This book provides a broad overview of parameter-setting theory in first and second language acquisition and refines the theory by revisiting and challenging the traditional assumptions that underlie it, based on cross-linguistic language data that cover a range of syntactic and phonological phenomena. From an historical perspective on parameter-setting theory to an introduction to its role in computational linguistics, neurolinguistics, and language change, the reader will find a critique of the most commonly made arguments, as well as an index of all the syntactic, phonological, lexical, and morphological parameters presented in the literature to date. A closer look at the theory itself addresses the following questions: What does a parameter-setting approach to language acquisition entail? What are the underpinnings of the theory? What issues and problems remain to be solved? The empirical studies carried out to test the null subject parameter and verb movement parameter are reviewed to re-examine long-standing theoretical assumptions as well as the learnability implications for first and second language acquisition.

One of the main difficulties of applying an evolutionary algorithm (or, as a matter of fact, any heuristic method) to a given problem is to decide on an appropriate set of parameter values. Typically these are specified before the algorithm is run and include population size, selection rate, operator probabilities, not to mention the representation and the operators themselves. This book gives the reader a solid perspective on the different approaches that have been proposed to automate control of these parameters as well as understanding their interactions. The book covers a broad area of evolutionary computation, including genetic algorithms, evolution strategies, genetic programming, estimation of distribution algorithms, and also discusses the issues of specific parameters used in parallel implementations, multi-objective evolutionary algorithms, and practical consideration for real-world applications. It is a recommended read for researchers and practitioners of evolutionary computation and heuristic methods.

Recent developments in linguistic theory have led to an important reorientation of research in related fields of linguistic inquiry as well as in linguistics itself. The developments I have in mind, viewed from the point of view of government-binding theory, have to do with the character ization of Universal Grammar (UG) as a set of subtheories, each with its set of central principles (perhaps just one principle central to each subtheory) and parameters (perhaps just one for each principle) according to which a principle can vary between an unmarked ('-') and a marked ('+') para metric value (Chomsky, 1985; 1986). For example, let us assume that there is an X-bar theory in explanation of those features of phrase structure irreducible to other subtheo ries of UG. Within X-bar theory variation among languages is then allowed only with respect to the position the head of a phrase occupies in rela t ion to its complemen ts such that the phrases of a language will be either right- or left-headed. Thus languages will vary between being right-headed in this respect (as in Japanese phrase structure) and being left-headed (as in English phrase structure). Everything else about the phrase structure of particular languages will be fixed within X-bar theory itself or else it will fallout from other subtheories of UG: Case theory; 0-theory, etc. (Chomsky, 1985:161-62; Chomsky, 1986:2-4; and references cited there). Hatters are the same in other modules of grammar.

In May 1985 the University of Massachusetts held the first conference on the parameter setting model of grammar and acquisition. The conference was conceived in the belief that there is a new possibility of tightly connecting grammatical studies and language acquisition studies, and that this new possibility has grown out of the new generation of ideas about the relation of Universal Grammar to the grammar of particular languages. The papers in this volume are all concerned in one way or another with the 'parametric' model of grammar, and with its role in explaining the acquisition of language. Before summarizing the accompanying papers, I would like to sketch the intellectual background of these new ideas. It has long been the acknowledged goal of grammatical theorists to explicate the relation between the experience of the child and the knowledge of the adult. Somehow, the child selects a unique grammar (by assumption) compatible with a random partially unreliable sample of some language. In the earliest work in generative grammar, starting with Chomsky's Aspects, and extending to such works as Jackendoffs Lexicalist Syntax (1977), the model of this account was the formal evaluation metric, accompanied by a general rule writing system. The model of acquisition was the following: the child composed a grammar by writing rules in the rule writing system, under the constraint that the rules must be compatible with the data, and that the grammar must be the one most highly valued by the evaluation metric.

The book addresses some important issues in second language acquisition research, such as access to Universal Grammar, the factors that account for success or failure in L2 learning, as well as parameter (re)setting, within the generative framework. The empirical data against which the research hypothesis is tested come from two main studies: a longitudinal one, involving beginner learners, and an experimental one, with elementary and advanced learners; all subjects are adults learning English in a formal instruction setting, with Romanian as L1. Starting from Chomsky’s (2005) account of first language acquisition in terms of three factors, namely genetic endowment, experience and non-linguistic principles, the first chapter of the book explores the question of whether the same three factors may be operating in second language learning. The rest of the book focuses on more specific research issues related to access to Universal Grammar and parameter (re)setting. Since L1 acquisition is defined mainly as the process of parameter setting on the basis of the input, parameter setting is expected to be at the core of L2 learning as well. However, in the literature, it has often been argued that one of the most important differences between L1 and L2 acquisition is related precisely to this process. The parameter which the book focuses on is the split IP parameter (Bobaljik and Thráinsson, 1998), in the case of Romanian L1 learners of English as L2. Romanian is (+ split IP) while English is (- split IP). In this case, the learners would need to move from the positive (+) value of the parameter to the negative (-) one, i.e. from a wider set of property values to a more restrictive one. There are three possible scenarios: (i) the L2 learners set the negative (-) value from the target language from the very beginning, on the basis of input alone; (ii) the learners first transfer the positive (+) value from their L1 and reset this value to the L2 negative (-) one as they progress in acquiring the L2; the final state is L2-like; (iii) the learners first transfer the positive (+) value from their L1 but resetting is impossible; the L2 underlying representations are different from those of native speakers even with advanced speakers, and what may account for final L2 state is parameter learning. These scenarios are explored with the help of the data provided by longitudinal and experimental studies.

This book develops a new approach called parameter advising for finding a parameter setting for a sequence aligner that yields a quality alignment of a given set of input sequences. In this framework, a parameter advisor is a procedure that automatically chooses a parameter setting for the input, and has two main ingredients: (a) the set of parameter choices considered by the advisor, and (b) an estimator of alignment accuracy used to rank alignments produced by the aligner. On coupling a parameter advisor with an aligner, once the advisor is trained in a learning phase, the user simply inputs sequences to align, and receives an output alignment from the aligner, where the advisor has automatically selected the parameter setting. The chapters first lay out the foundations of parameter advising, and then cover applications and extensions of advising. The content • examines formulations of parameter advising and their computational complexity, • develops methods for learning good accuracy estimators, • presents approximation algorithms for finding good sets of parameter choices, and • assesses software implementations of advising that perform well on real biological data. Also explored are applications of parameter advising to • adaptive local realignment, where advising is performed on local regions of the sequences to automatically adapt to varying mutation rates, and • ensemble alignment, where advising is applied to an ensemble of aligners to effectively yield a new aligner of higher quality than the individual aligners in the ensemble. The book concludes by offering future directions in advising research.

View the abstract.