The Development of Language Processing Strategies

A Cross-linguistic Study Between Japanese and English

Reiko Mazuka

• Publisher: Psychology Press
• Year: 2014
• Print ISBN: 9781138012318
• eBook ISBN: 9781317781547
• Topic: Personal Development
• Subtopic: History & Theory in Psychology

Ever since the notion of explanatory adequacy was promoted by Chomsky in his 1965 Aspects, linguists and psycholinguists have been in pursuit of a psychologically valid theory of grammar. To be explanatorily adequate, a theory of grammar can not only describe the general characteristics of a language but can also account for the underlying psychological processes of acquiring and processing that language. To be considered psychologically valid, a grammar must be learnable by ordinary children (the problem of acquisition) and must generate sentences that are parsable by ordinary people (the problem of processing). Ultimately, the fields of language acquisition and processing are concerned with the same goal: to build a theory that accounts for grammar as it is acquired by children; accessed in comprehension and production of speech; and represented within the human mind. Unfortunately, these two fields developed independently and have rarely been well-informed about each other's concerns. Both have experienced past difficulties as a result.

Recently, new models have been developed with full consideration to cross-linguistic diversity. Gone are many of the basic assumptions of conventional models, and in their place a variety of innovative and more flexible assumptions have emerged. However, in their attempt to address cross-linguistic issues, these processing models have yet to fully address the developmental challenge: How can a child without a stable grammar process language and still manage to acquire new grammar?

This book attempts to develop a model of language processing that addresses both cross-linguistic and developmental challenges. It proposes to link the setting of a basic configurational parameter during language acquisition to the different organization of processing strategies in left- and right-branching languages. Based primarily on Mazuka's doctoral dissertation, this volume incorporates various responses to the original proposal as well as the author's responses to the comments.

CHAPTER ONE
Introduction

The Problems

In many language development models, young children are assumed to process language in a qualitatively different manner than older children and adults. For example, in a model in which children learn grammatical rules by bootstrapping from functional categories (e.g., Pinker, 1984), the language processing strategies¹ that are used by young children are assumed to be primarily driven by functional relationships, such as agent and patient, rather than grammatical relations, such as subjects and objects.

In many of the formal models of language acquisition (Berwick & Weinberg, 1984; Gibson & Wexler, 1994; Wexler & Culicover, 1980), on the other hand, it is taken for granted that the processing strategies children use remain unchanged across development. As discussed in detail in chapter 2, this assumption is critical to these models because the basic mechanisms of language acquisition in these models are parser driven.

Despite the critical significance that both these models place on language processing strategies, few studies directly investigate the actual processing strategies that young children use in comprehending and producing sentences in a language.

In adult psycholinguistic literature, it has generally been assumed that processing strategies are universal, that is, although the grammar used to process a language differs from one language to another, the strategies that use such grammatical information remain constant across languages.

The processing of natural language by humans is extremely fast. For example, in the classic studies using the shadowing paradigm, Marslen-Wilson (1973, 1975) elegantly demonstrated that English-speaking adults restored semantic anomalies in stimulus sentences while they were shadowing them very closely, that is, within 300 to 800 msec. Evidence such as this has been interpreted to suggest that sentence processing must proceed incrementally, more or less from left to right.

To date, the majority of research in natural language processing, both formal and experimental, has been done on English. As discussed later in detail, English is a right-branching, head-initial language. This means that in principle, a main clause will precede a subordinate clause, and a grammatical head will precede its complement. On the other hand, in a left branching, head-final language such as Japanese, a subordinate clause will precede a main clause, and a complement will precede its grammatical head. Therefore, when a person hears a sentence and tries to comprehend it incrementally from left to right, the order of constituents that arrive in real time is reversed between English and Japanese.

If processing strategies for natural languages are universal, as has been hypothesized, both Japanese and English should be processed by the same strategies. However, when the processing must proceed extremely quickly from left to right in real time, can two languages be processed by the same strategies if the order of their constituents is essentially a mirror image of each other, as with Japanese and English?

We now know from numerous empirical studies that young children acquiring Japanese and English are sensitive at a very early stage to the systematic, parametric, and structural differences between the two languages (e.g., Lust, 1986, 1987, 1997). If processing strategies remain constant over development, as formal language acquisition models have proposed, young children acquiring Japanese and English should process these languages with the same strategies as Japanese and English adults. However, when children acquire languages that are as diverse as English and Japanese, can they be using the same strategies as English- and Japanese-speaking adults?

Alternatively, if children's processing strategies are qualitatively different from those of adults', do children who are acquiring different languages go through the same developmental course, or do they have to go through different trajectories according to their target languages?

In this book, the twin questions of the universality of natural language processing strategies across languages and over development are examined experimentally. These two questions are rarely investigated together, and the insights of one field are not always taken into consideration in the other. However, these questions can be systematically investigated only by taking into account both of the factors from the cross-linguistic and developmental variables. In this book, these questions are examined through a study of English and Japanese children's processing of complex sentences.

Linguists and psycholinguists have long been interested in the question of universality in our linguistic competence. However, universality of language processing strategies within the domain of language performance has rarely been scrutinized with the same intensity. Chomsky (1987) summarized this state of affairs as follows. "It has generally been assumed that the parser ... is fixed and does not undergo growth and maturation, or learning if such exists. The reasons for assuming this, apparently, are just that nothing is known about the matter, so we might as well adopt the simplest hypothesis" (p. 14).

The Theoretical Framework

In this book, the general framework of the Principles and Parameter Approach to language acquisition (Chomsky, 1975, 1981, 1986) is adopted as the basis of the analysis. According to this framework, Universal Grammar (UG) contains principles that characterize all natural languages, whereas grammatical parameters specify the dimensions along which languages may differ, and what die possible variations may be. Children are biologically programmed to entertain only those hypotheses about natural languages that conform to the principles of UG. They are also programmed to attend to specific types of possible language variation, and set the value of such parameters very early with only limited experience. Once a parameter is set, a child should, theoretically, be able to deduce various aspects of the language.

This approach provided a theoretical framework for numerous studies during the past few decades. A wide variety of studies using this framework have shown that this approach is productive in explaining complex phenomena of language acquisition (cf. Lust, 1986, 1987, 1997; Roeper & Williams, 1987; Wanner & Gleitman, 1982). In the following chapters of this book, various motivations for adopting this framework are presented. Specifically, the particular contrast in the basic structures of the two languages studied in this book, namely, Japanese and English, are captured in terms of a binary configurational parameter, the Principle Branching Direction Parameter (see ch. 4 for the definition of this parameter), and I propose that by linking the setting of this parameter and children's sentence processing strategies, we can provide a model in which the development of language processing strategies by Japanese- and English-speaking children can be accounted for.

The Structure of the Book

The structure of the book is as follows: In chapter 2, I first discuss the role of language processing in models of language acquisition. I show that although few studies have been done on the actual development of children's language processing strategies, most models of language acquisition critically depend on how children process linguistic input in order for a child to acquire grammar. I argue that the assumptions made by these models are paradoxical in the sense that children would have to be able to process not only all sentences that are grammatical according to the grammar the child currently holds, but also sentences that are beyond that.

Chapter 3 turns to models of language processing that have been developed on the basis of English. I show that the syntactic properties of a left-branching language such as Japanese challenges the basic assumptions of these models. In chapter 4, I propose a model in which setting of a basic configurational parameter, the Branching Direction Parameter, is linked to the different organization of language processing strategies for left- and right-branching languages. I first argue that the Branching Direction Parameter can be set preverbally using prosodic cues. As a deductive consequence of setting this parameter, children would adopt processing strategies that are appropriate for their languages.

Chapter 5 reviews alternative models that also attempt to account for cross-linguistic diversities between Japanese and English. The dominant approach in this endeavor is the single parser approach, in which the parser is modified to be powerful enough to handle both left- and right-branching languages. I show that the three proposals made within this framework—the head driven approach, minimal commitment approach, and full attachment approach—all share the difficult challenge of constraining the power of the parser such that it can still explain the limitation of human language processing behavior. In addition, I argue that without serious consideration of developmental issues as well, a model of language processing cannot be complete.

In chapter 6, I turn to experimental studies of sentence processing with both adults and children. I focus the review of literature on the processing of complex sentences, because it is the relevant domain over which the proposed model of parameterized parsing is defined. In particular, I focus on a study by Bever and Townsend (1979) that tested English adults' processing of main and subordinate clauses with semantic and lexical probe latency tasks. I show how the experimental data bear on predictions made by the single, universal parsing model and by the parameterized parsing model. These predictions are tested using semantic- and lexical-probe latency tasks (PLT) with matched experiments using English and Japanese children. Specifically, I tested semantic and lexical accessibility to main and subordinate clauses in children's processing of complex sentences, which Bever and Townsend (1979) studied with adult English speakers.

In chapter 7, I present data from four experiments: semantic PLT with children between ages 6 and 8, and lexical PLT with children between ages 4 and 6, in both English and Japanese. In chapter 8, these data are discussed in light of the predictions discussed in chapter 6. I argue that although the results from the current study cannot conclusively prove the parameterized parsing model, they are consistent with the predictions made by this model and offer advantages not only in the area of natural language processing, but also in an integral explanation of language acquisition and language processing.

¹The term processing strategy is used to mean a general heuristic that uses linguistic and norilinguistic information to comprehend natural language.

CHAPTER TWO
The Role of Language Processing in Language Acquisition: The Paradox of Processing Sentences to Learn Grammar

Traditionally, the focus of research in language development has been on what children know about language, whereas the question of how children process language has rarely been addressed directly (e.g., Frazier & de Villiers, 1990). Despite the lack of research in this area, most models of language acquisition are nevertheless contingent on particular assumptions about children's abilities to process incoming sentences in order to acquire the grammar of the language.

In this chapter, I discuss the intricacies of the relation between children's developing knowledge of grammar and their development of language processing strategies. I show that it is paradoxical to assume that children are able to parse a sentence in order to acquire grammar, whereas for adults it is assumed that grammar is required to parse a sentence.

Basic Terms

The term processing refers in general to "a mapping from one kind of information to another" (Marr, 1980, p. 24). Language processing refers to "a sequence of operations, each of which transforms a mental representation of a linguistic stimulus into a mental representation of a different form (usually more abstract in the case of perception"; Forster, 1979, p. 28). Of the various levels of language processing (e.g., phonetic, lexical, syntactic, semantic, or discourse), we are specifically concerned with natural language parsing, defined in general as the syntactic and semantic analysis in real time of a surface string into constituent structures. In this book, the term language processing strategies is used to refer to the actual strategies that children and adults use in processing language, and language parsing or parser to refer to formal models for natural language processing.

The Role of Processing in Language Acquisition Models

In any model of language acquisition, children will not be able to acquire language unless they can process the language that surrounds them. Models differ in terms of what the target grammar should look like and how the acquisition of such a grammar can be accomplished.

When children are born, the stimulus they are presented with is a continuous flow of speech. From this, they must somehow extract relevant linguistic units and map them onto linguistic forms. In models such as the Parameter Setting Approach (Chomsky, 1975, 1981, 1986), children must first process a string of speech in order to ascertain its grammatical parameters. In models where children are assumed to bootstrap into linguistic forms from other types of information—semantic (e.g., Pinker, 1984), distributional (e.g., Maratsos & Chalklery, 1980), or prosodic (cf. Gleitman & Wanner, 1982; Morgan & Newport, 1981)—they must process enough input to determine the correspondence between these types of information and the linguistic forms. In models that assume that children have to learn linguistic forms through generalization over individual words (cf. Tomasello, 1992, 1995), children must process linguistic input on the basis of prelinguistic categories before they gradually shift to more adultlike modes of processing. In each of these models, children are assumed to process the linguistic input in a particular way at the beginning, the outcome of which giving them the critical data necessary to gain access to linguistic categories.

In all these models, children process linguistic input differently depending on whether it is before or after they have gained access to linguistic forms. In a semantic bootstrapping model, for example, a child would process an incoming sentence on the basis of semantic categories, for example, agent and action, until he or she can successfully bootstrap into grammatical categories. Once he or she has gained access to linguistic categories such as nouns and verbs, the same child would use the newly gained grammatical categories to process the same sentence.

Models differ, however, in their predictions of when such a shift should occur. In models such as the Parameter Setting Approach, the way young children process language is assumed to be essentially the same as older children and adults, because except for during the very beginning stages, children have access to linguistic categories similar to those of adults. This view is called a continuity of language processing strategies. By contrast, in models that assume children take a long time to gain access to linguistic categories, young children are predicted to be using qualitatively different types of information in processing language compared to older children and adults. For example, according to a functionalist model proposed by Tomasello (1992, 1995), early word combinations are based on "children's conceptualizations of events (e.g., kissing) and their identification of discrete participant roles in these events (e.g., 'kisser' and 'kissee'"; Tomasello, 1995, p. 151), not on the grammatical relations such as subject and verb. Thus, children at this stage would process a sentence on the basis of events and participants, whereas older children and adults would process such sentences based on grammatical relations. This view is called a discontinuity of language processing strategies.

Competence and Performance

In adult psycholinguistics literature, it is generally assumed that a native speaker's grammatical knowledge is stable (altho...

Table of contents

• Cover
• Half Title
• Title
• Copyright
• Dedication
• Contents
• Preface
• CHAPTER 1 Introduction
• CHAPTER 2 The Role of Language Processing in Language Acquisition: The Paradox of Processing Sentences to Learn Grammar
• CHAPTER 3 The Challenge of Processing Japanese
• CHAPTER 4 Proposal: Linking a Grammatical Parameter to Processing Strategies
• CHAPTER 5 Alternative Proposals
• CHAPTER 6 Empirical Predictions of the Parameterized Parsing Model
• CHAPTER 7 Processing Experiments
• CHAPTER 8 Discussion and Conclusion
• References
• Author Index
• Subject Index