Models of Language Acquisition: Inductive and Deductive Approaches (review)

Focused on language acquisition and computational modeling, this collection sheds new light on familiar debates in studies of language acquisition. Its twelve chapters are grouped into three parts: words, word formation, and word order. The volume includes a detailed table of contents and an extensive index.

Ch. 1 by Peter Broeder and Jaap Murre serves as a general introduction. It reviews the connectionist paradigm and research methodology utilized in computational modeling. In Ch. 2 Brian MacWhinney responds to the problem of representing lexical items in neural network models with an approach called lexicalist connectionism, a framework that sees word learning as the association of large numbers of phonological and semantic features. In Ch. 3, Noel Sharkey, Amanda Sharkey, and Stuart Jackson ask if simple recurrent nets (SRNs) are sufficient for modeling acquisition. They report some positive results, but emphasize problems with the use of SRNs as psychological models.

In the next chapter, Antal van den Bosch and Walter Daelemans examine models for reading aloud. After offering an overview of other approaches, they present their SPC (subword-phoneme correspondences) model. It is a nonconnectionist model that uses subword chunks and a corpus of word-pronunciation pairs. The final paper in Part 1 is by Steven Gillus, Walter Daelemans, and Gert Durieux. They ask if stress patterns can be learned before syntactic structure. Additionally, they report that IBL (instance-based learning) or ‘lazy learning’ constitutes a valid alternative to rule-based paradigms.

In Ch. 6 Richard Shillcock, Paul Cairns, Nick Chater, and Joe Levy utilize connectionist and nonconnectionist approaches to formulate statistical solutions to the speech segmentation problem faced by infants. Next, Jeffrey Mark Siskind devises an algorithm for semantic inference that takes into consideration nonlinguistic context as well as length and quantity of utterances.

The next two chapters address arguments about whether connectionist models can effectively deal with low-frequency rules. In Ch. 8 Gary F. Marcus discusses cognitive mechanisms that underlie learning and generalization. In the following chapter Rainer Goebel and Peter Indefrey consider the learning of the German -s plural. They question assertions suggesting that it is German’s only regular plural and assert that other methods of pluralization also apply as defaults. In Ch. 10 Ramin Nakisa, Kim Plunkett, and Ulrike Hahn test three computational models (single-route and dual-route) on three different inflectional systems (Arabic and German plurals and English past tense).

Part 3 of the text begins with Partha Nyogi and Robert C. Berwick’s examination of the formal models for learning in the principles-and-parameters framework. They show that analyses of memoryless learning algorithms can be formalized using Markov chains. In the final chapter, Loekie Elbers argues that production rather than comprehension is the primary source of input for children’s processes of linguistic analysis. Drawing a sharp distinction between children’s processes of comprehension and analysis, the author describes a three-phase model of output-as-input.

This collection of essays offers a comprehensive and illuminating overview of recent research centered on the interface between language acquisition and computational linguistics. As the book offers a sophisticated and at times highly technical treatment of the aforementioned topics, it will probably be of most interest to readers with knowledge of how computational modeling is used in linguistics and the behavioral sciences.