Peter W. Jusczyk (1997), The Discovery of Spoken Language, MIT Press, Cambridge, Massachusetts and London, England, 314 pages
Reviewed by Linda Shocky <L.Shockey@reading.ac.uk>
Peter Jusczyk's The Discovery of Spoken Language is about how human infants acquire the phonetic skills needed to perceive and produce language. It starts with some stage-setting: Chapter 1 is an overview of the sort of task a newborn is faced with in acquiring the phonetics of a language; Chapter 2 is a brief bird's-eye view of language acquisition research. Chapter 3 gives a brief history of research on speech perception by infants. Chapter 4 discusses how the child's speech perception develops during the first year of life. It reviews that now-familiar notion that a very large number of possible language contrasts (though not all) are discriminable by infants shortly after birth, but that gradually children develop a preference for the sounds of their own language(s) and, eventually, an inability to distinguish a large number of contrasts which are distinctive only in other languages. A few pages into Chapter 4, Jusczyk begins to cite his own work in this area, though other researchers are not neglected.
One would be hard-pressed to remain unawed by Juszcyk's energy and dedication to infant language acquisition research. He has done an enormous number of experiments (most researchers would flag at the idea of simply finding and getting permission to experiment with so many babies), in what seems an impressively thoughtful and logical sequence: when do infants begin to begin to recognise words from their own language? Is this recognition of individual sounds or phonotactic sequences? Are they sensitive to frequency of occurrence? Do prosodic patterns play a part? When do they begin to find words in connected speech? When they do, are they using one cue or many?
Chapter 5, on the role of memory and attentional processes to the developement of speech perception reveals another impressive set of experiments which suggest that babies represent speech in terms of syllables rather than smaller phonetic segments and that they are capable of detecting differences in syllabic identity and in number of syllables in a short utterance. Also indicated are that infants are not yet very good at following speech in noise, that they ARE good at recognising voices they have heard before, and that they MAY BE good at making fine distinctions because they focus on smaller amounts of speech than older humans.
Chapter 6, How Attention to Sound Properties May Facilitate Learning Other Elements of Linguistic Organisation, suggests that infants are sensitive to prosodic markers of structure and gives evidence that they prefer listening to speech in which the phrase boundaries and their prosodic markers coincide in their native language. Evidence also exists that they prefer listening to speech which is phonotactically correct for their language, and to speech in which the words come in the correct order.
Chapter 7 begins with a cursory review of previous work on the acquisition of speech production and goes on to suggest important links between production and perception in early stages of language acquisition. Since the latter portion of their first year is when children are beginning to speak, it is suggested that the narrowing-down of ability to perceive foreign contrasts is related to the attempt to coordinate the perceptual and articulatory systems. I.e., at this time, children focus on the sounds they are learning to pronounce. The ability to distinguish most of the non-native sounds is then gradually lost because there is no articulatory-perceptual link.
Chapter 8 is called "Wrapping Things Up," which is a covert reference to Jusczyk's own model of infant speech perception, WRAPSA. In this chapter, he tries to assess where we now are in terms of understanding child language acquisition. He believes that the most likely model based on what we know now is one of "innately guided learning": humans, like other organisms, are programmed to learn particular kinds of things in particular ways. The infant cognitive system combines this propensity with whatever data is present in order to acquire the language(s) spoken in its environment. It does this through warping the perceptual space through foregrounding of the features which are significant to the target language. Just as correct pronunciation of a language involves a specific overall articulatory posture, learning to perceive includes language-specific tuning of perceptual facilities. As the infant brain matures, the child develops the ability to integrate information from different sources, so the movement is away from a general-purpose recogniser of differences and towards a much more sophisticated recogniser of a particular language.
The Word Recognition and Phonetic Structure Acquisition model assumes a preliminary level of analysis which stores away in the infant brain all of the perceptual features which the system is capable of extracting of EACH UTTERANCE. (Multiple tokens will thus have multiple representations) This is the portion of the system which allows young infants to detect differences irrelevant to their own target language: "the description that emerges is neutral with respect to the language that is spoken." (p.215) This repre- sentation decays quickly. The next component is a weighting scheme which focuses attention on crucial language-specific features. This may be mildly stimulated at birth, but with more language input very rapidly effects a language-specific transformation of perceptual space. The third component extracts patterns (including syllables and prosodic information) from the weighted output of the second and stores these patterns in memory, where they can be said to comprise the infant lexicon. The fourth compares new patterns with old ones to attempt a match. Feature similarities will cause old patterns (traces) to be stimulated. Though syllabic information is present, the goal is to recognise words. As more tokens of each utterance are collected, more traces will be activated by new tokens, so recognition of patterns will soon become more efficient and will eventually lead to extraction of words.
This model differs from any other extant model of speech or word recognition, creatively combining elements of earlier models in a novel fashion. It predicts most of the experimental results discussed in Chapter 4, though, as Jusczyk himself says, "more empirical data about the details and the course of development of speech perception capacities are essential for [evaluation]." (p. 229)
An appendix describes techniques used in studies of infant speech perception.
I join the star-studded legion which praise the book on its dust jacket: it is a must-read for anyone with a serious interest in language acquisition.
One minor point: Jusczyk should have asked someone to check his phonetic transcriptions of English.
Two quibbles with the format of the book: 1) Jysczyk uses very few visual aids: there is a total of four illustrations, all of which occur in the first 47 pages. This makes the rest of the book seem quite heavy going. There may be people who can read this book straight through, but I missed points where I could stop, take a breath, and summarise what I had assimilated. I would have liked to see a table like this at the end of each section:
Experimenter(s) Year Question Investigated Technique Result (Can infants detect..)
Jusczyk et al 1993 native words? 9mo/LT Yes Friederici et al 1993 Dutch syllable boundaries? 9mo/LT Yes
This would be useful not only to clarify the first reading, but also as a memory aid when returning to the book.
2) We are consistently told the results of Jusczyk's experiments without any evidence of the validity of his interpretation: we don't know how many infants participated in each experiment, what tests of significance were used and what level of significance was regarded as passable. One notes, upon perusing a selection of his articles in Perception & Pschophysics and JASA that Jusczyk includes both illustrations and statistics in journal articles. Granted that his Discontinuity in Sucking Behaviour graphs have a certain 'seen one, seen them all' quality....as a reader of the book, you haven't necessarily seen one, and doing so strikes me as useful.
Presumably, the lack of illustrations and statistics is part of an effort to keep the book a reasonable length and affordable, and this must be applauded, if reluctantly.
Linda Shockey is a lecturer in linguistics at The University of Reading, England. She specialises in acoustic phonetics and the phonology of connected speech, especially English.