Roger P. Levy

Songlin Yang, Roger P. Levy, Yoon Kim · mit

We study grammar induction with mildly context-sensitive grammars for unsupervised discontinuous parsing. Using the probabilistic linear context-free rewriting system (LCFRS) formalism, our approach fixes the rule structure in advance and focuses on parameter learning with maximum likelihood. To reduce the computational complexity of both parsing and parameter estimation, we restrict the grammar formalism to LCFRS-2 (i.e., binary LCFRS with fan-out two) and further discard rules that require O(n^6) time to parse, reducing inference to O(n^5). We find that using a large number of nonterminals is beneficial and thus make use of tensor decomposition-based rank-space dynamic programming with an embedding-based parameterization of rule probabilities to scale up the number of nonterminals. Experiments on German and Dutch show that our approach is able to induce linguistically meaningful trees with continuous and discontinuous structures

Ethan Gotlieb Wilcox, Tiago Pimentel, Clara Meister et al. · cambridge, harvard

A fundamental result in psycholinguistics is that less predictable words take a longer time to process. One theoretical explanation for this finding is Surprisal Theory (Hale, 2001; Levy, 2008), which quantifies a word's predictability as its surprisal, i.e. its negative log-probability given a context. While evidence supporting the predictions of Surprisal Theory have been replicated widely, most have focused on a very narrow slice of data: native English speakers reading English texts. Indeed, no comprehensive multilingual analysis exists. We address this gap in the current literature by investigating the relationship between surprisal and reading times in eleven different languages, distributed across five language families. Deriving estimates from language models trained on monolingual and multilingual corpora, we test three predictions associated with surprisal theory: (i) whether surprisal is predictive of reading times; (ii) whether expected surprisal, i.e. contextual entropy, is predictive of reading times; (iii) and whether the linking function between surprisal and reading times is linear. We find that all three predictions are borne out crosslinguistically. By focusing on a more diverse set of languages, we argue that these results offer the most robust link to-date between information theory and incremental language processing across languages.

Tiwalayo Eisape, Vineet Gangireddy, Roger P. Levy et al.

Next-word predictions from autoregressive neural language models show remarkable sensitivity to syntax. This work evaluates the extent to which this behavior arises as a result of a learned ability to maintain implicit representations of incremental syntactic structures. We extend work in syntactic probing to the incremental setting and present several probes for extracting incomplete syntactic structure (operationalized through parse states from a stack-based parser) from autoregressive language models. We find that our probes can be used to predict model preferences on ambiguous sentence prefixes and causally intervene on model representations and steer model behavior. This suggests implicit incremental syntactic inferences underlie next-word predictions in autoregressive neural language models.

James A. Michaelov, Catherine Arnett, Tyler A. Chang et al. · mit

How does the extent to which a model is open or closed impact the scientific inferences that can be drawn from research that involves it? In this paper, we analyze how restrictions on information about model construction and deployment threaten reliable inference. We argue that current closed models are generally ill-suited for scientific purposes, with some notable exceptions, and discuss ways in which the issues they present to reliable inference can be resolved or mitigated. We recommend that when models are used in research, potential threats to inference should be systematically identified along with the steps taken to mitigate them, and that specific justifications for model selection should be provided.

Stephan C. Meylan, Ruthe Foushee, Nicole H. Wong et al.

Children's early speech often bears little resemblance to that of adults, and yet parents and other caregivers are able to interpret that speech and react accordingly. Here we investigate how these adult inferences as listeners reflect sophisticated beliefs about what children are trying to communicate, as well as how children are likely to pronounce words. Using a Bayesian framework for modeling spoken word recognition, we find that computational models can replicate adult interpretations of children's speech only when they include strong, context-specific prior expectations about the messages that children will want to communicate. This points to a critical role of adult cognitive processes in supporting early communication and reveals how children can actively prompt adults to take actions on their behalf even when they have only a nascent understanding of the adult language. We discuss the wide-ranging implications of the powerful listening capabilities of adults for theories of first language acquisition.

James A. Michaelov, Roger P. Levy · mit

Recent work has found that contemporary language models such as transformers can become so good at next-word prediction that the probabilities they calculate become worse for predicting reading time. In this paper, we propose that this can be explained by reading time being sensitive to simple n-gram statistics rather than the more complex statistics learned by state-of-the-art transformer language models. We demonstrate that the neural language models whose predictions are most correlated with n-gram probability are also those that calculate probabilities that are the most correlated with eye-tracking-based metrics of reading time on naturalistic text.

Sophia Zhi, Roger P. Levy, Stephan C. Meylan

One of the many tasks facing the typically-developing child language learner is learning to discriminate between the distinctive sounds that make up words in their native language. Here we investigate whether multimodal information--specifically adult speech coupled with video frames of speakers' faces--benefits a computational model of phonetic learning. We introduce a method for creating high-quality synthetic videos of speakers' faces for an existing audio corpus. Our learning model, when both trained and tested on audiovisual inputs, achieves up to a 8.1% relative improvement on a phoneme discrimination battery compared to a model trained and tested on audio-only input. It also outperforms the audio model by up to 3.9% when both are tested on audio-only data, suggesting that visual information facilitates the acquisition of acoustic distinctions. Visual information is especially beneficial in noisy audio environments, where an audiovisual model closes 67% of the loss in discrimination performance of the audio model in noise relative to a non-noisy environment. These results demonstrate that visual information benefits an ideal learner and illustrate some of the ways that children might be able to leverage visual cues when learning to discriminate speech sounds.

Yingshan Susan Wang, Linlu Qiu, Zhaofeng Wu et al.

Grammaticality and likelihood are distinct notions in human language. Pretrained language models (LMs), which are probabilistic models of language fitted to maximize corpus likelihood, generate grammatically well-formed text and discriminate well between grammatical and ungrammatical sentences in tightly controlled minimal pairs. However, their string probabilities do not sharply discriminate between grammatical and ungrammatical sentences overall. But do LMs implicitly acquire a grammaticality distinction distinct from string probability? We explore this question through studying internal representations of LMs, by training a linear probe on a dataset of grammatical and (synthetic) ungrammatical sentences obtained by applying perturbations to a naturalistic text corpus. We find that this simple grammaticality probe generalizes to human-curated grammaticality judgment benchmarks and outperforms LM probability-based grammaticality judgments. When applied to semantic plausibility benchmarks, in which both members of a minimal pair are grammatical and differ in only plausibility, the probe however performs worse than string probability. The English-trained probe also exhibits nontrivial cross-lingual generalization, outperforming string probabilities on grammaticality benchmarks in numerous other languages. Additionally, probe scores correlate only weakly with string probabilities. These results collectively suggest that LMs acquire to some extent an implicit grammaticality distinction within their hidden layers.

Guangyuan Jiang, Matthias Hofer, Jiayuan Mao et al.

One hallmark of human language is its combinatoriality -- reusing a relatively small inventory of building blocks to create a far larger inventory of increasingly complex structures. In this paper, we explore the idea that combinatoriality in language reflects a human inductive bias toward representational efficiency in symbol systems. We develop a computational framework for discovering structure in a writing system. Built on top of state-of-the-art library learning and program synthesis techniques, our computational framework discovers known linguistic structures in the Chinese writing system and reveals how the system evolves towards simplification under pressures for representational efficiency. We demonstrate how a library learning approach, utilizing learned abstractions and compression, may help reveal the fundamental computational principles that underlie the creation of combinatorial structures in human cognition, and offer broader insights into the evolution of efficient communication systems.

Jennifer Hu, Ethan Gotlieb Wilcox, Siyuan Song et al.

What have language models (LMs) learned about grammar? This question remains hotly debated, with major ramifications for linguistic theory. However, since probability and grammaticality are distinct notions in linguistics, it is not obvious what string probabilities can reveal about an LM's underlying grammatical knowledge. We present a theoretical analysis of the relationship between grammar, meaning, and string probability, based on simple assumptions about the generative process of corpus data. Our framework makes three predictions, which we validate empirically using 280K sentence pairs in English and Chinese: (1) correlation between the probability of strings within minimal pairs, i.e., string pairs with minimal semantic differences; (2) correlation between models' and humans' deltas within minimal pairs; and (3) poor separation in probability space between unpaired grammatical and ungrammatical strings. Our analyses give theoretical grounding for using probability to learn about LMs' structural knowledge, and suggest directions for future work in LM grammatical evaluation.

Linlu Qiu, Cedegao E. Zhang, Joshua B. Tenenbaum et al. · mit

Language use is shaped by pragmatics -- i.e., reasoning about communicative goals and norms in context. As language models (LMs) are increasingly used as conversational agents, it becomes ever more important to understand their pragmatic reasoning abilities. We propose an evaluation framework derived from Wavelength, a popular communication game where a speaker and a listener communicate about a broad range of concepts in a granular manner. We study a range of LMs on both language comprehension and language production using direct and Chain-of-Thought (CoT) prompting, and further explore a Rational Speech Act (RSA) approach to incorporating Bayesian pragmatic reasoning into LM inference. We find that state-of-the-art LMs, but not smaller ones, achieve strong performance on language comprehension, obtaining similar-to-human accuracy and exhibiting high correlations with human judgments even without CoT prompting or RSA. On language production, CoT can outperform direct prompting, and using RSA provides significant improvements over both approaches. Our study helps identify the strengths and limitations in LMs' pragmatic reasoning abilities and demonstrates the potential for improving them with RSA, opening up future avenues for understanding conceptual representation, language understanding, and social reasoning in LMs and humans.

James A. Michaelov, Roger P. Levy, Benjamin K. Bergen · mit

We show that across architecture (Transformer vs. Mamba vs. RWKV), training dataset (OpenWebText vs. The Pile), and scale (14 million parameters to 12 billion parameters), autoregressive language models exhibit highly consistent patterns of change in their behavior over the course of pretraining. Based on our analysis of over 1,400 language model checkpoints on over 110,000 tokens of English, we find that up to 98% of the variance in language model behavior at the word level can be explained by three simple heuristics: the unigram probability (frequency) of a given word, the $n$-gram probability of the word, and the semantic similarity between the word and its context. Furthermore, we see consistent behavioral phases in all language models, with their predicted probabilities for words overfitting to those words' $n$-gram probabilities for increasing $n$ over the course of training. Taken together, these results suggest that learning in neural language models may follow a similar trajectory irrespective of model details.

Jiayuan Mao, Haoyue Shi, Jiajun Wu et al.

We present Grammar-Based Grounded Lexicon Learning (G2L2), a lexicalist approach toward learning a compositional and grounded meaning representation of language from grounded data, such as paired images and texts. At the core of G2L2 is a collection of lexicon entries, which map each word to a tuple of a syntactic type and a neuro-symbolic semantic program. For example, the word shiny has a syntactic type of adjective; its neuro-symbolic semantic program has the symbolic form λx. filter(x, SHINY), where the concept SHINY is associated with a neural network embedding, which will be used to classify shiny objects. Given an input sentence, G2L2 first looks up the lexicon entries associated with each token. It then derives the meaning of the sentence as an executable neuro-symbolic program by composing lexical meanings based on syntax. The recovered meaning programs can be executed on grounded inputs. To facilitate learning in an exponentially-growing compositional space, we introduce a joint parsing and expected execution algorithm, which does local marginalization over derivations to reduce the training time. We evaluate G2L2 on two domains: visual reasoning and language-driven navigation. Results show that G2L2 can generalize from small amounts of data to novel compositions of words.

Ethan Gotlieb Wilcox, Pranali Vani, Roger P. Levy

We present a targeted, scaled-up comparison of incremental processing in humans and neural language models by collecting by-word reaction time data for sixteen different syntactic test suites across a range of structural phenomena. Human reaction time data comes from a novel online experimental paradigm called the Interpolated Maze task. We compare human reaction times to by-word probabilities for four contemporary language models, with different architectures and trained on a range of data set sizes. We find that across many phenomena, both humans and language models show increased processing difficulty in ungrammatical sentence regions with human and model `accuracy' scores (a la Marvin and Linzen(2018)) about equal. However, although language model outputs match humans in direction, we show that models systematically under-predict the difference in magnitude of incremental processing difficulty between grammatical and ungrammatical sentences. Specifically, when models encounter syntactic violations they fail to accurately predict the longer reaction times observed in the human data. These results call into question whether contemporary language models are approaching human-like performance for sensitivity to syntactic violations.

Stephan C. Meylan, Ruthe Foushee, Elika Bergelson et al.

How do adults understand children's speech? Children's productions over the course of language development often bear little resemblance to typical adult pronunciations, yet caregivers nonetheless reliably recover meaning from them. Here, we employ a suite of Bayesian models of spoken word recognition to understand how adults overcome the noisiness of child language, showing that communicative success between children and adults relies heavily on adult inferential processes. By evaluating competing models on phonetically-annotated corpora, we show that adults' recovered meanings are best predicted by prior expectations fitted specifically to the child language environment, rather than to typical adult-adult language. After quantifying the contribution of this "child-directed listening" over developmental time, we discuss the consequences for theories of language acquisition, as well as the implications for commonly-used methods for assessing children's linguistic proficiency.

Noga Zaslavsky, Jennifer Hu, Roger P. Levy

What computational principles underlie human pragmatic reasoning? A prominent approach to pragmatics is the Rational Speech Act (RSA) framework, which formulates pragmatic reasoning as probabilistic speakers and listeners recursively reasoning about each other. While RSA enjoys broad empirical support, it is not yet clear whether the dynamics of such recursive reasoning may be governed by a general optimization principle. Here, we present a novel analysis of the RSA framework that addresses this question. First, we show that RSA recursion implements an alternating maximization for optimizing a tradeoff between expected utility and communicative effort. On that basis, we study the dynamics of RSA recursion and disconfirm the conjecture that expected utility is guaranteed to improve with recursion depth. Second, we show that RSA can be grounded in Rate-Distortion theory, while maintaining a similar ability to account for human behavior and avoiding a bias of RSA toward random utterance production. This work furthers the mathematical understanding of RSA models, and suggests that general information-theoretic principles may give rise to human pragmatic reasoning.

Jennifer Hu, Jon Gauthier, Peng Qian et al.

While state-of-the-art neural network models continue to achieve lower perplexity scores on language modeling benchmarks, it remains unknown whether optimizing for broad-coverage predictive performance leads to human-like syntactic knowledge. Furthermore, existing work has not provided a clear picture about the model properties required to produce proper syntactic generalizations. We present a systematic evaluation of the syntactic knowledge of neural language models, testing 20 combinations of model types and data sizes on a set of 34 English-language syntactic test suites. We find substantial differences in syntactic generalization performance by model architecture, with sequential models underperforming other architectures. Factorially manipulating model architecture and training dataset size (1M--40M words), we find that variability in syntactic generalization performance is substantially greater by architecture than by dataset size for the corpora tested in our experiments. Our results also reveal a dissociation between perplexity and syntactic generalization performance.

Richard Futrell, Roger P. Levy

RNN language models have achieved state-of-the-art results on various tasks, but what exactly they are representing about syntax is as yet unclear. Here we investigate whether RNN language models learn humanlike word order preferences in syntactic alternations. We collect language model surprisal scores for controlled sentence stimuli exhibiting major syntactic alternations in English: heavy NP shift, particle shift, the dative alternation, and the genitive alternation. We show that RNN language models reproduce human preferences in these alternations based on NP length, animacy, and definiteness. We collect human acceptability ratings for our stimuli, in the first acceptability judgment experiment directly manipulating the predictors of syntactic alternations. We show that the RNNs' performance is similar to the human acceptability ratings and is not matched by an n-gram baseline model. Our results show that RNNs learn the abstract features of weight, animacy, and definiteness which underlie soft constraints on syntactic alternations.