Roger Philip Levy

An essential feature of human communication is robustness to errors: even when speakers and writers make mistakes, listeners and readers can usually determine the intended meaning. A leading hypothesis is that comprehenders achieve robust error correction by internally deploying a “noisy‐channel” model of the language production process. Crucially, to best achieve error robustness, a comprehender's noisy‐channel model should be finely calibrated to the environmental statistics of language production, incorporating not only prior expectations about communicative intents but also structure‐sensitive likelihoods of different types of speaker error. This environmentally calibrated noisy‐channel hypothesis has been difficult to test, because the rates of most types of language production errors are hard to estimate. Here, we test this hypothesis by experimentally investigating comprehenders' interpretations of subject–verb agreement errors, for which context‐contingent language production error rates are feasible to estimate. In a novel free‐form error correction task, participants edited sentences with subject–verb agreement mismatches to indicate what they think was intended. Bayesian analysis of the resulting data shows that comprehenders' interpretations reflect both item‐specific prior expectations and structure‐sensitive error likelihoods that closely correspond to error rates in language production. These results provide quantitative evidence that humans closely track prior statistics over linguistic forms and deploy a noisy‐channel model finely calibrated to statistics of the linguistic environment to achieve robust language understanding.

Human language comprehension is remarkably robust---we are good at grasping what speakers and writers intend and can even understand them when their productions are not perfectly well-formed. But it is also susceptible to breakdown: we sometimes don’t reach any possible interpretation of the linguistic input and even arrive at interpretations which are incompatible with the literal meaning. We argue that the most promising overall theory of both robustness and breakdown in language comprehension and processing is the Noisy Channel theory, which integrates the roles of probabilistic inference and cognitive (e.g., memory) constraints in language understanding and provides a unified account across many levels of language representation. We articulate the key components and commitments of this theory and review the substantial body of evidence which is consistent with its predictions. We further identify distinctive novel predictions of the Noisy Channel theory and show that it often provides a straightforward explanation for long-standing puzzles. We conclude by reviewing open questions and future directions for the theory.

An essential feature of human communication is robustness to errors: even when speakers and writers make mistakes, listeners and readers can usually determine the intended meaning. A leading hypothesis is that comprehenders achieve robust error correction by internally deploying a "noisy-channel" model of the language production process. Crucially, to best achieve error robustness, a comprehender's noisy-channel model should be finely calibrated to the environmental statistics of language production, incorporating not only prior expectations about communicative intents but also structure-sensitive likelihoods of different types of speaker error. This environmentally-calibrated noisy-channel hypothesis has been difficult to test, because the rates of most types of language production errors are hard to estimate. Here we test this hypothesis by experimentally investigating comprehenders' interpretations of subject-verb agreement errors, for which context-contingent language production error rates are feasible to estimate. In a novel free-form error correction task, participants edited sentences with subject-verb agreement mismatches to indicate what they think was intended. Bayesian analysis of the resulting data shows that comprehenders' interpretations reflect both item-specific prior expectations and structure-sensitive error likelihoods that closely correspond to error rates in language production. These results provide quantitative evidence that humans closely track prior statistics over linguistic forms and deploy a noisy-channel model finely calibrated to statistics of the linguistic environment to achieve robust language understanding.

We present OneStop Eye Movements, a large-scale corpus of eye movements in reading, in which native (L1) speakers read newswire texts in English and answer reading comprehension questions. OneStop has 152 hours of eye movement recordings from 360 participants for 2.6 million word tokens, more data than all the existing public broad coverage English L1 eye tracking datasets combined. The eye movement data was collected for extensively piloted reading comprehension materials comprising 486 reading comprehension questions and auxiliary text annotations geared towards behavioral analyses of reading comprehension. Importantly, OneStop includes multiple reading regimes: ordinary reading, information seeking, repeated reading of the same text, and reading simplified text. The combination of the unprecedented size, high-quality reading comprehension materials and multiple reading scenarios, aims to enable new research avenues in the study of reading and human language processing. It further aims to facilitate the integration of eye tracking data in Natural Language Processing (NLP), Artificial Intelligence (AI), Human Computer Interaction (HCI) and educational applications.

We present OneStop Eye Movements, a large-scale corpus of eye movements in reading, in which native speakers read newswire texts in English and answer reading comprehension questions. OneStop has 152 hours of passage reading eye movement recordings from 360 participants for 2.6 million word tokens, more data than all the existing public broad coverage native English eye tracking datasets combined. The eye movement data was collected for extensively piloted reading comprehension materials comprising 486 reading comprehension questions and auxiliary text annotations geared towards behavioral analyses of reading comprehension. OneStop includes multiple reading regimes: ordinary reading, information seeking, repeated reading of the same text, and reading simplified text. The combination of the unprecedented size, high-quality reading comprehension materials and multiple reading regimes, aims to enable new research avenues in the study of reading and human language processing. It further aims to facilitate the integration of eye tracking data in Natural Language Processing (NLP), Artificial Intelligence (AI), Human Computer Interaction (HCI) and educational applications.

Human spoken language uses a continuous stream of acoustic signals to communicate about continuous features of the world, by using discrete forms — words — that segment the world into categories. Here we investigate how discreteness (the segmentation of a continuous signal space into discrete forms) and systematicity (the consistent alignment of these forms with what they refer to in the world) can emerge under communicative pressure. In an exploratory study, participants were paired with one another and played a game in which they varied the pitch of auditory signals to communicate about a continuous color space, generalizing from a small, shared set of signal-color pairings. The emergent systems exhibited both discreteness and systematicity, but only systematicity robustly predicted successful communication. These findings offer insight into the cognitive strategies that could support the creation and evolution of language, highlighting how pressures for effective communication can shape continuous signal spaces into structured, learnable systems.

A central endeavor in psycholinguistic research has been to determine the processing profile of syntactically ambiguous strings. Previous work investigating syntactic attachment ambiguities has shown that discarding a locally grammatically available, but globally failing, parse is costly. However, little is known about how comprehenders cope with semantic parsing ambiguities. Using the case study of scopally ambiguous definite descriptions such as the rabbit in the big hat, we examine whether comparable penalties arise for non-lexical semantic ambiguities. In a series of reference resolution tasks, we find dispreference for strings that are globally defined but fail to refer under alternative semantic parses, compared to strings where all readings successfully refer to the same individual. Crucially, this effect is only detectable when the alternative failing reading gives rise to a REFERENTIAL GARDEN PATH, where a dynamic constraint evaluation process temporarily settles on a unique referent before eventually failing. We conclude that failing alternative readings cause dispreference for a definite description, but only when the failing interpretation constitutes a red herring.

Human spoken language uses a continuous stream of acoustic signals to communicate about continuous features of the world, by using discrete forms—words—that segment the world into categories. Here we investigate how discreteness (the segmentation of a continuous signal space into discrete forms) and systematicity (the consistent alignment of these forms with what they refer to in the world) can emerge under communicative pressure. In an exploratory study, participants were paired with one another and played a game in which they varied the pitch of auditory signals to communicate about a continuous color space, generalizing from a small, shared set of signal-color pairings. The emergent systems exhibited both discreteness and systematicity, but only systematicity robustly predicted successful communication. These findings offer insight into the cognitive strategies that could support the creation and evolution of language, highlighting how pressures for effective communication can shape continuous signal spaces into structured, learnable systems.

Sometimes sentences sound acceptable when they are ungrammatical or semanticallyimplausible. In this paper, we study “comparative illusion” (CI) sentences where people oftenrate a sentence like More people have been to Russia than I have to be acceptable while in factit is semantically anomalous. We provide a potential explanation for this language illusion fromthe noisy-channel framework. We hypothesize that comprehenders make rational inferencesover the perceived sentence by entertaining alternative “close” plausible interpretations, wherecloseness is determined by possible production errors. In four experiments, (1) we identified alinguistic construction that elicits a salient CI illusion effect; (2) we established a range ofplausible interpretations of the CI sentence; and (3) we found that the probability forcomprehenders to assign a certain plausible interpretation to the CI sentence is proportional tohow likely they think that that interpretation is to be produced as the CI sentence during noisylanguage communication.