NIH Grant Abstract

Grant Number: 7R29DC03082-05
PI Name: BRENT, MICHAEL R.
PI Title:
Project Title: MECHANISMS OF SPEECH SEGMENTATION AND WORD DISCOVERY

Abstract: DESCRIPTION (Adapted from the Investigator's Abstract): By 7.5 months, infants show signs of recognizing the sounds of familiar words in sentential context and learning to recognize novel words presented in sentential context. This ability appears to be necessary for the growth of passive vocabulary that occurs between 10 and 16 months. The proposed research investigates the mechanisms underlying speech segmentation and its use in the isolation of novel words from sentential contexts. Understanding how these processes work in normal children is important for the early detection and effective remediation of communication disorders. It has been proposed that children's segmentation procedures may be guided by non-lexical cues, including stress cues, allophonic cues, phonotactic cues, and statistical cues. In addition, recognizing familiar words in an utterance can help to isolate the unfamiliar words. The INCDROP model is an explicit theory of the latter strategy and its relation to non-lexical cues. The aims of the proposed research are: (1) To test certain predictions of INCDROP by exposing adult subjects to speech consisting of artificial word-like units and asking them to make judgments about the units they perceive. (2) To test the effectiveness of INCDROP segmentation by computer simulation using large input samples of transcribed child-directed English and Dutch, represented both as phonemes and as atomic syllables. (3) To record, transcribe, and make available to others a longitudinal and cross-sectional sample of English spoken to children between the ages of 10 and 16 months. This will fill gaps in the available samples of child-directed speech at a time when such samples are becoming increasingly important. (4) To augment INCDROP so that it can take advantage of stress cues and statistical cues. This project will involve computer simulations and is expected to lead to further empirical predictions. These projects are part of a larger research program that brings multiple technologies to bear on understanding how children solve two intertwined problems: discovering the general sound systems of their languages and discovering the particular words of their languages.

Thesaurus Terms:
infant human (0-1 year), language development, preschool child (1-5), sensory discrimination, sequential perception, speech recognition, verbal learning
communication disorder, cue, developmental psychology, longitudinal human study, speech
behavioral /social science research tag, clinical research, computer simulation, human subject

Institution: WASHINGTON UNIVERSITY
LINDELL AND SKINKER BLVD
ST. LOUIS, MO 63130
Fiscal Year: 1999
Department: COMPUTER SCIENCE
Project Start: 01-AUG-96
Project End: 31-JUL-01
ICD: NAT'L. INST. ON DEAFNESS & OTHER COMMUNICATION DISORDERS
IRG: CMS