Search

We evaluate whether BERT, a widely used neural network forsentence processing, acquires an inductive bias towards form-ing structural generalizations through pretraining on raw data.We conduct four experiments testing its preference for struc-tural vs. linear generalizations in different structure-dependentphenomena. We find that BERT makes a structural general-ization in 3 out of 4 empirical domains—subject-auxiliary in-version, reflexive binding, and verb tense detection in embed-ded clauses—but makes a linear generalization when tested onNPI licensing. We argue that these results are the strongest ev-idence so far from artificial learners supporting the propositionthat a structural bias can be acquired from raw data. If this con-clusion is correct, it is tentative evidence that some linguisticuniversals can be acquired by learners without innate biases.However, the precise implications for human language acqui-sition are unclear, as humans learn language from significantlyless data than BERT.

This study investigates the second language (L2) acquisition of a constraint on definiteness in Japanese floating numeral quantifiers (NQs) by native English and Korean speakers. The constraint arises because of the specific structural relation between a floating NQ and its associated noun, resulting in an obligatorily indefinite interpretation. The indirect – or, covert – encoding of definiteness in this structure allows investigation of predictions based on the cline of difficulty proposed by Cho and Slabakova (2014), whereby L2 acquisition of a covert property may be facilitated if the first language (L1) expresses the relevant feature overtly. English is such a language, having overt morphology to express definiteness, whereas Korean has floating NQs that are obligatorily, and covertly, indefinite, as in Japanese. Sensitivity to definiteness in Japanese floating NQs was measured using an acceptability judgement task. Although both L1-Korean and L1-English speakers of Japanese showed sensitivity to the constraint at group level, follow-up analyses suggested that the Korean group had more consistent knowledge. We argue that the complexity of the acquisition task – which was greater for the English-speakers than the Korean-speakers – played a bigger role in attainment than overt versus covert encoding of the relevant feature in the L1. [ABSTRACT FROM AUTHOR]

Syntactic rules in natural language typically need to make ref-erence to hierarchical sentence structure. However, the simpleexamples that language learners receive are often equally com-patible with linear rules. Children consistently ignore theselinear explanations and settle instead on the correct hierarchi-cal one. This fact has motivated the proposal that the learner’shypothesis space is constrained to include only hierarchicalrules. We examine this proposal using recurrent neural net-works (RNNs), which are not constrained in such a way. Wesimulate the acquisition of question formation, a hierarchicaltransformation, in a fragment of English. We find that someRNN architectures tend to learn the hierarchical rule, suggest-ing that hierarchical cues within the language, combined withthe implicit architectural biases inherent in certain RNNs, maybe sufficient to induce hierarchical generalizations. The like-lihood of acquiring the hierarchical generalization increasedwhen the language included an additional cue to hierarchy inthe form of subject-verb agreement, underscoring the role ofcues to hierarchy in the learner’s input.

This paper sheds light on the acquisition of wh-islands in L2 English spoken by native speakers of Spanish and L2 Spanish spoken by native speakers of English as well as on the distribution of wh-islands in L1 Spanish. A grammaticality judgment task with a 7-point Likert scale provides evidence that wh-island effects are present in L1 and L2 Spanish as well as L1 and L2 English. The L1 Spanish facts challenge the received view of wh-islands in this language, in keeping with recent developments which show that islands are more widely attested across languages than previously thought. These facts also highlight the dialogue between L2 research and replication studies thanks to the use of native control groups.

This paper seeks to make progress in our understanding of the non-UG components of Chomsky’s (2005) Three Factors model. In relation to the input (Factor 2), I argue for the need to formulate a suitably precise hypothesis about which aspects of the input will qualify as ‘intake’ and, hence, serve as the basis for grammar construction. In relation to Factor 3, I highlight a specific cognitive bias that appears well motivated outside of language, while also having wide-ranging consequences for our understanding of how I-language grammars are constructed, and why they should have the crosslinguistically comparable form that generativists have always argued human languages have. This is Maximise Minimal Means (MMM). I demonstrate how its incorporation into our model of grammar acquisition facilitates understanding of diverse facts about natural language typology, acquisition, both in “stable” and “unstable” contexts, and also the ways in which linguistic systems may change over time.

This paper sheds light on the acquisition of wh-islands in L2 English spoken by native speakers of Spanish and L2 Spanish spoken by native speakers of English as well as on the distribution of wh-islands in L1 Spanish. A grammaticality judgment task with a 7-point Likert scale provides evidence that wh-island effects are present in L1 and L2 Spanish as well as L1 and L2 English. The L1 Spanish facts challenge the received view of wh-islands in this language, in keeping with recent developments which show that islands are more widely attested across languages than previously thought. These facts also highlight the dialogue between L2 research and replication studies thanks to the use of native control groups.

Noam Chomsky's early work was at the core of the "cognitive revolution" in the 1950s-60s, leading to a paradigm shift from a behavioralist to a mentalistic approach to human psychology. Central to this revolution has been the question of how infants learn language. Here, we provide an overview of how the generative enterprise has shaped research on language acquisition over the last decades. We argue that a large body of empirical knowledge about infants' representation of grammar has accumulated. Many of these questions would most likely not have been investigated empirically without the impetus of such a mentalistic approach. [ABSTRACT FROM AUTHOR]

Pater's (2019) target article proposes that neural networks will provide theories of learning that generative grammar lacks. We argue that his enthusiasm is premature since the biases of neural networks are largely unknown, and he disregards decades of work on machine learning and learnability. Learning biases form a two-way street: all learners have biases, and those biases constrain the space of learnable grammars in mathematically measurable ways. Analytical methods from the related fields of computational learning theory and grammatical inference allow one to study language learning, neural networks, and linguistics at an appropriate level of abstraction. The only way to satisfy our hunger and to make progress on the science of language learning is to confront these core issues directly. [ABSTRACT FROM AUTHOR]

Joe Pater's (2019) target article calls for greater interaction between neural network research and linguistics. I expand on this call and show how such interaction can benefit both fields. Linguists can contribute to research on neural networks for language technologies by clearly delineating the linguistic capabilities that can be expected of such systems, and by constructing controlled experimental paradigms that can determine whether those desiderata have been met. In the other direction, neural networks can benefit the scientific study of language by providing infrastructure for modeling human sentence processing and for evaluating the necessity of particular innate constraints on language acquisition. [ABSTRACT FROM AUTHOR]

Copyright of Catalan Journal of Linguistics is the property of Universitat Autonoma de Barcelona and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

The aim of this paper is to confront the main arguments employed by advocates of Universal Moral Grammar (UMG) with the results obtained in empirical studies from different fields of cognitive science and also with constructivist arguments proposed by Prinz and Sterelny, among others. From this analysis, I conclude that the Poverty of the Stimulus Argument is unconvincing when applied to the field of morality. The research with the trolley problems, often used to support the existence of UMG, is also insufficient to sustain the thesis that morality works similarly to language, especially when the existence of simpler explanations is considered. Furthermore, in spite of the universality of morality, the unlimited variation in moral norms across groups is a serious problem for any account relying on the principles and parameters model of the linguistic analogy.

This article deals with the classic example of "poverty of the linguistic stimulus", which concerns the formation of yes-no interrogative sentences in English. Young children have been claimed to acquire a subject-auxiliary inversion rule (SAI) without being exposed to the data necessary to learn it. According to the traditional conclusion, that acquisition is made possible by an innate structure-dependence (SD) principle. We first show that young English-speaking children do have access to some of the supposedly "relevant evidence" in this example and that other kinds of data might be just as important. Next, we examine Chomsky's recent "minimalist" analysis of yes-no interrogatives, which abandons the SD principle but introduces new equally problematic hypotheses. Finally, we argue that the early acquisition of a general question-formation operation such as SAI is not compatible with the results of studies on young children's questions and other linguistic productions. Those results are instead shown to be consistent with a constructional and usage-based approach to language.

This paper presents an algorithm for learning the construction grammar of a language from a large corpus. This grammar induction algorithm has two goals: first, to show that construction grammars are learnable without highly specified innate structure; second, to develop a model of which units do or do not constitute constructions in a given dataset. The basic task of construction grammar induction is to identify the minimum set of constructions that represents the language in question with maximum descriptive adequacy. These constructions must (1) generalize across an unspecified number of units while (2) containing mixed levels of representation internally (e.g., both item-specific and schematized representations), and (3) allowing for unfilled and partially filled slots. Additionally, these constructions may (4) contain recursive structure within a given slot that needs to be reduced in order to produce a sufficiently schematic representation. In other words, these constructions are multi-length, multi-level, possibly discontinuous co-occurrences which generalize across internal recursive structures. These co-occurrences are modeled using frequency and the ΔP measure of association, expanded in novel ways to cover multi-unit sequences. This work provides important new evidence for the learnability of construction grammars as well as a tool for the automated corpus analysis of constructions. [ABSTRACT FROM AUTHOR]

A generative approach to language acquisition is no different from any other in assuming that target language input is crucial for language acquisition. This discussion note addresses the place of input in generative second language acquisition (SLA) research and the perception in the wider field of SLA research that generative SLA ‘ignores’ the input. This impression may have arisen because generative SLA researchers have tended not to systematically study quantitative distribution of input properties, nor qualitative properties of the input available to learners. We argue that precisely these sorts of studies would be at least beneficial, if not indispensable, to the development of a comprehensive (generative) theory of SLA. Furthermore, a welcome side-effect of more systematic engagement with input would be potential for greater accessibility for generative SLA in the wider field of second language (L2) studies. [ABSTRACT FROM AUTHOR]

Universal Grammar (UG) is a suspect concept. There is little agreement on what exactly is in it; and the empirical evidence for it is very weak. This paper critically examines a variety of arguments that have been put forward as evidence for UG, focussing on the three most powerful ones: universality (all human languages share a number of properties), convergence (all language learners converge on the same grammar in spite of the fact that they are exposed to different input), and poverty of the stimulus (children know things about language which they could not have learned from the input available to them). I argue that these arguments are based on premises which are either false or unsubstantiated. Languages differ from each other in profound ways, and there are very few true universals, so the fundamental crosslinguistic fact that needs explaining is diversity, not universality. A number of recent studies have demonstrated the existence of considerable differences in adult native speakers' knowledge of the grammar of their language, including aspects of inflectional morphology, passives, quantifiers, and a variety of more complex constructions, so learners do not in fact converge on the same grammar. Finally, the poverty of the stimulus argument presupposes that children acquire linguistic representations of the kind postulated by generative grammarians; constructionist grammars such as those proposed by Tomasello, Goldberg and others can be learned from the input. We are the only species that has language, so there must be something unique about humans that makes language learning possible. The extent of crosslinguistic diversity and the considerable individual differences in the rate, style and outcome of acquisition suggest that it is more promising to think in terms of a languagemaking capacity, i.e., a set of domain-general abilities, rather than an innate body of knowledge about the structural properties of the target system. [ABSTRACT FROM AUTHOR]

This article argues that mirror neurons originate in sensorimotor associative learning and therefore a new approach is needed to investigate their functions. Mirror neurons were discovered about 20 years ago in the monkey brain, and there is now evidence that they are also present in the human brain. The intriguing feature of many mirror neurons is that they fire not only when the animal is performing an action, such as grasping an object using a power grip, but also when the animal passively observes a similar action performed by another agent. It is widely believed that mirror neurons are a genetic adaptation for action understanding; that they were designed by evolution to fulfill a specific socio-cognitive function. In contrast, we argue that mirror neurons are forged by domain-general processes of associative learning in the course of individual development, and, although they may have psychological functions, they do not necessarily have a specific evolutionary purpose or adaptive function. The evidence supporting this view shows that (1) mirror neurons do not consistently encode action “goals”; (2) the contingency- and context-sensitive nature of associative learning explains the full range of mirror neuron properties; (3) human infants receive enough sensorimotor experience to support associative learning of mirror neurons (“wealth of the stimulus”); and (4) mirror neurons can be changed in radical ways by sensorimotor training. The associative account implies that reliable information about the function of mirror neurons can be obtained only by research based on developmental history, system-level theory, and careful experimentation. [ABSTRACT FROM AUTHOR]

This paper examines the logical problem of language acquisition drawing upon an experimental study on children’s knowledge of anaphoric one by Lidz, Waxman and Freedman (2003). The finding was that, upon being presented with the instruction “Look! A yellow bottle. Do you see another one?”, 18-month-old children prefer to look at a yellow bottle rather than to a bottle of a different color. According to Lidz et al. (2003), the results that children cannot interpret one as anaphoric to head nouns. We point out that the experimental findings are not explained under the hypothesis offered by the authors of that study. Secondly, we consider whether, under current assumptions, children’s knowledge of anaphoric one can be inferred from the properties of the final state. Thirdly, we reaffirm the validity of the Poverty of the Stimulus argument, despite the challenge posed by the learning model proposed by Regier and Gahl (2004). Finally, we draw upon recent psycholinguistic work to propose an explanation for the findings documented by Lidz et al. (2003) that is independent from – though consistent with – their knowledge of the constraint on anaphoric one.

Abstract: There is much debate over the degree to which language learning is governed by innate language-specific biases, or acquired through cognition-general principles. Here we examine the probabilistic language acquisition hypothesis on three levels: We outline a novel theoretical result showing that it is possible to learn the exact generative model underlying a wide class of languages, purely from observing samples of the language. We then describe a recently proposed practical framework, which quantifies natural language learnability, allowing specific learnability predictions to be made for the first time. In previous work, this framework was used to make learnability predictions for a wide variety of linguistic constructions, for which learnability has been much debated. Here, we present a new experiment which tests these learnability predictions. We find that our experimental results support the possibility that these linguistic constructions are acquired probabilistically from cognition-general principles. [Copyright &y& Elsevier]

This review article evaluates the intersection of the content of two recent anthologies in second language (L2) phonology. One of the books lays out both the methodological context and theoretical underpinnings of the field, whereas the other volume reports 11 empirical studies on the L2 acquisition of several aspects of pronunciation by adult learners of English whose native language is either Spanish or Brazilian Portuguese. The criteria applied for the evaluation lead to the conclusion that, for the most part, the studies are successful in presenting an interesting array of second-language pronunciation patterns, and that such an intersection of domains is a fruitful way to advance the field of interphonology. [ABSTRACT FROM PUBLISHER]

While rules and exemplars are usually viewed as opposites, this paper argues that they form end points of the same distribution. By representing both rules and exemplars as (partial) trees, we can take into account the fluid middle ground between the two extremes. This insight is the starting point for a new theory of language learning that is based on the following idea: If a language learner does not know which phrase-structure trees should be assigned to initial sentences, s/he allows (implicitly) for all possible trees and lets linguistic experience decide which is the “best” tree for each sentence. The best tree is obtained by maximizing “structural analogy” between a sentence and previous sentences, which is formalized by the most probable shortest combination of subtrees from all trees of previous sentences. Corpus-based experiments with this model on the Penn Treebank and the Childes database indicate that it can learn both exemplar-based and rule-based aspects of language, ranging from phrasal verbs to auxiliary fronting. By having learned the syntactic structures of sentences, we have also learned the grammar implicit in these structures, which can in turn be used to produce new sentences. We show that our model mimicks children’s language development from item-based constructions to abstract constructions, and that the model can simulate some of the errors made by children in producing complex questions. [ABSTRACT FROM AUTHOR]

It is widely held that children’s linguistic input underdetermines the correct grammar, and that language learning must therefore be guided by innate linguistic constraints. Here, we show that a Bayesian model can learn a standard poverty-of-stimulus example, anaphoric one, from realistic input by relying on indirect evidence, without a linguistic constraint assumed to be necessary. Our demonstration does, however, assume other linguistic knowledge; thus, we reduce the problem of learning anaphoric one to that of learning this other knowledge. We discuss whether this other knowledge may itself be acquired without linguistic constraints. [ABSTRACT FROM AUTHOR]

Abstract: Several phonological and prosodic properties of words have been shown to relate to differences between grammatical categories. Distributional information about grammatical categories is also a rich source in the child’s language environment. In this paper we hypothesise that such cues operate in tandem for developing the child’s knowledge about grammatical categories. We term this the Phonological-Distributional Coherence Hypothesis (PDCH). We tested the PDCH by analysing phonological and distributional information in distinguishing open from closed class words and nouns from verbs in four languages: English, Dutch, French, and Japanese. We found an interaction between phonological and distributional cues for all four languages indicating that when distributional cues were less reliable, phonological cues were stronger. This provides converging evidence that language is structured such that language learning benefits from the integration of information about category from contextual and sound-based sources, and that the child’s language environment is less impoverished than we might suspect. [Copyright &y& Elsevier]

In a head-final language, V-raising is hard to detect since there is no evidence from the string to support a raising analysis. If the language has a cliticlike negation that associates with the verb in syntax, then scope facts concerning negation and a quantified object NP could provide evidence regarding the height of the verb. Even so, such facts are rare, especially in the input to children, and so we might expect that not all speakers exposed to a head-final language acquire the same grammar as far as V-raising is concerned. Here, we present evidence supporting this expectation. Using experimental data concerning the scope of quantified NPs and negation in Korean, elicited from both adults and 4-year-old children, we show that there are two populations of Korean speakers: one with V-raising and one without. [ABSTRACT FROM AUTHOR]

Lidz et al. [Lidz, J., Waxman, S., & Freedman, J. (2003). What infants know about syntax but couldn''t have learned: Experimental evidence for syntactic structure at 18 months. Cognition, 89, B65–B73.] claim experimental substantiation of an argument from the poverty of the stimulus, in the sense of Pullum and Scholz [Linguist. Rev. 19 (2002) 9]. They cite a specific feature of English—the assignment of appropriate antecedents for anaphoric one—that cannot possibly be learned from experience because the evidence needed is found only in utterances of a type too rare to be encountered. Their argument involves three empirical claims. In this note we dispute all three. [Copyright &y& Elsevier]

Syntactic knowledge is widely held to be partially innate, rather than learned. In a classic example, it is sometimes argued that children know the proper use of anaphoric one, although that knowledge could not have been learned from experience. Lidz et al. [Lidz, J., Waxman, S., & Freedman, J. (2003). What infants know about syntax but couldn’t have learned: Experimental evidence for syntactic structure at 18 months. Cognition, 89, B65–B73.] pursue this argument, and present corpus and experimental evidence that appears to support it; they conclude that specific aspects of this knowledge must be innate. We demonstrate, contra Lidz et al., that this knowledge may in fact be acquired from the input, through a simple Bayesian learning procedure. The learning procedure succeeds because it is sensitive to the absence of particular input patterns—an aspect of learning that is apparently overlooked by Lidz et al. More generally, we suggest that a prominent form of the “argument from poverty of the stimulus” suffers from the same oversight, and is as a result logically unsound. [Copyright &y& Elsevier]

Lidz, Waxman, and Freedman [Lidz, J., Waxman, S., & Freedman, J. (2003). What infants know about syntax but couldn’t have learned: Evidence for syntactic structure at 18-months. Cognition, 89, B65–B73.] argue that acquisition of the syntactic and semantic properties of anaphoric one in English relies on innate knowledge within the learner. Several commentaries have now been published questioning this finding. We defend the original finding by identifying both empirical and logical flaws in the critiques. [Copyright &y& Elsevier]

Generative linguistic theory stands on the hypothesis that grammar cannot be acquired solely on the basis of an analysis of the input, but depends, in addition, on innate structure within the learner to guide the process of acquisition. This hypothesis derives from a logical argument, however, and its consequences have never been examined experimentally with infant learners. Challenges to this hypothesis, claiming that an analysis of the input is indeed sufficient to explain grammatical acquisition, have recently gained attention. We demonstrate with novel experimentation the insufficiency of this countervailing view. Focusing on the syntactic structures required to determine the antecedent for the pronoun one, we demonstrate that the input to children does not contain sufficient information to support unaided learning. Nonetheless, we show that 18-month-old infants do have command of the syntax of one. Because this syntactic knowledge could not have been gleaned exclusively from the input, infants'' mastery of this aspect of syntax constitutes evidence for the contribution of innate structure within the learner in acquiring a grammar. [Copyright &y& Elsevier]