An official website of the United States government
The experimental study of artificial language learning has become a widely used means of investigating the predictions of theories of language learning and representation. Although much is now known about the generalizations that learners make from various kinds of data, relatively little is known about how those representations affect speech processing. This paper presents an event-related potential (ERP) study of brain responses to violations of lab-learned phonotactics. Novel words that violated a learned phonotactic constraint elicited a larger Late Positive Component (LPC) than novel words that satisfied it. Similar LPCs have been found for violations of natively acquired linguistic structure, as well as for violations of other types of abstract generalizations, such as musical structure. We argue that lab-learned phonotactic generalizations are represented abstractly and affect the evaluation of speech in a manner that is similar to natively acquired syntactic and phonological rules.
more »
« less
CAN NEURAL NETWORKS ACQUIRE A STRUCTURAL BIAS FROM RAW LINGUISTIC DATA?
We evaluate whether BERT, a widely used neural network for sentence processing, acquires an inductive bias towards forming structural generalizations through pretraining on raw data. We conduct four experiments testing its preference for structural vs. linear generalizations in different structure-dependent phenomena. We find that BERT makes a structural generalization in 3 out of 4 empirical domains---subject-auxiliary inversion, reflexive binding, and verb tense detection in embedded clauses---but makes a linear generalization when tested on NPI licensing. We argue that these results are the strongest evidence so far from artificial learners supporting the proposition that a structural bias can be acquired from raw data. If this conclusion is correct, it is tentative evidence that some linguistic universals can be acquired by learners without innate biases. However, the precise implications for human language acquisition are unclear, as humans learn language from significantly less data than BERT.
more »
« less
- Award ID(s):
- 1850208
- PAR ID:
- 10233708
- Date Published:
- Journal Name:
- Proceedings of the Annual Meeting of the Cognitive Science Society
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Relations between words are governed by hierarchical structure rather than linear ordering. Sequence-to-sequence (seq2seq) models, despite their success in downstream NLP applications, often fail to generalize in a hierarchy sensitive manner when performing syntactic transformations—for example, transforming declarative sentences into questions. However, syntactic evaluations of seq2seq models have only observed models that were not pre-trained on natural language data before being trained to perform syntactic transformations, in spite of the fact that pre-training has been found to induce hierarchical linguistic generalizations in language models; in other words, the syntactic capabilities of seq2seq models may have been greatly understated. We address this gap using the pre-trained seq2seq models T5 and BART, as well as their multilingual variants mT5 and mBART. We evaluate whether they generalize hierarchically on two transformations in two languages: question formation and passivization in English and German. We find that pre-trained seq2seq models generalize hierarchically when performing syntactic transformations, whereas models trained from scratch on syntactic transformations do not. This result presents evidence for the learnability of hierarchical syntactic information from non-annotated natural language text while also demonstrating that seq2seq models are capable of syntactic generalization, though only after exposure to much more language data than human learners receive.more » « less
-
null (Ed.)One reason pretraining on self-supervised linguistic tasks is effective is that it teaches models features that are helpful for language understanding. However, we want pretrained models to learn not only to represent linguistic features, but also to use those features preferentially during fine-turning. With this goal in mind, we introduce a new English-language diagnostic set called MSGS (the Mixed Signals Generalization Set), which consists of 20 ambiguous binary classification tasks that we use to test whether a pretrained model prefers linguistic or surface generalizations during finetuning. We pretrain RoBERTa from scratch on quantities of data ranging from 1M to 1B words and compare their performance on MSGS to the publicly available RoBERTa_BASE. We find that models can learn to represent linguistic features with little pretraining data, but require far more data to learn to prefer linguistic generalizations over surface ones. Eventually, with about 30B words of pretraining data, RoBERTa_BASE does consistently demonstrate a linguistic bias with some regularity. We conclude that while self-supervised pretraining is an effective way to learn helpful inductive biases, there is likely room to improve the rate at which models learn which features matter.more » « less
-
Abstract Language models are typically evaluated on their success at predicting the distribution of specific words in specific contexts. Yet linguistic knowledge also encodes relationships between contexts, allowing inferences between word distributions. We investigate the degree to which pre-trained transformer-based large language models (LLMs) represent such relationships, focusing on the domain of argument structure. We find that LLMs perform well in generalizing the distribution of a novel noun argument between related contexts that were seen during pre-training (e.g., the active object and passive subject of the verb spray), succeeding by making use of the semantically organized structure of the embedding space for word embeddings. However, LLMs fail at generalizations between related contexts that have not been observed during pre-training, but which instantiate more abstract, but well-attested structural generalizations (e.g., between the active object and passive subject of an arbitrary verb). Instead, in this case, LLMs show a bias to generalize based on linear order. This finding points to a limitation with current models and points to a reason for which their training is data-intensive.1more » « less
-
This study examines whether children acquiring Tseltal (Mayan) demonstrate a noun bias – an overrepresentation of nouns in their early vocabularies. Nouns, specifically concrete and animate nouns, are argued to universally predominate in children’s early vocabularies because their referents are naturally available as bounded concepts to which linguistic labels can be mapped. This early advantage for noun learning has been documented using multiple methods and across a diverse collection of language populations. However, past evidence bearing on a noun bias in Tseltal learners has been mixed. Tseltal grammatical features and child–caregiver interactional patterns dampen the salience of nouns and heighten the salience of verbs, leading to the prediction of a diminished noun bias and perhaps even an early predominance of verbs. We here analyze the use of noun and verb stems in children’s spontaneous speech from egocentric daylong recordings of 29 Tseltal learners between 0;9 and 4;4. We find weak to no evidence for a noun bias using two separate analytical approaches on the same data; one analysis yields a preliminary suggestion of a flipped outcome (i.e. a verb bias). We discuss the implications of these findings for broader theories of learning bias in early lexical development.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
