190 LANGUAGE, VOLUME 75, NUMBER 1 (1999) ond language (L2) acquisition by untutored adult immigrants . Already published are studies of utterance structure, temporality, and the lexicon, all based on data (stored in CHILDES) from the same 40 learners. The approach is cross-linguistic and longitudinal, with data collected 20-25 times over a two-year interval from learners of varied first languages (Lis) acquiring varied L2s ab initio. Each observational session elicited learner language referring to spatial relations via picture description and stage direction tasks, supplemented occasionally by other communicative activities and by the informants' incidental uses of spatial terms in conversation with native L2speaking experimenters. This design was held consistent by research teams at different sites, with members of those teams contributing separate chapters to the book. Ch. 1, by Angelika Becker and Clive Perdue, gives general background. In Ch. 2, B presents a theoretical framework pitched at somewhat too abstract a level. More liberal use of examples and a visual display of the relationships among the complex taxonomy ofspatial relations and terms would help. Ch. 3, by Mary Carroll , reports the acquisition ofEnglish by two Italian speakers; B's Ch. 4 addresses acquisition of German by four Italian speakers and one Turkish speaker; Ch. 5, by Jorge Giacobbe, Clive Perdue, and Rémi Porquier describes the acquisition of French by two Spanish speakers and one Moroccan Arabic speaker. Carroll's Ch. 6 concludes with a survey ofthe results. In the absence of a theory of acquisition in the relevant domain, the approach is explicitly 'datadriven and inductive' (2), making it easy to get lost in the masses of descriptive data presented for each informant in Chs. 3-5. Those data are revealing both with regard to the process of acquisition and with regard to what they communicate about the properties of the source and target languages. They also, happily, model what it means to view L2 acquisition (relatively) unobstructed by the 'comparative fallacy ', that is, free from the tendency to see interlanguage grammars as flawed or partial versions of native speakers' grammars. I can only summarize here some of the overall findings. All early learners arrive at a comparable 'basic variety' of their L2. They may first represent spatial relations by simple juxtaposition, then start to encode dynamic relations (movement along an axis, or movement toward or away) in whatever formal devices the L2 exploits. For static spatial relations , topological terms (inside, near) emerge before axis-based terms (above, left) because the latter require perspective-based computation. Past the basic variety, how space is conceived and represented in the Ll shapes learners' interlanguage systems, as evidenced in acquisitional order variations between learners of different Lis. The research finds that divergence between Ll and L2 in the conceptualization of space and in the encoding of spatial relations leads to slower and more idiosyncratic developmental paths. In addition, greater conceptual complexity (for example, that represented in the word between) delays acquisition across the board. [Margaret Thomas, Boston College.] Computational approaches to language acquisition. Ed. by Michael R. Brent. Cambridge, MA: MIT Press, 1997. Pp. 199. There has been a movement in recent years away from rule-based theories of language acquisition to understanding how developmental and statistical constraints on natural language development give rise to infant speech and language processing ability. These models can be divided into two broad categories : First, artificial neural network models have provided a limited developmental account (in reading, for example) of how mappings from orthography to phonology can be 'learned' by statistical approximation using many ubiquitous nonlinear predictors. The second approach has sought to determine the statistical conditions necessary for developing language and specifying processes which might satisfy these conditions (the 'bootstrapping' approach). Autonomous bootstrapping is the subject of this volume of papers reprinted from a special issue of Cognition. Brent presents an overview of computational approaches to the acquisition of natural language in infancy which are based on bootstrapping hypotheses and integrates key results from the four research papers reprinted in the volume. These papers address semantic , prosodie, and syntactic bootstrapping issues although an account of the acquisition of regular and irregular verb morphologies is absent. The papers are presented in a complementary way...
Read full abstract