning—the prior—and the likelihood that the intended sentence may be corrupted into the perceived sentence. In this study, we test the noisy‐channel model in Mandarin Chinese, a language taxonomically different from English. We present native Mandarin speakers sentences in a written modality (Experiment 1) and an auditory modality (Experiment 2) in three pairs of syntactic alternations. The critical materials are literally implausible but require differing numbers and types of edits in order to form more plausible sentences. Each sentence is followed by a comprehension question that allows us to infer whether the speakers interpreted the item literally, or made an inference toward a more likely meaning. Similar to previous research on related English constructions, Mandarin participants made the most inferences for implausible materials that could be inferred as plausible by deleting a single morpheme or inserting a single morpheme. Participants were less likely to infer a plausible meaning for materials that could be inferred as plausible by making an exchange across a preposition. And participants were least likely to infer a plausible meaning for materials that could be inferred as plausible by making an exchange across a main verb. Moreover, we found more inferences in written materials than spoken materials, possibly a result of a lack of word boundaries in written Chinese. Overall, the fact that the results were so similar to those found in related constructions in English suggests that the noisy‐channel proposal is robust. 

We studied the learnability of English filler-gap dependencies and the “island” constraints on them by assessing the generalizations made by autoregressive (incremental) language models that use deep learning to predict the next word given preceding context. Using factorial tests inspired by experimental psycholinguistics, we found that models acquire not only the basic contingency between fillers and gaps, but also the unboundedness and hierarchical constraints implicated in the dependency. We evaluated a model’s acquisition of island constraints by demonstrating that its expectation for a filler-gap contingency is attenuated within an island environment. Our results provide empirical evidence against the argument from the poverty of the stimulus for this particular structure. 

Behavioral measures of word-by-word reading time provide experimental evidence to test theories of language processing. A-maze is a recent method for measuring incremental sentence processing that can localize slowdowns related to syntactic ambiguities in individual sentences. We adapted A-maze for use on longer passages and tested it on the Natural Stories corpus. Participants were able to comprehend these longer text passages that they read via the Maze task. Moreover, the Maze task yielded useable reaction time data with word predictability effects that were linearly related to surprisal, the same pattern found with other incremental methods. Crucially, Maze reaction times show a tight relationship with properties of the current word, with little spillover of effects from previous words. This superior localization is an advantage of Maze compared with other methods. Overall, we expanded the scope of experimental materials, and thus theoretical questions, that can be studied with the Maze task. 

Abstract The detailed study of eye movements in reading has shed considerable light into how language processing unfolds in real time. Yet eye movements in reading remain inadequately studied in non-native (L2) readers, even though much of the world’s population is multilingual. Here we present a detailed analysis of the quantitative functional influences of word length, frequency, and predictability on eye movement measures in reading in a large, linguistically diverse sample of non-native English readers. We find many similar qualitative effects as in L1 readers, but crucially also a proficiency-sensitive “lexicon-context tradeoff”. The most proficient L2 readers’ eye movements approach an L1 pattern, but as L2 proficiency diminishes, readers’ eye movements become less sensitive to a word’s predictability in context and more sensitive to word frequency, which is context-invariant. This tradeoff supports a rational, experience-dependent account of how context-driven expectations are deployed in L2 language processing. 

Abstract Scalar inferences (SI) are a signature example of how humans interpret language based on unspoken alternatives. While empirical studies have demonstrated that human SI rates are highly variable—both within instances of a single scale, and across different scales—there have been few proposals that quantitatively explain both cross- and within-scale variation. Furthermore, while it is generally assumed that SIs arise through reasoning about unspoken alternatives, it remains debated whether humans reason about alternatives as linguistic forms, or at the level of concepts. Here, we test a shared mechanism explaining SI rates within and across scales: context-driven expectations about the unspoken alternatives. Using neural language models to approximate human predictive distributions, we find that SI rates are captured by the expectedness of the strong scalemate as an alternative. Crucially, however, expectedness robustly predicts cross-scale variation only under a meaning-based view of alternatives. Our results suggest that pragmatic inferences arise from context-driven expectations over alternatives, and these expectations operate at the level of concepts.1