- Email: [email protected]
Relations between word meanings and identification of words in spoken sentences by hearing-impaired children
JOURNAL
OF COMMUNICATION
DISORDERS
15 (1982), 363-373
363
RELATIONS BETWEEN WORD MEANINGS AND IDENTIFICATION OF WORDS IN SPOKEN SENTENCES BY HEARING-IMPAIRED CHILDREN MICHAEL
STINSON
National Technical Institute forthe Deaf, Rochester
Institute of Technology
To examine effects of contextual words upon the identification of test words in sentences, 21 hearing-impaired children, 7-12 yr of age, received word-discrimination training followed by sentence testing. Word-discrimination training was conducted to ensure that children could discriminate the contextual words in isolation. The subsequent sentence testing was the more important phase of the experiment, since it permitted observation of the use of contextual cues to facilitate the identification of the correct test word. The relationship between the contextual words and test words was either congruous, incongruous, or neutral, as predetermined from normative data. After hearing each sentence, children decided whether they had heard the test word or its alternative. The alternative word differed in either the initial or final consonant. Significantly more correct discriminations were made on congruous and neutral sentences than on incongruous sentences. These results indicated that hearing-impaired children used contextual cues to identify the correct alternative.
Introduction Hearing-impaired persons often have difficulty following speech because they cannot hear or lip-read certain speech sounds that permit the identification of particular words. It may be possible, however, for hearing-impaired persons to extract semantic and syntactic information from intelligible words and construct hypotheses about unintelligible words based on this information. Recent research with the Speech Perception in Noise (SPIN) test has, in fact, demonstrated the use of contextual cues by adults having mild or moderate hearing losses (Kalikow et al., 1977; Kalikow et al., 1976). However, the extent to which a hearing-impaired child can use contextual cues is not clear for two reasons. First, if children lose their hearing early in life, they may acquire a different linguistic base than normal-hearing children. Consequently, the children may experience difficulty using syntactic and semantic information to facilitate their understanding of difficult-to-hear speech. Second, the decrement in hearing ability may force these listeners to code each word as
Address correspondence to Michael Stinson, Department of Educational Research and Development, National Technical Institute for the Deaf, Rochester Institute of Technology, One Lomb Memorial Drive, Rochester, NY 14623.
B Elsevier Science Publishing Company, Inc., 1982 52 Vanderbilt Ave., New York, NY 10017
0021-9924/82/050363-l
1$02.75
364
MICHAEL
STINSON
they hear it because they cannot identify enough words to form a useful contextual base. The perception of spoken sentences by hearing-impaired children has received little research attention, with studies conducted by Wilcox and Tobin (1974), Davis and Blasdell (1975), and Erber and McMahon (1976) being exceptions. None of these studies, however, investigated effects of sentence context upon identification of a particular word in a sentence. Investigation of hearingimpaired childrens’ use of contextual cues when listening to sentences may have implications for aural rehabilitation. Results of the present study were expected to provide information pertaining to semantic aspects of context that may be important in programming sentences for auditory training of hearing-impaired children. No attempt was made, however, to identify materials for clinical testing. This study examined the effect of contextual words upon the identification of test words by hearing-impaired children. The following hypotheses were made about the manipulation of contextual words in auditory presentations of sentences: 1. Discriminating between two similar sounding words is more dtflicult when test words are presented in a sentence with neutral context than when the test words are presented in a sentence with a biasing word that is relatively easy to discriminate from a corresponding word in a second sentence. For example, it should be relatively easy for children to discriminate betweenpail and tail if they occur in the respective sentences: The little girl has a pail. The little dog has a tail. After hearing one of these sentences, children are asked whether they heard pail or tail. In this case, children will usually select the correct alternative because its meaning is consistent with that of the biasing word. On the other hand, it should be relatively difficult for children to discriminate betweenpail and tail if they hear one of them in the neutral context, “The was grey.” Similar sounding pairs of words, such as pail and tail, are likely to be confused by hearing-impaired children. In the experiment the alternatives for the test word always differed by one consonant and were based on consonant confusions for impaired hearers reported by Owens et al., ( 1972). 2. Children will more frequently identtfi the correct word in a pair when it is in a sentence where the relationship between the biasing and test word is congruous than in a sentence where the relationship is incongruous (e.g., “The little girl has a tail”). If listeners perform better on sentences with a relatively congruous relationship between the biasing and the test word than on sentences with an incongruous relationship, this pattern would suggest that biasing words are being related to the test word. Presumably, listeners would
WORD MEANINGS
365
AND IDENTIFICATION
do worse on the incongruous sentences because the biasing word would induce the listeners to ignore the accoustical information in the test word. 3. Children will less frequently identify the test word correctly when it is in a sentence with an incongruous relationship between the biasing and the test word than when it is in a sentence with neutral context. In this study, hearing-impaired children were given auditory-discrimination training with the contextual words prior to participating in the test conditions. The discrimination training was conducted to ensure that the children could discriminate the biasing words in isolation. If children could discriminate the biasing words in isolation, it was more reasonable to assume (a) that the children could identify the biasing words in sentences and (b) that varying the words in sentences would produce semantic manipulations. Method Overview There were two phases in the experiment. In the first phase normative data were collected from normal-hearing children. In the second, more important phase word-discrimination training and sentence testing was conducted with hearing-impaired children. The second phase was divided into two periods. In period 1, sentences were presented that contained a congruous relationship between the biasing and test words. In the period 2, sentences were presented that contained an incongruous relationship between the words. In addition, sentences having a neutral context were presented in both periods. All sentences in both periods were presented through headsets. Collection Sentences
of Normative
Data for Congruous
and Neutral
Forty-four normal-hearing third and fourth graders between the ages of 9 and 11 yr (mean 9.8 yr) were group-tested in their classrooms in order to identify (a) congruous sentences, in which the test word was predictable from the context, and (b) neutral sentences, in which the test word was not predictable from the context. Materials. The paper-and-pencil instrument consisted of 11 sentence triads such as: Incomplete
Sentence
Sentence
Type
Responses
1. The little girl has a ~.
Congruous
pailltail
2. The little dog has a ~.
Congruous
pailltail
Neutral
pailltail
3. The
was grey.
366
MICHAEL
STINSON
For each of the 11 triads, there were two alternative responses. The congruous sentences were written with the expectation that there would be congruity between the context and one of the alternatives. (They are called “congruous” in order to contrast them with the “incongruous” sentences which were used in the experiment proper.) The neutral sentences were written with the expectation that contextual cues would be absent. These sentences were randomly distributed and printed in a booklet with two sentences per page. For each of these incomplete sentences, the two alternative words for completing the sentence (e.g., pail/tail) were printed below the sentence, with a picture for each alternative. The experimenter read each incomplete sentence out loud, and then the children read the sentence silently and checked the word-picture alternative that best completed the sentence. Three practice items preceded the 33 test items. Selection of trials with congruous und neutrul sentences. The procedure for selecting the congruous and neutral sentences for further study had two steps: (a) calculating, for each sentence, the percentage of completions with each of the two alternative words; and (b) selecting the sentences with relatively optimal response distributions. For the 1I neutral sentences, the optimal distribution was a 50%/50% split. This distribution would verify that contextual cues were absent from the sentence; furthermore, this distribution would indicate that experience and word familiarity did not create a consistent bias for one of the words in the pair. Thus, if the sentence was presented in an acoustical mode, performance should depend solely upon the acoustical cues of the test word. For the 22 congruous sentences, the optimal distribution was 100%/O% in favor of the more appropriate alternative (e.g., “The little dog has a tail (vs. pail).“) This distribution would indicate that context governed selection of the test word. Three of the four neutral sentences that were retained for further study had distributions that were closer to 50%/50% than did the eight other sentences. The fourth neutral sentence that was retained was randomly selected from among three sentences with distributions of 64%/36%. The distributions, in percentages, for the retained sentences are listed below: 1. 2. 3. 4.
The (pailltail) was grey. Please give him the (map/match). Look at the (goat/boat) over there. The (can/pan) goes on the shelf.
53147 50150 5914 I 64136
The distributions for the seven sentences that were discarded ranged from the aforementioned 64%/36% to 89%/l 1%. (The more frequent response is listed first when the distribution is not 50%/50%). The eight congruous sentences that were retained belonged to the same three-sentence triads as did the retained neutral sentences. These four pairs of congruous sentences are shown in Table I. For the seven discarded pairs of congruous sentences, the distributions ranged from 98%/2% to 100%/O% (the
WORD
Normal-Hearing
MEANINGS
Children’s
AND IDENTIFICATION
TABLE 1 Responses
to Incomplete
Pairs of incomplete sentences with possible responses in ( ) and the more appropriate one in italics
367
Sentences
Percentage of More and Less Appropriate Responses (N=44) more Appropriate
la. lb. 2a. 2b. 3a. 3b. 4a. 4b.
The little The little The cook The cook We need We need You find You find
girl has a (pail/tail). dog has a (rail/pail). took the coffee from the (can/pan). took the chicken from the (pan/can). a (march/map) for the fire. a (map/match) for the trip. (goats/boats) on the farm. (boats/goats) on the sea.
82 98 98 94 100 100 100 100
Less Appropriate 18 2 2 6 0 0 0 0
more frequent response is listed first). Thus for both the retained and discarded congruous sentences there was virtually uniform selection of one alternative even though the less frequent selection sometimes made sense. (e.g. “The cook took more frequent the chicken from the can,” less frequent selection; vs. “pan,” selection.) Furthermore, since the distributions for the retained and discarded congruous sentences were similar, selection of the retained sentences was actually based upon the neutral ones. Note that the selection of sentences for further study was based upon data summarizing normal-hearing children’s responses. This selection procedure identified sentences with contextual cues that the normal-hearing children recognized. Thus, it was subsequently possible to evaluate hearing-impaired children’s responses to contextual cues generally recognized by normal-hearing children. Word-Discrimination
Training and Sentence
Testing
Subjects. Subjects were 26 hearing-impaired children 7 to 12 yr of age (mean 9.9 yr) enrolled in programs for the hearing-impaired at local public schools. The severity of hearing loss as measured by the average of the pure-tone thresholds for the SOO-2000-Hz range in the better ear, varied from 60 to 103 dB, with a mean of 79.8 dB. All had measurable hearing in the better ear at 500, 1000, and 2000 Hz. Word-discrimination truining. During both periods in this phase of the
368
MICHAEL
STINSON
experiment, hearing-impaired children first learned to make discriminations for a set of biasing words presented in isolation. After discrimination training, the children were tested with sentences in which the same biasing words were expected to serve as context. Materials for word-discrimination training consisted of the following: 1. Two word lists were recorded on tapes by a male talker. Each had two practice words, followed by 12 words to be used subsequently in congruous or incongruous sentences. For List I, the training words were selected from the pairs chicken/coffee and girl/dog, with each word being presented three times in the list in randomized order. For List 2, the training words were selected from the pairs fire/trip and farm/sea. 2. For each list, a separate sheet showed the words in type and corresponding pictures. 3. There were five response cards per list. Each card had four pictures: two depicting words on the list and two depicting foil words. For example, one of the cards had pictures of “dog,” “girl,” “star,” and “broom,” with the words from the list being “girl” and “dog.” All testing was conducted individually and face-to-face in a quiet room. Instructions were given in the communication mode that was ordinarily used in the classroom (either speech or speech with signs). First, children were shown the sheet for a randomly selected list and asked to say or sign each word as they pointed to the corresponding picture. All children did this task correctly. Second, children were instructed to listen, using headsets, to words from the 12-item list and to point to the picture for the word on the response card. Whenever children were correct, the trainer said “right.” The criterion for completion of training was 12 consecutive correct discriminations. Twenty-one of the 26 children achieved criterion by the fifth presentation of the 12.item list. Five children who did not meet criterion by the fifth presentation were dropped from the study. Testing with sentences. Testing with sentences was the most important part of the experiment since it permitted observation of effects due to the manipulation of context. The congruous and neutral sentences that were identified on the basis of the normative data were modified by inserting the test word in the sentence. In addition, a matching set of incongruous sentences was generated by substituting the alternativt response as the test word (e.g., inserting pi1 for tail, to make. “The little dog had a pail”). The criterion for designating a sentence as congruous or incongruous was the normative data that showed near-unanimous agreement that one alternative was more congruent with the context. The same test words were used in congruous, incongruous and neutral sentences in order to control for the effect of variation in test words. The biasing and test words were ones that the children were likely to know. All words were among the most frequent 4500 in Carroll et al. ( 197 I). Words of
WORD
MEANINGS
AND
IDENTIFICATION
369
this frequency level are commonly found in text materials written for normalhearing, third-grade children. In addition, teachers had indicated that, in general, hearing-impaired children in the 7- 12-yr age range would be familiar with the words. The experiment was designed so that children did not hear the same sentence in each of the two periods. In one period, they heard four of the test words in congruous sentences and the other four words in neutral sentences. In the other period, they heard the latter four words in incongruous sentences, and the former four in neutral ones. Presentation of congruous or incongruous sentences was counterbalanced across periods. Eight lists were prepared for recording in order to achieve the counterbalancing and to vary the presentation order of test items. Each of these lists had 32 sentences, broken down into: (a) two practice sentences, (b) six foil sentences, and (c) 24 test sentences. Each of the eight test sentences (four neutral, four congruous, or four incongruous) was used three times in a list. The foil sentences were inserted in order to provide some relatively easy items and to increase the interval between repetitions of a test item. For each sentence, there was a card that depicted the test word and a similar sounding word (e.g., pictures ofpail and tail). Instructions were given face-to-face. Children were told that some of the sentences might be hard for them to understand. In addition, before receiving the incongruous sentences, they were told that some of the sentences would be funny and would not make sense. Immediately after hearing each sentence, children saw one of the cards depicting a test word and the other word in the pair. Selection of the picture for the word in the just-presented sentence constituted the measure of performance. Sentence-presentation rate was paced by the subject. In each period, children heard three repetitions of one 32-item list and three repetitions of a second list that contained the same items arranged in a different order. Results The neutral sentences were divided into the set that accompanied the congruous sentences and the set that accompanied the incongruous ones. Table 2 shows the mean percent correct for congruous, incongruous, and neutral sentences (with the data displayed separately for the two sets of neutral ones). These data are the proportion of correct selections of the test word for three pairs of list-presentation trials. Since in responding to a sentence the child always decided between two alternatives, chance performance level was 0.50. These data were entered into a repeated-measures analysis of variance comprised of three within-subject factors and one between-subject factor. These were, respectively, (a) congruous/incongruous vs. neutral sentences, (b) congruous and
370
MICHAEL
TABLE Mean Percent Correct for Congruous.
STINSON
2 Incongruous
and Neutral
Sentences
Biasing Sentences
____
Pairs of Trials 3-4
l-2
5-6
M
SlI
,&I
SD
M
SD
0.83
0. II
0.80
0. I6
0.w
0.16
0.35
0.20
0.33
0.2
0.33
0.24
Neutral
I
Sentences
Set ac~rnpan~~l bq congruous
0.65
0.14
0.65
0. 14
0.63
0.16
Set accompanied incongruous
0.63
0. I6
0.65
0.17
0.63
0. 16
bq
accompanying sentences, periods,
neutral sentences vs. incongruous (c) pairs of trails. and (d) presentation
It was predicted when
the sentences
and accompanying order of the sentence
neutral testing
that children would select the correct word more frequently were congruous than when they were incongruous. It was
also predicted that children would not differ in the frequency selected the correct word when the two sets of sentences were
in which they neutral. Given
these predictions. the analysis of variance was expected to yield a.significant interaction of the two relevant factors (congruous/incongruous vs. neutral and congruous/neutral vs. incongruous/neutral sentences). In fact, this was the outcome. F( 1.19) = 49.89. p < 0.001. Followup analyses with Dunn’s multiple-comparison more frequently
procedure showed that children selected when the sentences were congruous than
the correct when they
word were
incongruous (p < 0.01). Furthermore. there was not a significant difference in performance when the two sets were neutral. Further analyses with Dunn’s procedure showed that (a) children made more correct selections when the sentences were neutral than when they were incongruous (p < 0.01). and (b) that children made significantly more correct selections in the congruous sentences than in the neutral ones (p < 0.01). These results had also been predicted. The analysis of variance yielded only one other significant effect: Children selected the correct word more frequently when the sentences were the congruous and accompanying neutral ones than when they were the incongruous and accompanying neutral ones, F( 1,19) = 44.04, 17 < 0.01).
WORD MEANINGS
AND IDENTIFICATION
371
Discussion The principal findings were (a) that the hearing-impaired children made more correct selections of the test word in the congruous sentences than in the incongruous ones, (b) that they made more correct selections in the congruous sentences than in the neutral ones, and (c) that they made fewer correct selections in the incongruous sentences than in the neutral ones. These findings suggest that, in general, children extracted meaning from the biasing context and used this information to select the test word. In hearing a sentence such as, “The little dog has a tail, ” children may have registered all the words but the test one. For this word they could not tell acoustically whether they had heard a ip/ or a /t/. However, they could compensate for this confusion by using semantic knowledge that indicated that tail was the appropriate word. On the other hand, poor performance on incongruous sentences suggests that, in this case, contextual information overrode the acoustic cues. Of course, if the test alternatives had not sounded so similar, context might not have had as strong an effect. Intermediate performance when the sentences were neutral is also consistent with the interpretation that relations between word meanings governed selection of the test word. These sentences did not have word-meaning information that could either facilitate or interfere with selection of the correct test word. An assumption of this study was that the biasing words that served as cues for the normal-hearing children who responded to the paper-and-pencil instrument also served as cues for the hearing-impaired children who responded to the auditory presentation. This assumption was necessary because the sentences used to examine the effects of context upon hearing-impaired children were identified on the basis of the normative data provided by normal-hearing children. This assumption does not imply, however, that the two groups of children have had identical experiences with language. The discrimination training for the biasing words may have played an essential role in enabling children to use the contextual cues. For the congruous sentences, the discrimination training may have increased the extent children registered the biasing words in subsequent sentence testing; consequently, more cues were available for inferring the correct test word. On the other hand, for the incongruous sentences, the discrimination training may have made more cues available that were misleading. Consequently, the listener was more likely to ignore the acoustic cues in the test word. Results from a pilot study suggest that the word-discrimination training played an essential role in using contextual cues (Stinson, 1976). This pilot study did not include discrimination training, and the results indicated that children were not able to use contextual cues. These results suggest that it may sometimes be desirable to use discrimination training with words in conjunction with auditory training with sentences. The
312
MICHAEL
STINSON
first step in instruction for the use of contextual cues might consist of listening to sentences that include the same biasing words. In this procedure it should be relatively easy for the listener to register the biasing words and use them as contextual cues. Of course, the inclusion of discrimination training in the present study limits the generality of the findings. In normal conversation hearing-impaired children are not likely to have the benefit of discrimination training for the clues that might help them understand biasing words in spoken sentences. Additional research is needed to determine whether hearing-impaired children can generalize use of contextual cues to situations where no discrimination training is provided and where the sentences approximate those in normal conversation. The findings of the present study are consistent with the proposition that sentence comprehension involves an intricate interplay between linguistic and perceptual processes. While there seem to be many techniques available for diagnosing deficiencies in the auditory-perceptual process, there are few techniques for diagnosing and remediating the skills that link perceptual and linguistic processes. Further description and understanding of these “linking” skills is needed since they are crucial in sentence comprehension, and consequently, in communication. This reseurch was supported by NINCDS post-doctoral fellowship NS05 165 und was conducted in the Depurtment of Otolaryngology, UniversiQ of California, Sun Francisco. Prepuration of the manuscript was supported in part by the National Technical Institute for the Deaf in the course of an agreement with the U.S. Department of Health, Education und Welfare. Appreciation is extended to E. Owens fclr his though@ suggestions throughout the research and to D. Coggiola, D. Sims and J. Webster for comments on the paper. Appreciation is also extended to the Walnut Creek, Richmond and Vallejo (California) school districts for their cooperution. References Carrel,
J., Dawes,
P., and Richman, B. (1971). The Amrrrcun Heritage Word Frequenc~y Book. Heritage Publishing Co. Davis, J., and Blasdell, R. (1975). Perceptual strategies employed by normal-hearing and hearing-impaired children in the comprehension of sentences containing relative clauses. J. Speech Hmt-. RPS. 18:281-295. Erber, N., and McMahan, D. (1976). Effects of sentence context on recognition of words through lipreading by deaf children. J. Speech Hear. Res. 19:112- 119. Kalikow, D.. Stevens, K., and Elliot. L. (1977). Development of a test of speech intelligibility in noise using sentence materials with controlled word predictability. .I Acoust. Sot. Am. 61:1337-1351. Kalikow, D., Stevens. K., Gerstman. H., and Morrison, R. (1976). The speech perception in noise New York: American
WORD
MEANINGS
AND IDENTIFICATION
373
test: description and clinical data. Paper presented at the convention of the American Speech and Hearing Association. Houston, Texas, November. Owens, E., Benedict, M., and Schubert, E. (1972). Consonant phonemic errors associated with pure-tone configurations and certain kinds of hearing impairment. .I. Speech Hear. Rex. 15308-322. Stinson, M. (1976). Audiology and Psycholinguistics. Progress Report to NINCDS. San Francisco, California: Department of Otolaryngology, University of California, San Francisco. Wilcox, J., and Tobin, H. (1974). Linguistic performance of hard of hearing and normal hearing children. .I. Speech Hear. Res. 17:286-293.
OF COMMUNICATION
DISORDERS
15 (1982), 363-373
363
RELATIONS BETWEEN WORD MEANINGS AND IDENTIFICATION OF WORDS IN SPOKEN SENTENCES BY HEARING-IMPAIRED CHILDREN MICHAEL
STINSON
National Technical Institute forthe Deaf, Rochester
Institute of Technology
To examine effects of contextual words upon the identification of test words in sentences, 21 hearing-impaired children, 7-12 yr of age, received word-discrimination training followed by sentence testing. Word-discrimination training was conducted to ensure that children could discriminate the contextual words in isolation. The subsequent sentence testing was the more important phase of the experiment, since it permitted observation of the use of contextual cues to facilitate the identification of the correct test word. The relationship between the contextual words and test words was either congruous, incongruous, or neutral, as predetermined from normative data. After hearing each sentence, children decided whether they had heard the test word or its alternative. The alternative word differed in either the initial or final consonant. Significantly more correct discriminations were made on congruous and neutral sentences than on incongruous sentences. These results indicated that hearing-impaired children used contextual cues to identify the correct alternative.
Introduction Hearing-impaired persons often have difficulty following speech because they cannot hear or lip-read certain speech sounds that permit the identification of particular words. It may be possible, however, for hearing-impaired persons to extract semantic and syntactic information from intelligible words and construct hypotheses about unintelligible words based on this information. Recent research with the Speech Perception in Noise (SPIN) test has, in fact, demonstrated the use of contextual cues by adults having mild or moderate hearing losses (Kalikow et al., 1977; Kalikow et al., 1976). However, the extent to which a hearing-impaired child can use contextual cues is not clear for two reasons. First, if children lose their hearing early in life, they may acquire a different linguistic base than normal-hearing children. Consequently, the children may experience difficulty using syntactic and semantic information to facilitate their understanding of difficult-to-hear speech. Second, the decrement in hearing ability may force these listeners to code each word as
Address correspondence to Michael Stinson, Department of Educational Research and Development, National Technical Institute for the Deaf, Rochester Institute of Technology, One Lomb Memorial Drive, Rochester, NY 14623.
B Elsevier Science Publishing Company, Inc., 1982 52 Vanderbilt Ave., New York, NY 10017
0021-9924/82/050363-l
1$02.75
364
MICHAEL
STINSON
they hear it because they cannot identify enough words to form a useful contextual base. The perception of spoken sentences by hearing-impaired children has received little research attention, with studies conducted by Wilcox and Tobin (1974), Davis and Blasdell (1975), and Erber and McMahon (1976) being exceptions. None of these studies, however, investigated effects of sentence context upon identification of a particular word in a sentence. Investigation of hearingimpaired childrens’ use of contextual cues when listening to sentences may have implications for aural rehabilitation. Results of the present study were expected to provide information pertaining to semantic aspects of context that may be important in programming sentences for auditory training of hearing-impaired children. No attempt was made, however, to identify materials for clinical testing. This study examined the effect of contextual words upon the identification of test words by hearing-impaired children. The following hypotheses were made about the manipulation of contextual words in auditory presentations of sentences: 1. Discriminating between two similar sounding words is more dtflicult when test words are presented in a sentence with neutral context than when the test words are presented in a sentence with a biasing word that is relatively easy to discriminate from a corresponding word in a second sentence. For example, it should be relatively easy for children to discriminate betweenpail and tail if they occur in the respective sentences: The little girl has a pail. The little dog has a tail. After hearing one of these sentences, children are asked whether they heard pail or tail. In this case, children will usually select the correct alternative because its meaning is consistent with that of the biasing word. On the other hand, it should be relatively difficult for children to discriminate betweenpail and tail if they hear one of them in the neutral context, “The was grey.” Similar sounding pairs of words, such as pail and tail, are likely to be confused by hearing-impaired children. In the experiment the alternatives for the test word always differed by one consonant and were based on consonant confusions for impaired hearers reported by Owens et al., ( 1972). 2. Children will more frequently identtfi the correct word in a pair when it is in a sentence where the relationship between the biasing and test word is congruous than in a sentence where the relationship is incongruous (e.g., “The little girl has a tail”). If listeners perform better on sentences with a relatively congruous relationship between the biasing and the test word than on sentences with an incongruous relationship, this pattern would suggest that biasing words are being related to the test word. Presumably, listeners would
WORD MEANINGS
365
AND IDENTIFICATION
do worse on the incongruous sentences because the biasing word would induce the listeners to ignore the accoustical information in the test word. 3. Children will less frequently identify the test word correctly when it is in a sentence with an incongruous relationship between the biasing and the test word than when it is in a sentence with neutral context. In this study, hearing-impaired children were given auditory-discrimination training with the contextual words prior to participating in the test conditions. The discrimination training was conducted to ensure that the children could discriminate the biasing words in isolation. If children could discriminate the biasing words in isolation, it was more reasonable to assume (a) that the children could identify the biasing words in sentences and (b) that varying the words in sentences would produce semantic manipulations. Method Overview There were two phases in the experiment. In the first phase normative data were collected from normal-hearing children. In the second, more important phase word-discrimination training and sentence testing was conducted with hearing-impaired children. The second phase was divided into two periods. In period 1, sentences were presented that contained a congruous relationship between the biasing and test words. In the period 2, sentences were presented that contained an incongruous relationship between the words. In addition, sentences having a neutral context were presented in both periods. All sentences in both periods were presented through headsets. Collection Sentences
of Normative
Data for Congruous
and Neutral
Forty-four normal-hearing third and fourth graders between the ages of 9 and 11 yr (mean 9.8 yr) were group-tested in their classrooms in order to identify (a) congruous sentences, in which the test word was predictable from the context, and (b) neutral sentences, in which the test word was not predictable from the context. Materials. The paper-and-pencil instrument consisted of 11 sentence triads such as: Incomplete
Sentence
Sentence
Type
Responses
1. The little girl has a ~.
Congruous
pailltail
2. The little dog has a ~.
Congruous
pailltail
Neutral
pailltail
3. The
was grey.
366
MICHAEL
STINSON
For each of the 11 triads, there were two alternative responses. The congruous sentences were written with the expectation that there would be congruity between the context and one of the alternatives. (They are called “congruous” in order to contrast them with the “incongruous” sentences which were used in the experiment proper.) The neutral sentences were written with the expectation that contextual cues would be absent. These sentences were randomly distributed and printed in a booklet with two sentences per page. For each of these incomplete sentences, the two alternative words for completing the sentence (e.g., pail/tail) were printed below the sentence, with a picture for each alternative. The experimenter read each incomplete sentence out loud, and then the children read the sentence silently and checked the word-picture alternative that best completed the sentence. Three practice items preceded the 33 test items. Selection of trials with congruous und neutrul sentences. The procedure for selecting the congruous and neutral sentences for further study had two steps: (a) calculating, for each sentence, the percentage of completions with each of the two alternative words; and (b) selecting the sentences with relatively optimal response distributions. For the 1I neutral sentences, the optimal distribution was a 50%/50% split. This distribution would verify that contextual cues were absent from the sentence; furthermore, this distribution would indicate that experience and word familiarity did not create a consistent bias for one of the words in the pair. Thus, if the sentence was presented in an acoustical mode, performance should depend solely upon the acoustical cues of the test word. For the 22 congruous sentences, the optimal distribution was 100%/O% in favor of the more appropriate alternative (e.g., “The little dog has a tail (vs. pail).“) This distribution would indicate that context governed selection of the test word. Three of the four neutral sentences that were retained for further study had distributions that were closer to 50%/50% than did the eight other sentences. The fourth neutral sentence that was retained was randomly selected from among three sentences with distributions of 64%/36%. The distributions, in percentages, for the retained sentences are listed below: 1. 2. 3. 4.
The (pailltail) was grey. Please give him the (map/match). Look at the (goat/boat) over there. The (can/pan) goes on the shelf.
53147 50150 5914 I 64136
The distributions for the seven sentences that were discarded ranged from the aforementioned 64%/36% to 89%/l 1%. (The more frequent response is listed first when the distribution is not 50%/50%). The eight congruous sentences that were retained belonged to the same three-sentence triads as did the retained neutral sentences. These four pairs of congruous sentences are shown in Table I. For the seven discarded pairs of congruous sentences, the distributions ranged from 98%/2% to 100%/O% (the
WORD
Normal-Hearing
MEANINGS
Children’s
AND IDENTIFICATION
TABLE 1 Responses
to Incomplete
Pairs of incomplete sentences with possible responses in ( ) and the more appropriate one in italics
367
Sentences
Percentage of More and Less Appropriate Responses (N=44) more Appropriate
la. lb. 2a. 2b. 3a. 3b. 4a. 4b.
The little The little The cook The cook We need We need You find You find
girl has a (pail/tail). dog has a (rail/pail). took the coffee from the (can/pan). took the chicken from the (pan/can). a (march/map) for the fire. a (map/match) for the trip. (goats/boats) on the farm. (boats/goats) on the sea.
82 98 98 94 100 100 100 100
Less Appropriate 18 2 2 6 0 0 0 0
more frequent response is listed first). Thus for both the retained and discarded congruous sentences there was virtually uniform selection of one alternative even though the less frequent selection sometimes made sense. (e.g. “The cook took more frequent the chicken from the can,” less frequent selection; vs. “pan,” selection.) Furthermore, since the distributions for the retained and discarded congruous sentences were similar, selection of the retained sentences was actually based upon the neutral ones. Note that the selection of sentences for further study was based upon data summarizing normal-hearing children’s responses. This selection procedure identified sentences with contextual cues that the normal-hearing children recognized. Thus, it was subsequently possible to evaluate hearing-impaired children’s responses to contextual cues generally recognized by normal-hearing children. Word-Discrimination
Training and Sentence
Testing
Subjects. Subjects were 26 hearing-impaired children 7 to 12 yr of age (mean 9.9 yr) enrolled in programs for the hearing-impaired at local public schools. The severity of hearing loss as measured by the average of the pure-tone thresholds for the SOO-2000-Hz range in the better ear, varied from 60 to 103 dB, with a mean of 79.8 dB. All had measurable hearing in the better ear at 500, 1000, and 2000 Hz. Word-discrimination truining. During both periods in this phase of the
368
MICHAEL
STINSON
experiment, hearing-impaired children first learned to make discriminations for a set of biasing words presented in isolation. After discrimination training, the children were tested with sentences in which the same biasing words were expected to serve as context. Materials for word-discrimination training consisted of the following: 1. Two word lists were recorded on tapes by a male talker. Each had two practice words, followed by 12 words to be used subsequently in congruous or incongruous sentences. For List I, the training words were selected from the pairs chicken/coffee and girl/dog, with each word being presented three times in the list in randomized order. For List 2, the training words were selected from the pairs fire/trip and farm/sea. 2. For each list, a separate sheet showed the words in type and corresponding pictures. 3. There were five response cards per list. Each card had four pictures: two depicting words on the list and two depicting foil words. For example, one of the cards had pictures of “dog,” “girl,” “star,” and “broom,” with the words from the list being “girl” and “dog.” All testing was conducted individually and face-to-face in a quiet room. Instructions were given in the communication mode that was ordinarily used in the classroom (either speech or speech with signs). First, children were shown the sheet for a randomly selected list and asked to say or sign each word as they pointed to the corresponding picture. All children did this task correctly. Second, children were instructed to listen, using headsets, to words from the 12-item list and to point to the picture for the word on the response card. Whenever children were correct, the trainer said “right.” The criterion for completion of training was 12 consecutive correct discriminations. Twenty-one of the 26 children achieved criterion by the fifth presentation of the 12.item list. Five children who did not meet criterion by the fifth presentation were dropped from the study. Testing with sentences. Testing with sentences was the most important part of the experiment since it permitted observation of effects due to the manipulation of context. The congruous and neutral sentences that were identified on the basis of the normative data were modified by inserting the test word in the sentence. In addition, a matching set of incongruous sentences was generated by substituting the alternativt response as the test word (e.g., inserting pi1 for tail, to make. “The little dog had a pail”). The criterion for designating a sentence as congruous or incongruous was the normative data that showed near-unanimous agreement that one alternative was more congruent with the context. The same test words were used in congruous, incongruous and neutral sentences in order to control for the effect of variation in test words. The biasing and test words were ones that the children were likely to know. All words were among the most frequent 4500 in Carroll et al. ( 197 I). Words of
WORD
MEANINGS
AND
IDENTIFICATION
369
this frequency level are commonly found in text materials written for normalhearing, third-grade children. In addition, teachers had indicated that, in general, hearing-impaired children in the 7- 12-yr age range would be familiar with the words. The experiment was designed so that children did not hear the same sentence in each of the two periods. In one period, they heard four of the test words in congruous sentences and the other four words in neutral sentences. In the other period, they heard the latter four words in incongruous sentences, and the former four in neutral ones. Presentation of congruous or incongruous sentences was counterbalanced across periods. Eight lists were prepared for recording in order to achieve the counterbalancing and to vary the presentation order of test items. Each of these lists had 32 sentences, broken down into: (a) two practice sentences, (b) six foil sentences, and (c) 24 test sentences. Each of the eight test sentences (four neutral, four congruous, or four incongruous) was used three times in a list. The foil sentences were inserted in order to provide some relatively easy items and to increase the interval between repetitions of a test item. For each sentence, there was a card that depicted the test word and a similar sounding word (e.g., pictures ofpail and tail). Instructions were given face-to-face. Children were told that some of the sentences might be hard for them to understand. In addition, before receiving the incongruous sentences, they were told that some of the sentences would be funny and would not make sense. Immediately after hearing each sentence, children saw one of the cards depicting a test word and the other word in the pair. Selection of the picture for the word in the just-presented sentence constituted the measure of performance. Sentence-presentation rate was paced by the subject. In each period, children heard three repetitions of one 32-item list and three repetitions of a second list that contained the same items arranged in a different order. Results The neutral sentences were divided into the set that accompanied the congruous sentences and the set that accompanied the incongruous ones. Table 2 shows the mean percent correct for congruous, incongruous, and neutral sentences (with the data displayed separately for the two sets of neutral ones). These data are the proportion of correct selections of the test word for three pairs of list-presentation trials. Since in responding to a sentence the child always decided between two alternatives, chance performance level was 0.50. These data were entered into a repeated-measures analysis of variance comprised of three within-subject factors and one between-subject factor. These were, respectively, (a) congruous/incongruous vs. neutral sentences, (b) congruous and
370
MICHAEL
TABLE Mean Percent Correct for Congruous.
STINSON
2 Incongruous
and Neutral
Sentences
Biasing Sentences
____
Pairs of Trials 3-4
l-2
5-6
M
SlI
,&I
SD
M
SD
0.83
0. II
0.80
0. I6
0.w
0.16
0.35
0.20
0.33
0.2
0.33
0.24
Neutral
I
Sentences
Set ac~rnpan~~l bq congruous
0.65
0.14
0.65
0. 14
0.63
0.16
Set accompanied incongruous
0.63
0. I6
0.65
0.17
0.63
0. 16
bq
accompanying sentences, periods,
neutral sentences vs. incongruous (c) pairs of trails. and (d) presentation
It was predicted when
the sentences
and accompanying order of the sentence
neutral testing
that children would select the correct word more frequently were congruous than when they were incongruous. It was
also predicted that children would not differ in the frequency selected the correct word when the two sets of sentences were
in which they neutral. Given
these predictions. the analysis of variance was expected to yield a.significant interaction of the two relevant factors (congruous/incongruous vs. neutral and congruous/neutral vs. incongruous/neutral sentences). In fact, this was the outcome. F( 1.19) = 49.89. p < 0.001. Followup analyses with Dunn’s multiple-comparison more frequently
procedure showed that children selected when the sentences were congruous than
the correct when they
word were
incongruous (p < 0.01). Furthermore. there was not a significant difference in performance when the two sets were neutral. Further analyses with Dunn’s procedure showed that (a) children made more correct selections when the sentences were neutral than when they were incongruous (p < 0.01). and (b) that children made significantly more correct selections in the congruous sentences than in the neutral ones (p < 0.01). These results had also been predicted. The analysis of variance yielded only one other significant effect: Children selected the correct word more frequently when the sentences were the congruous and accompanying neutral ones than when they were the incongruous and accompanying neutral ones, F( 1,19) = 44.04, 17 < 0.01).
WORD MEANINGS
AND IDENTIFICATION
371
Discussion The principal findings were (a) that the hearing-impaired children made more correct selections of the test word in the congruous sentences than in the incongruous ones, (b) that they made more correct selections in the congruous sentences than in the neutral ones, and (c) that they made fewer correct selections in the incongruous sentences than in the neutral ones. These findings suggest that, in general, children extracted meaning from the biasing context and used this information to select the test word. In hearing a sentence such as, “The little dog has a tail, ” children may have registered all the words but the test one. For this word they could not tell acoustically whether they had heard a ip/ or a /t/. However, they could compensate for this confusion by using semantic knowledge that indicated that tail was the appropriate word. On the other hand, poor performance on incongruous sentences suggests that, in this case, contextual information overrode the acoustic cues. Of course, if the test alternatives had not sounded so similar, context might not have had as strong an effect. Intermediate performance when the sentences were neutral is also consistent with the interpretation that relations between word meanings governed selection of the test word. These sentences did not have word-meaning information that could either facilitate or interfere with selection of the correct test word. An assumption of this study was that the biasing words that served as cues for the normal-hearing children who responded to the paper-and-pencil instrument also served as cues for the hearing-impaired children who responded to the auditory presentation. This assumption was necessary because the sentences used to examine the effects of context upon hearing-impaired children were identified on the basis of the normative data provided by normal-hearing children. This assumption does not imply, however, that the two groups of children have had identical experiences with language. The discrimination training for the biasing words may have played an essential role in enabling children to use the contextual cues. For the congruous sentences, the discrimination training may have increased the extent children registered the biasing words in subsequent sentence testing; consequently, more cues were available for inferring the correct test word. On the other hand, for the incongruous sentences, the discrimination training may have made more cues available that were misleading. Consequently, the listener was more likely to ignore the acoustic cues in the test word. Results from a pilot study suggest that the word-discrimination training played an essential role in using contextual cues (Stinson, 1976). This pilot study did not include discrimination training, and the results indicated that children were not able to use contextual cues. These results suggest that it may sometimes be desirable to use discrimination training with words in conjunction with auditory training with sentences. The
312
MICHAEL
STINSON
first step in instruction for the use of contextual cues might consist of listening to sentences that include the same biasing words. In this procedure it should be relatively easy for the listener to register the biasing words and use them as contextual cues. Of course, the inclusion of discrimination training in the present study limits the generality of the findings. In normal conversation hearing-impaired children are not likely to have the benefit of discrimination training for the clues that might help them understand biasing words in spoken sentences. Additional research is needed to determine whether hearing-impaired children can generalize use of contextual cues to situations where no discrimination training is provided and where the sentences approximate those in normal conversation. The findings of the present study are consistent with the proposition that sentence comprehension involves an intricate interplay between linguistic and perceptual processes. While there seem to be many techniques available for diagnosing deficiencies in the auditory-perceptual process, there are few techniques for diagnosing and remediating the skills that link perceptual and linguistic processes. Further description and understanding of these “linking” skills is needed since they are crucial in sentence comprehension, and consequently, in communication. This reseurch was supported by NINCDS post-doctoral fellowship NS05 165 und was conducted in the Depurtment of Otolaryngology, UniversiQ of California, Sun Francisco. Prepuration of the manuscript was supported in part by the National Technical Institute for the Deaf in the course of an agreement with the U.S. Department of Health, Education und Welfare. Appreciation is extended to E. Owens fclr his though@ suggestions throughout the research and to D. Coggiola, D. Sims and J. Webster for comments on the paper. Appreciation is also extended to the Walnut Creek, Richmond and Vallejo (California) school districts for their cooperution. References Carrel,
J., Dawes,
P., and Richman, B. (1971). The Amrrrcun Heritage Word Frequenc~y Book. Heritage Publishing Co. Davis, J., and Blasdell, R. (1975). Perceptual strategies employed by normal-hearing and hearing-impaired children in the comprehension of sentences containing relative clauses. J. Speech Hmt-. RPS. 18:281-295. Erber, N., and McMahan, D. (1976). Effects of sentence context on recognition of words through lipreading by deaf children. J. Speech Hear. Res. 19:112- 119. Kalikow, D.. Stevens, K., and Elliot. L. (1977). Development of a test of speech intelligibility in noise using sentence materials with controlled word predictability. .I Acoust. Sot. Am. 61:1337-1351. Kalikow, D., Stevens. K., Gerstman. H., and Morrison, R. (1976). The speech perception in noise New York: American
WORD
MEANINGS
AND IDENTIFICATION
373
test: description and clinical data. Paper presented at the convention of the American Speech and Hearing Association. Houston, Texas, November. Owens, E., Benedict, M., and Schubert, E. (1972). Consonant phonemic errors associated with pure-tone configurations and certain kinds of hearing impairment. .I. Speech Hear. Rex. 15308-322. Stinson, M. (1976). Audiology and Psycholinguistics. Progress Report to NINCDS. San Francisco, California: Department of Otolaryngology, University of California, San Francisco. Wilcox, J., and Tobin, H. (1974). Linguistic performance of hard of hearing and normal hearing children. .I. Speech Hear. Res. 17:286-293.