Linguistic Nativism and the Poverty of the Stimulus

Alexander Clark, Shalom Lappin

• Publisher: Wiley-Blackwell
• Year: 2010
• ISBN:
  9781444390551
• Edition: 1
• Topic: Languages & Linguistics
• Subtopic: Linguistics
• Index: Languages & Linguistics

This unique contribution to the ongoing discussion of language acquisition considers the Argument from the Poverty of the Stimulus in language learning in the context of the wider debate over cognitive, computational, and linguistic issues.

• Critically examinesthe Argument from the Poverty of the Stimulus - the theory that the linguistic input which children receive is insufficient to explain the rich and rapid development of their knowledge of their first language(s) through general learning mechanisms
• Focuses on formal learnability properties of the class of natural languages, considered from the perspective of several learning theoretic models
• The only current book length study of arguments for the poverty of the stimulus which focuses on the computational learning theoretic aspects of the problem

1

Introduction: Nativism in Linguistic Theory

Clearly human beings have an innate, genetically specified cognitive endowment that allows them to acquire natural language. The precise nature of this endowment is, however, a matter of scientific controversy. A variety of views on this issue have been proposed. We take two positions as representative of the spectrum. The first takes language acquisition and use as mediated primarily by genetically determined language-specific representations and mechanisms. The second regards these processes as largely or entirely the result of domain-general learning procedures.

The debate between these opposing perspectives does not concern the existence of innately specified cognitive capacities. While humans learn languages with a combinatorial syntax, productive morphology, and (in all cases but sign language) phonology, other species do not. Hence, people have a unique, species-specific ability to learn language and process it. What remains in dispute is the nature of this innate ability, and, above all, the extent to which it is a domain-specific linguistic device. This is an empirical question, but there is a dearth of direct evidence about the actual brain and neural processes that support language acquisition. Moreover, invasive experimental work is often impossible for ethical or practical reasons. The problem has frequently been addressed abstractly, through the study of the mathematical and computational processes required to produce the outcome of learning from the data available to the learner. As a result, choosing among competing hypotheses on the basis of tangible experimental or observational evidence is generally not an option.

The concept of innateness is, itself, acutely problematic. It lacks an agreed biological or psychological characterization, and we will avoid it wherever possible. It is instructive to distinguish between innateness as a biological concept from the idea of innateness that has figured in the history of philosophy, and we will address this difference in section 1.2. More generally, innateness as a genetic property is notoriously difficult to define, and its use is generally discouraged by biologists. Mameli and Bateson (2006) point out that it conflates a variety of different, often not fully compatible, ideas. These include canalization, genetic determinism, presence from birth, and others.

It is uncontroversial, if obvious, that the environment of the child has an important influence on the linguistic abilities that he/she acquires. Children who are raised in English-speaking homes grow up to speak English, while those in Japanese-speaking families learn Japanese. When a typically developing infant is adopted very early, there is no apparent delay or distortion in the language acquisition process. By contrast, if a child is deprived of language and social interaction in the early years of life, then language does not develop normally, and, in extreme cases, fails to appear at all. It is safe to assume, then, that adult linguistic competence emerges through the interaction between the innate learning ability of the child, and his/her exposure to linguistic data in a social context, primarily through interaction with caregivers, as well as access to ambient adult speech in the environment.

The interesting and important issue in this discussion is whether language learning depends heavily on an ability that is special purpose in character, or whether it is the result of general learning methods that the child applies to other cognitive tasks. It seems clear that general-purpose learning algorithms play some role in certain aspects of the language acquisition task. However, it is far from obvious how domain-specific and general-learning procedures divide this task between them. Linguists have frequently assumed that lexical acquisition, for example, is largely the result of data-driven learning, while other aspects of linguistic knowledge, such as syntax, depend heavily on rich domain-specific mechanisms.

Another long-running debate concerns whether the capacity of adults to speak languages can be properly described as knowledge (Devitt, 2006). This is a philosophical question that falls outside the scope of this study. We do not yet know anything substantive about how learning mechanisms or the products of these mechanisms are represented in the brain. We cannot tell whether they are encoded as propositions in some symbolic system, or are emergent properties of a neural network. We do not yet have the evidence necessary to resolve these sorts of questions, or even to formulate them precisely. The technical term cognizing has occasionally been used in place of knowing, since knowledge of language has different properties from other paradigm cases of knowledge. Unlike the latter, it is not conscious, and the question of epistemic justification does not arise. We will pass over this issue here. It is not relevant to our concerns, and none of the arguments that we develop in this book depend upon it.

The idea of domain specificity is less problematic, and it provides the focus of our interest. At one extreme we have details that are clearly specific to language, such as parts of speech. At the other we have general properties of semantic representation, which seem to be domain general in character. We can distinguish clearly between semantic concepts such as agent and purely syntactic concepts such as subject,noun, and noun phrase, even though systematic relations may connect them. Hierarchical structure offers a less clear-cut case. It is generally considered to be a central element of linguistic description at various levels of representation, but it is arguably present as an organizing principle across a variety of nonlinguistic modes of cognition. There are clearly gray areas where a learning algorithm originally evolved for one purpose might be co-opted for another. Most specific proposals for a domain-specific theory of language acquisition do not allow for this sort of ambiguity. Instead, they posit a set of principles and formal objects that are decidedly language specific in nature.

A related question is whether a phenomenon is species specific. Given that language is restricted to humans, if a property is language specific, then it must be unique to people. Learning mechanisms present in a nonhuman species cannot be language specific.

Humans do exhibit domain-general learning capabilities. They learn skills like chess, which cannot plausibly be attributed to a domain-specific acquisition device. One way to understand the difference between domain-general and domain-specific learning is to consider an idealized form of learning. One of the most general such formulations is Bayesian learning. It abstracts away from computational considerations and considers the optimal use of information to update the knowledge of a situation. On this approach we can achieve a precise characterization of the contribution that domain knowledge makes, in the form of a prior probability distribution. In domain-specific learning, the prior distribution tightly restricts the learner to a small set of hypotheses. The prior knowledge is thus very important to the final learning outcome. By contrast, in domain-general learning, the prior distribution is very general in character. It allows a wide range of possibilities, and the hypothesis on which the learner eventually settles is conditioned largely by the information supplied by the input data. This latter form of learning is sometimes called empiricist or data-driven learning. Here the learned hypothesis, in this case the grammar of the language, is largely extracted from the dataset through processes of induction.

Language acquisition presents some unusual characteristics, which we will discuss further in the next chapter. First, languages are very complex and hard for adults to learn. Learning a second language as an adult requires a significant commitment of time, and the end result generally falls well short of native proficiency. Second, children learn their first languages without explicit instruction, and with no apparent effort. Third, the information available to the child is fairly limited. He/she hears a random subset of short sentences. The putative difficulty of this learning task is one of the strongest intuitive arguments for linguistic nativism. It has become known as The Argument from the Poverty of the Stimulus (APS).

The term universal grammar (UG) is problematic in that it is not used in a consistent manner in the linguistics literature. On the standard description of UG, it is the initial state of the language learner. However, it is also used in a number of alternative ways. It can refer to the universal properties of natural languages, the set of principles, formal objects, and operations shared by all natural languages. Alternatively, it is interpreted as the class of possible human languages. To avoid equivocation, we will take UG in the sense of the term that seems to us to be standard in current linguistic theory. We intend UG to be the species-specific cognitive mechanism that allows a child to acquire its first language(s). Equivalently, we take it to be the initial state of the language learner, independent of the data to which he/she is exposed in his/her environment. We will pass over the systematic ambiguity between UG taken as the actual initial state of the learner, and UG construed as the theory of this state, as this distinction is not likely to cause confusion here. Given this interpretation of UG, its existence is uncontroversial. The interesting empirical questions turn on its richness, and the extent to which it is domain specific. These are the issues that drive this study.

1.1 Historical Development

Chomsky has been the most prominent advocate of linguistic nativism over the past 50 years, though he has largely resisted the use of this term. His view of universal grammar as the set of innate constraints that a language faculty imposes on the form of possible grammars for natural language has dominated theoretical linguistics during most of this period. To get a clearer idea of what is involved in this notion of the language faculty we will briefly consider the historical development of the connection between UG and language acquisition in Chomsky’s work.

Chomsky (1965) argues that, given the relative paucity of primary data and the (putative) fact that statistical methods of induction cannot yield knowledge of syntax, the essential form of any possible grammar of a natural language must be part of the cognitive endowment that humans bring to the language acquisition task. He characterizes UG as containing the following components (p. 31):

1 (a) an enumeration of the class s₁, s₂, … of possible sentences;

(b) an enumeration of the class SD₁, SD₂, … of possible structural descriptions;

(c) an enumeration of the class G₁, G₂, … of possible generative grammars;

(d) specification of a function f such that SD_f(i,j) is the structural description assigned to sentence s_i by grammar G_j, for arbitrary i,j;

(e) specification of a function m such that m(i) is an integer associated with the grammar G_i as its value (with, let us say, lower value indicated by higher number).

1(c) is the hypothesis space of possible grammars for natural languages. 1(a) is the set of strings that each grammar generates. 1(b) is the set of syntactic representations that these grammars assign to the strings that they produce, where this assignment can be a one-to-many relation in which a string receives alternative descriptions. 1(d) is the function that maps a grammar to the set of representations for a string. 1(e) is an evaluation measure that ranks the possible grammars. Specifically, it determines the most highly valued grammar from among those that generate the same string set.

Chomsky (1965) posits this UG as an innate cognitive module that supports language acquisition. It parses the input stream of primary linguistic data (PLD) into phonetic sequences that comprise distinct sentences, and it defines the hypothesis space of possible grammars with which a child can assign syntactic representations to these strings. In cases where several grammars are compatible with the data, the evaluation measure selects the preferred one.

Chomsky distinguishes between a theory of grammar that is descriptively adequate from one that achieves explanatory adequacy. The former generates and assigns syntactic representations to the sentences of a language in a way that captures their observed structural properties. The latter incorporates an evaluation measure that encodes the function that children apply to select a single grammar from among several incompatible grammars, all of which are descriptively adequate for the data to which the child has been exposed. This notion of explanatory adequacy is formulated in terms of a theory of UG’s capacity to account for central aspects of language acquisition.

The evaluation measure in the Aspects model of UG is an awkward and problematic device. It is required in order to resolve conflicts among alternative grammars that are compatible with the PLD. However, it is not clear how it can be specified, and what sort of evidence should be invoked to motivate an account of its design. By assumption, it ranks grammars that enjoy the same degree of descriptive adequacy, and so the PLD cannot help with the selection.

Notions of formal simplicity of the sort used to choose among rival scientific theories do not offer an appropriate grammar-ranking procedure for at least two reasons. First, they are notoriously difficult to formulate as global metrics that are both precise and consistent. Second, if one could define a workable simplicity measure of this kind, then it would not be part of a domain-specific UG but an instance of a general principle for deciding among competing theories across cognitive domains. Chomsky (1965, p. 38) suggests that the evaluation measure is a domain-specific simplicity measure internal to UG.

If a particular formulation of (i)–(iv) [1(a)–1(d)] is assumed, and if pairs ( D₁,G₁), ( D₂,G₂) …of primary linguistic data and descriptively adequate grammars are given, the problem of defining “simplicity” is just the problem of discovering how G_i is determined by D_i for each i. Suppose, in other words, that we regard an acquisition model for a language as an input-output device that determines a particular generative grammar as “output,” given certain primary linguistic data as input. A proposed simplicity measure, taken together with a specification (i)–(iv), constitutes a hypothesis concerning the nature of such a device. Choice of a simplicity measure is therefore an empirical matter with empirical consequences.

The problem here is that Chomsky does not indicate the sort of evidence that can be used to evaluate such a simplicity metric. If observable linguistic data and general notions of theoretical simplicity are excluded, then we have only the facts of language acquisition to go on. But it is not obvious how these can be used to define a UG internal evaluation function. If, at the final stage of the acquisition process, several descriptively adequate grammars are available for a language L, then how will we know which of these a child’s evaluation metric selects as the most highly valued grammar for L? We seem to be left with a mechanism whose description is inaccessible to the empirical assessment that Chomsky insists is the only basis for understanding its design.

A solution to this problem was proposed with the emergence of the Principles and Parameters (P&P) model of UG. Chomsky (1981) suggests that UG consists of schematic constraints on the representations that comprise the syntactic derivation of a sentence, and on the movement operation which specifies the mappings between adjacent levels in the derivation. These constraints include parameters that allow for a finite number of possible values (ideally they are binary). Assigning v...

Table of contents

• Cover
• Halftitle page
• Title page
• Copyright
• Dedication
• Contents
• Preface
• Chapter 1: Introduction: Nativism in Linguistic Theory
• Chapter 2: Clarifying the Argument from the Poverty of the Stimulus
• Chapter 3: The Stimulus: Determining the Nature of Primary Linguistic Data
• Chapter 4: Learning in the Limit: The Gold Paradigm
• Chapter 5: Probabilistic Learning Theory for Language Acquisition
• Chapter 6: A Formal Model of Indirect Negative Evidence
• Chapter 7: Computational Complexity and Efficient Learning
• Chapter 8: Positive Results in Efficient Learning
• Chapter 9: Grammar Induction through Implemented Machine Learning
• Chapter 10: Parameters in Linguistic Theory and Probabilistic Language Models
• Chapter 11: A Brief Look at Some Biological and Psychological Evidence
• Chapter 12: Conclusion
• References
• Author Index
• Subject Index