Richard G. Schwartz

Do children pick and choose? An examination of phonological selection and avoidance in early lexical acquisition.

The influence of phonological selection and avoidance upon early lexical acquisition was examined within an experimental paradigm. During 10 bi-weekly experimental sessions, 12 children (1;0.21 to 1;3.15 at the outset) were presented with 16 contrived lexical concepts, each consisting of a nonsense word and four unfamiliar referents. For each child, eight words involved phonological characteristics which had been evidenced in production (in) and eight had characteristics which had not been evidenced in production or selection (out), in words were produced imitatively and non-imitatively in greater numbers and in earlier sessions than OUT words, providing evidence for the influence of selection and avoidance. The degree of phonetic accuracy of these two types of productions did not differ. These findings are discussed in terms of a proposal concerning early phonological representation and acquisition.

Speaker adaptation from a speaker-independent training corpus

A technique for using the speech of multiple reference speakers as a basis for speaker adaptation in large-vocabulary continuous-speech recognition is introduced. In contrast to other methods that use a pooled reference model, this technique normalizes the training speech from multiple reference speakers to a single common feature space before pooling it. The normalized and pooled speech is then treated as if it came from a single reference speaker for training the reference hidden Markov model (HMM). The usual probabilistic spectrum transformation can be applied to the reference HMM to model a new speaker. Preliminary experimental results are reported from applying this approach to over 100 reference speakers from the speaker-independent portion of the DARPA 1000-Word Resource Management Database.<<ETX>>

The forward-backward search algorithm

The authors introduce a technique that greatly speeds up expensive time-synchronous beam searches in speech recognition. The algorithm is called the forward-backward search and is mathematically related to the Baum-Welch forward-backward training algorithm. It uses a simplified forward pass followed by a detailed backward search. The information stored in the forward pass is used to decrease the computation in the backward pass by a large factor. An increase in speed of a factor of 40 with no increase in search errors was observed. The authors also describe how they have incorporated this algorithm into a real-time speaker-independent spoken language understanding system. One version of this is based on the 1000 word Resource Management vocabulary and is directed by a statistical class grammar. Another version has been incorporated into a military transportation planning application called DART (Dynamic Analysis Re-planning Tool).<<ETX>>

Neurophysiological Indexes of Speech Processing Deficits in Children with Specific Language Impairment

We used neurophysiological and behavioral measures to examine whether children with specific language impairment (SLI) have deficits in automatic processing of brief, phonetically similar vowels, and whether attention plays a role in such deficits. The neurophysiological measure mismatch negativity (MMN) was used as an index of discrimination in two tasks; one in which children ignored the auditory stimuli and watched a silent video and a second in which they attended to the auditory modality. Children with SLI showed good behavioral discrimination, but significantly poorer behavioral identification of the brief vowels than the children with typical language development (TLD). For the TLD children, two neurophysiological measures (MMN and a later negativity, LN) indexed discrimination of the vowels in both tasks. In contrast, only the LN was elicited in either task for the SLI group. We did not see a direct correspondence between the absence of MMN and poor behavioral performance in the children with SLI. This pattern of findings indicates that children with SLI have speech perception deficiencies, although the underlying cause may vary.

Non‐word repetition in Spanish‐speaking children with Specific Language Impairment (SLI)

BACKGROUND A number of previous studies have revealed that children with Specific Language Impairment have limitations in Phonological Working Memory as revealed by a task that requires them to repeat non-words of increasing syllable length. However, most published studies have used non-words that are phonotactically English. AIMS The purpose was to examine the repetition of non-words that are consistent with the phonotactic patterns of Spanish. The study also examined the relationship between non-word repetition performance and other language measures. METHODS & PROCEDURES Eleven Spanish-speaking children with Specific Language Impairment and 11 age-matched children with typical language development aged 8;3-10;11, who were part of a larger study of sentence processing, participated in the study. The primary data were the childrens repetition of 20 non-words, four at each syllable length (one, two, three, four and five syllables). The childrens productions were transcribed and scored for non-word, segmental and cluster accuracy as well as for error type. OUTCOMES & RESULTS The children with Specific Language Impairment performed more poorly on almost all measures of accuracy, but particularly in their production of three-, four-, and five-syllable non-words. Substitutions were the most frequent error type for both groups. Likelihood ratios indicated that non-word repetition performance is a highly accurate identifier of language status in these preselected groups. The childrens non-word repetition was highly correlated with most of the standardized language measures that were administered to the children. CONCLUSIONS The repetition of non-words consistent with Spanish phonotactics reveals word-length effects and error patterns similar to those found in previous studies. It extends these findings to older school-age Spanish-speaking children with Specific Language Impairment. Given the limited choices for instruments that can be used to identify children with Specific Language Impairment, a Spanish Non-word Repetition Task has the potential to be a valuable screening test for clinical and research purposes.

50% Efficient Solar Cell Architectures and Designs

Very high efficiency solar cells (VHESC) for portable applications that operate at greater than 55 percent efficiency in the laboratory and 50 percent in production are being created. We are integrating the optical design with the solar cell design, and have entered previously unoccupied design space that leads to a new paradigm. This project requires us to invent, develop and transfer to production these new solar cells. Our approach is driven by proven quantitative models for the solar cell design, the optical design and the integration of these designs. We start with a very high performance crystalline silicon solar cell platform. Examples will be presented. Initial solar cell device results are shown for devices fabricated in geometries designed for this VHESC program

The role of input frequency in lexical acquisition.

The influence of frequency of occurrence in input upon early lexical acquisition was examined within an experimental paradigm. Twelve children (1; 0·21 to 1; 3·15) were presented with 16 contrived lexical concepts, each involving a nonsense word and four referents, over ten experimental sessions. Within each concept two exemplars were presented frequently and two were presented infrequently. Overall the children named more frequently presented exemplars than infrequently presented exemplars. However, when the absolute number of presentations was held constant, distributed (infrequent) presentations led to greater acquisition than massed (frequent) presentations.

A mixed‐source model for speech compression and synthesis

This paper presents an excitation source model for speech compression and synthesis that allows the degree of voicing to be varied continuously by mixing voiced (pulse) and unvoiced (noise) excitations in a frequency‐selective manner. The mix is achieved by dividing the speech spectrum into two regions, with the pulse source exciting the low‐frequency region and the noise source exciting the high‐frequency region. The degree of voicing is specified by a parameter Fc, which corresponds to the cut‐off frequency between the voiced and unvoiced regions. For speech compression applications, Fc can be extracted automatically from the speech spectrum and transmitted. Experiments performed with the new model indicate its power in synthesizing natural sounding voiced fricatives and in largely eliminating the ’’buzzy’’ quality of vocoded speech. A functional definition of buzziness and naturalness is given in terms of the model.

Lexical access in children with and without specific language impairment: a cross‐modal picture–word interference study

Two experiments examined the time course of lexical information availability in 20 adults, 20 children (8;0-10;0) with typical language development, and in 20 children (8;0-10;0) with specific language impairment. A cross-modal picture-word interference paradigm was used in which participants named the pictures as quickly as possible while ignoring the phonologically and semantically related interfering words. A novel early phonological interference effect appeared in all groups. Similar temporal patterns were revealed for the adults and the typical language development group, supporting the notion of similar underlying lexicalization mechanisms. Parametric differences were found in overall response times and errors, with children responding slower and producing more errors than adults. The presence of a phonological facilitation effect suggests that children with specific language impairment utilize phonological primes to ease lexical access. Children with specific language impairment exhibited lingering semantic inhibition and a late semantic inhibition effect suggesting difficulty in processing semantic information. Data from all participants support the cascaded processing model of lexical access.

Automatic detection of new words in a large vocabulary continuous speech recognition system

A preliminary investigation of techniques that automatically detect when a speaker has used a word that is not in the vocabulary of a continuous-speech recognition system is described. A technique that uses a general model for the acoustics of any word to recognize the existence of new words is developed. Using this general word model, the correct detection of new words versus the false alarm rate is measured. Experiments were run using the DARPA 1000-word Resource Management Database indicate a detection rate for new words of 74% with a false alarm rate of 3.4%.<<ETX>>