British Journal of Disorders of Communication, 24, 211-239 (1989)
© The College of Speech Therapists, London
C. Adams Centre of Audiology, Education of the Deaf and Speech Pathology, University of Manchester
and D. V. M. Bishop Department of Psychology, University of Manchester
Conversational samples were obtained from 57 children aged from 8 to 12 years with specific language impairment, and 67 control children aged from 4 to 12 years. Fourteen of the language impaired children fitted the clinical description of semantic-pragmatic disorder. It was found that exchange structure, turntaking. conversational repair and use of cohesive devices could be assessed with adequate inter-rater and test-retest reliability. Children with semantic-pragmatic disorder produced more initiations than other children. Some of these children also violated turntaking rules by interrupting the conversational partner to an unusual degree. Use of cohesion was normal for children with semantic-pragmatic disorder, but limited in other language-impaired children. Analysis of conversations may be more useful than conventional language tests for identifying linguistic abnormalities in children with semantic-pragmatic disorder.
Key words: children, language disorder, semantics/pragmatics,. discourse analysis.
The majority of language assessment procedures available to speech therapists are concerned with the child's ability to use grammar, vocabulary and phonology. In recent years there has been increasing recognition that many children who are reasonably competent in these areas may nevertheless have problems with semantics and pragmatics. Thus although their speech may be fluent and grammatically well formed, the content of what they say has an odd quality, and the way in which they use language in social interactions may be unusual. Several authors have suggested that this constitutes a specific subtype of language disorder, variously termed 'semantic-pragmatic syndrome' (Rapin & Allen, 1983), 'semantic-pragmatic disorder' Bishop & Rosenbloom, 1987) or 'conversational disability' (Conti-Ramsden & Gunn, 1986). Rapin (1987) includes the following in her characterisation of semantic-pragmatic syndrome speech fluent with adequate articulation; verbose; comprehension deficits for the meaning of verbal messages, notably questions; tendency to interpret messages quite literally; tendency to respond to one or two words in a sentence rather than to the entire message; incessant chatter, perseveration, use of or circumlocutions, semantic paraphasias and lack of semantic specificity; impairment in the ability to take turns and to maintain a topic in discourse.
One problem for the researcher in this area is that we lack suitable instruments for objectively assessing such disorders. Most procedures for analysing conversational data are designed to handle normal interactions and also tend to be geared either to very young children (e.g. Dore, 1974; Blank & Pranklin, 1980) or to specific types of conversational interaction such as classroom teaching (e.g. Coulthard, 1985).
The few available procedures suitable for analysing conversational impairments tend to be of two kinds. The first type is exemplified by the work of Prutting and Kirchner (1983, 1987), who described an analysis in which an individual's contribution to a 15-minute conversation is rated as appropriate or inappropriate on a range of dimensions of pragmatic behaviour, such as topic introduction, turntaking contingency and cohesion. This procedure has the advantage that inter-rater reliability is high for trained observers, but the disadvantage that the resulting measure is rather gross, with pragmatic impairments being rated as either present or absent, rather than there being any quantification of the degree of impairment. This sort of measure can be of value in research studies, but gives insufficient detail for therapists wishing to analyse pragmatic problems in detail. The second type of approach has the converse advantages and disadvantages and is typified by the analytical frameworks described by Damico (1985) and McTear (1985). These procedures have the potential for being used for much more fine-grained analysis, but seem more appropriate for qualitative description of single cases, rather than for measuring conversational behaviours in order to compare different groups. Their inter-rater and test-retest reliabilities have yet to be demonstrated.
The aim of this study was to establish how far it is possible to quantify conversational behaviour in a way that is reliable and valid but also sensitive enough to give a detailed characterisation of disorder for therapists working with language-impaired children.
Reliability refers to the replicability of results across different situations and different raters. This is an important issue when dealing with conversational data, where many concepts are ill defined. Two analysts may disagree if their subjective interpretation of coding categories differs. Another aspect of reliability concerns stability of the measures obtained across different situations. A child's conversational behaviour will depend on two things: the context of conversation. including whom the conversation is with, what it is about, where and when it takes place, and so on, and on the characteristics of the child, including transient factors such as mood, as well as more long-lasting features such as whether the child has a specific problem in comprehension. The main interest of the speech therapist will usually be in evaluating the more stable characteristics of the child. It is therefore necessary to demonstrate that, if similar procedures are followed for obtaining conversational samples, the same child will obtain a similar profile of conversational ability from one occasion to another.
It is important to demonstrate that a procedure is reliable, but this does not ensure that it is valid. Validity refers to the extent to which an assessment measures what it purports to measure. It is possible to imagine an assessment procedure that is reliable but not at all valid: for example, if we were to propose an index of 'poor conversation' in which points were awarded for full sentences but subtracted for elliptical ones, we may obtain a measure that was stable across different raters and different occasions, but that was quite meaningless, because it would end up classifying normal children as conversationally impaired. There are no hard and fast criteria for establishing the validity of an assessment procedure, but in general it is important to demonstrate that it does discriminate language-impaired from normally developing children in a meaningful way. In addition, we would usually expect a language assessment to show systematic developmental changes in normal children.
Since we aimed to find a means of objectively characterising language problems of children with 'semantic-pragmatic disorder', a further test of the validity of our conversational analysis is the extent to which it discriminates such children from other language-impaired children. A major problem is that we have no objective criteria for classifying language-impaired children into subtypes, and we must rely on subjective impressions of how far a child fits a clinical description when using terms such as 'semantic-pragmatic disorder'. There is an inevitable circularity in this exercise: first we classify children according to our informal impressions of their conversational behaviour, then we devise a procedure expressly to assess those aspects of behaviour thought to characterise this subgroup and see whether this is sensitive to their problems. It could be argued that a positive result would not be particularly surprising. Nevertheless, this is a necessary first step towards devising more rigorous and objective criteria for classification. If we find that our measures do relate to more informal classifications, this would provide some justification for using them as categorisation criteria in future studies.
In this study two complementary approaches were adopted. The first, described in this paper, involved applying procedures developed for the analysis of normal conversational behaviour to children with language impairment. The second approach, described in a companion paper (see p.241), worked from the opposite end by identifying abnormal conversational behaviours in the language of language-impaired children, and then attempting to categorise these.
Many researchers describe analysis of conversational data from clinical groups (e.g. Prutting & Kirchner, 1987), but there is rarely any information about what was talked about. It seemed inadvisable to rely solely on data from 'natural' conversations between a child and an adult stranger because results might depend on the topic of conversation and, if there were wide variability in this, it would be difficult to establish how far the content of the conversation (as opposed to characteristics of the individual) was contributing to variability.
For instance, certain language-impaired children may mask difficulties by keeping the conversation on familiar topics, rather than having to listen to and interact with the conversational partner. To prevent this, a semi-structured situation was used, in which conversation was developed around a set of photographs. The aim was to keep the conversation going for 5-10 minutes, which typically gave between 100 and 150 conversational turns and took around 1 hour to transcribe. Although longer samples are often recommended for linguistic analysis. this recommendation is based more on conventional wisdom than on any objective criteria, and it has the effect that many analysis procedures are far too lengthy to be of use to clinical practitioner:. It seemed preferable to establish whether reliable information could be obtained from a sample of this length rather than assuming, as others have done. that much longer samples are necessary.
The conversational analysis that was used with this material draws heavily on the framework specified by McTear (1985), focusing specifically on exchange structure, turntaking, repairs and cohesion. For each of these aspects of behaviour. children's conversations were analysed to establish:
1. How far occurrence of conversational features can be reliably rated by different individuals.
2. Whether conversational features are stable over time for a given child.
3. If there are developmental trends in normal children.
4. Whether these conversational features distinguish language-impaired children from children with normal language.
5. Whether they distinguish children with semantic-pragmatic disorders from other language-impaired children.
In order to limit the amount of control data to be collected, consideration was restricted to 8-to 12-year-old language-impaired children, an age by which most language-impaired children are capable of engaging in intelligible conversation with an adult. Control data were collected for normal children in this age range, and for younger children. This made it possible to see whether differences between language-impaired and age-matched controls corresponded to immaturities or whether they included conversational features not observed in the course of normal development.Subjects
Language-impaired children were recruited from eight schools and units specialising in the education of children with specific language impairments. plus a further three children referred by speech therapists aware of our special interest in children with semantic-pragmatic disorder. All children attending the schools and units who were in the age range 8-12 years inclusive were seen, after excluding those with physical speech disorders, those whose speech was unintelligible, children with profound comprehension problems and pupils whose nonverbal abilities, as measured by the Block Design and Picture Completion sub-tests of the WISC-R (Wechsler Intelligence Scale of Children - Revised, Wechsler, 1976), were below normal limits, defined as having a summed scaled score on these subtests below 13. A wide range of types of language problem was included.
This gave a sample of 57 language-impaired children, who were divided into two broad groups of 'semantic-pragmatic disorder' (n =14) and 'other' (n = 43), depending on how well they fitted the clinical picture described by Bishop and Rosenbloom (1987). Where school staff used the concept of semantic-pragmatic disorder, we depended upon their judgements to decide whom to include in this group. Where staff did not use this descriptive category, we judged whether children fitted this clinical picture or not on the basis of impressions formed during a series of language assessments which did not include the conversational procedure.
The aim was to see 10 control children (5 boys and 5 girls) at each of the following ages: 4, 5, 6, 8, 10 and 12 years. Some additional 4 year olds were included so that children in a nursery class did not feel left out. Control children were drawn from primary schools and nursery classes in Manchester and Liverpool. All had English as the first language spoken at home. Children receiving speech therapy were excluded.
Each conversation was developed around a set of three photographs. Two sets of three photographs were used. The first set consisted of pictures from the Winslow Press Sequences II series, showing: Al - a doctor examining a sick boy; A2 - a girl having a birthday party; A3 - a man and girl with a broken down car. The second set consisted of family snapshots and showed: BI - a group of children and adults on a roundabout in a playground; B2 - a family having a picnic; and B3 - a boy at the seaside with his grandmother.
Each picture was presented and the child was encouraged to talk about similar experiences he or she had had. It was intended to avoid as far as possible the typical 'teaching' interaction where the adult only asks questions to which he or she already knows the answer. The picture was removed once conversation on that topic had been initiated. The conversation was allowed to develop naturally, but the adult could refer to a set of prepared questions when she judged that the conversation needed moving on. There was no requirement that all of these questions be asked in the course of the conversation, and the adult was at liberty to ask other questions as appropriate. A few questions were specifically designed to tap particular problems of understanding. For example, we asked children: 'Would you say that was a good place to break down?', and 'Can you tell me about it?' anticipating that some very young or language-impaired children might interpret the first question literally as a command to repeat the words 'that was a good place to break down' and give a simple reply of 'yes' to the second (compare Leonard, Wilcox, Fulmer & Davis, 1978).
A range of standardised and experimental language assessment procedures as well as the WISC-R Performance subtests was administered. These included two standardised comprehension tests (WISC-R Verbal Comprehension and Test for Reception of Grammar) and a story-retelling procedure devised by Bishop and Adams (unpublished data) that was used to compute a mean length of utterance (MLU) in morphemes. Previous work showed that MLU measured in this context was a sensitive index for distinguishing language-impaired from normally developing children.
Each language-impaired and control child was tape-recorded in conversation (using photograph set A) with one of the authors, the tape being transcribed by the adult who had participated in the conversation.
In a second phase of the study, speech therapists at a residential school for children with specific language impairments collected conversations with 20 children to provide data for assessing the test-retest reliability of the analysis procedure. Fourteen of the children in this second phase of the study had been seen by one of the authors approximately 9 months previously for the first phase. Each child in phase 2 had two conversations with a speech therapist who was not regularly working with that child: one conversation was based on the original set of topics (photograph set A), and the other on the new set (photograph set B). The conversations occurred on different days within 2 weeks of one another. The order of the two sets of topics was balanced. The tape-recorded conversations were transcribed and analysed by one of the authors. In order to establish inter-rater reliability, each author analysed the first conversation from all children in phase 2. To establish test-retest reliability, the sample of children was divided in two, and each author analysed both conversations from children in one half.
Samples were transcribed as soon as possible after recording by one author. Limited resources prevented the assessment of inter-rater reliability of transcription but common procedures were followed as below. A future study is intended to further investigate transcription reliability. The conventions of Crystal, Fletcher and Garman (1976) were followed to mark pauses, overlaps and unintelligible segments. Intonation was not usually transcribed, unless unusual or ambiguous, but was used to make decisions about utterance boundaries in the course of transcription.
When relying on audio-tape non-verbal responses, such as nodding, cannot be recorded. However, it was seldom difficult to deduce where these behaviours had occurred, because the nature of a non-verbal turn will determine the adult's response, e.g.
adult: do you like sausages?/ (silence on tape) adult: mmm, so do I/ Compare: adult: do you like sausages?/ (5 seconds silence on tape) adult: not sure?/
It seems reasonable to assume a nod occurred in the first exchange, but not in the second. It was always assumed that no response had occurred unless the adult's turn gave good grounds for believing that the child had responded non-verbally. In the first example above, the silence on tape would be coded as '(nods)', whereas in the second, it would be transcribed as '(5 seconds)'.
Each adult (A) and child (C) turn was numbered for easier identification later on. (In this paper, turns are numbered consecutively for ease of reference.) A turn was defined as an interval filled by:
1. One or more utterances by one speaker.
2. Silence or 'erm' etc. as the other speaker waits for a contribution.
3. Non-verbal response such as nodding.
Each utterance was written on a new line, terminated by a slash. The definition of 'utterance' is problematic for any language analysis procedure, and will to some extent be arbitrary. The following criteria were used here:
i. A turn consisting of a single clause or part of clause was treated as one utterance, even if there was a gap before or during the turn.
ii. A sequence of clauses within a turn joined by connective(s) was treated as one utterance, unless the clause preceding the connective was terminated by failing intonation. A series of clauses where only the last was marked by an overt connective was treated as one utterance if it formed a sequence in a child's account of a series of events, e.g.
1 A: what happens then/ 2 C: when you come into school, we come in, we see if people here have registrar, then I think we have musical chairs/
3 C: I don't like orange juice/ I only like coca cola/
Tags and comment clauses were treated as belonging to the main clause that they were connected to and not as separate utterances.
iii. The adult sometimes interjected 'yes', 'mmm', 'did you?' or other encouragements before the child had completed an utterance, so that a clause produced by the child was spread across several turns, e.g.:
4 A: what did you eat?/ 5 C: jelly and icecream-(rise intonation) 6 A: mmmm/ (encouragingly) 7 C: -crisps-(rise intonation) 8 A: yes/ 9 C: -and sausages/ (fall)
iv. 'Yes' and 'no' were treated as full utterances, except where they were followed by a related clause which was so closely linked to the assertion/negation that the 'yes' or 'no' would sound acceptable if moved after the following clause, e.g.
10 A: do you like going to parties/ 11 C: yes, I do/ compare: 12 A: would you say that boy looks ill?/ 13 C: no/ he's just got a cold/
v. Incomplete fragments and false starts were treated with the following clause if they appeared to represent attempts to say the same thing. However, if the child appeared to abandon an utterance without completing it, and started a new one, then the incomplete part was treated as a separate utterance:
14 A: where did you go on holiday? 15 C: went to______/ 16 A: forgotten the name?/
Space does not permit more than an outline of the analytical framework and the procedures used for coding language samples. The codes that were used are shown in Table 1, and will now be briefly described.
Table 1: Codes used to classify conversational behaviour__________________________________________________________________________________________
1. Exchange structure Initiation Question/command IQ Statement IS Response Minimal, non-verbal RMn Minimal. Verbal RMv Extended RE Continuation Statement CS Follow-up F Unintelligible Un Incomplete X 2. Turntaking Gap <G> Inadvertent overlap <I> Violating overlap <V> Adult interrupt <A> 3. Repairs Appropriate response to request for clarification <R1> Inadequate response to request for clarification <R2> Child request for clarification <R3> Child self-repair <R4> 4. Cohesion (pro-forms/demonstratives) Referent recoverable from linguistic context (anaphoric/cataphoric) <L> Referent recoverable from situational conical (exophoric) <S> Unestablished referent <U>__________________________________________________________________________________________
Units of Exchange Structure
Conversation does not consist of a random sequence of utterances by two individuals: there should be a clear relationship between utterances in sequence. Linguists have attempted to identify general rules that distinguish acceptable from unacceptable sequences of utterances in a conversation, but little agreement has been reached (Stubbs, 1983). McTear (1985) adapted a categorisation of utterances used by Coulthard (1985) to analyse teacher-pupil interactions, in which the basic unit of conversational exchange is defined as an initiation by A followed by a response by B. An initiation is defined as an item which begins anew and sets up an expectation of a response; a response is predicted but itself sets up no expectations. An additional optional element of exchange structure is the follow-up, which neither elicits nor provides information, but which acts to acknowledge a response, serving merely to echo or reinforce information that has already been given.
The following passage illustrates these types of item:
17 A: have you ever been ill like that?/ I 18 C: yeah, sick/ R 19 A: what happened to you then?/ I 20 C: oh I was sick lots of times I was_____/ R had to kept me off school/ 21 A: oh dear/ F
It will be noted that one utterance in this example is uncoded: C's second utterance at turn 20. In this analysis, this would be coded as continuation (C). McTear (1985) defined a continuation as an utterance which continues or adds to a previous utterance within a turn. This definition was extended to encompass instances of continuation lasting beyond one turn, with adult follow-ups or (ignored) initiation intervening.
Two different types of initiation can be distinguished: a question (to which the response is expected to provide information) and a statement (that itself provides information, and to which the response is some kind of acknowledgement).
Questions and requests for action (Q)
22 A. my car broke down once/ IS 23 C: have you got a mini?/ IQ
In general, initiations classified as Q corresponded to the grammatical forms of question or command, but the criterion for using this code was the illocutionary force of the utterance, rather than the grammatical form. Thus the rater had to judge what the speaker intended, rather than the literal meaning of an utterance. It is, for instance, possible to ask a question using the grammatical form of a statement, e.g.
24 C: I suppose you haven't got a pencil
which has the illocutionary status of a yes/no question (i.e. Have you got a pencil?).
Requests for action (e.g. 'Guess what I did yesterday?') did not seem common enough to merit a separate category, and 80 were classified together with questions. Both types of utterances make a strong demand on the conversational partner to respond.
These are usually straightforward to recognise, consisting of utterances where one speaker provides information in contexts other than in direct response to a request for information.
25 C: I'm getting a new dress today/ IS it's going to be for my party/ CS I'm having a birthday party on Saturday/ CS
Utterances such as these make a less strong demand on the partner to respond, but nevertheless they set up the expectation that the other will make some response. rather than just starting a new initiation.
Minimal responses (RMv and RMn)
These codes were used for responses such as 'yes', 'no' or 'don't know', or the assumed non-verbal equivalents of nodding, shaking the head or shrugging, which provide no new information other than confirmation, denial or an indication of inability to respond.
Extended responses (RE)
The term 'extended' refers to the content of a response, rather than its form: an 'extended' response was coded for any response that gave more information than just a minimal yes/no/don't know reply. and it could consist of just a single word.
26 A: where did you go on holiday/ IQ 27 C: Blackpool/ RE
Where an utterance consisted of 'yes' or 'no' plus additional new information, this was coded as RE, e.g.
28 A: did you ride your bike?/ IQ 29 C: yes, yesterday/ RE
30 A: is your sister bigger than you?/ IQ 31 C: yes, she is/ RMv(although the child literally says more than 'yes' the utterance contains no more information than confirmation).
These are frequently used by the adult, but may also be found in the child's utterances: sometimes there is a sequence of two follow-ups:
32 A: what did you see there?/ IQ 33 C: there was this massive great tank/ RE 34 A: a big one/ F 35 C: yeah/ F
Sometimes a child failed to respond at all to an initiation, i.e. there was a blank interval while the adult waited for a response, after which the adult tried again to elicit a response. Such 'empty turns' were left uncoded. However, if the child showed signs of trying to respond, for instance by saying 'er....', then this was coded as an incomplete utterance.
Unintelligible (Un) and Incomplete (X) utterances
No attempt was made to analyse utterances further if they were so unintelligible as to make interpretation dubious or impossible. These were simply coded Un. However, presence of an unintelligible segment of an utterance did not always preclude analysis.
Incomplete utterances (i.e. false starts, instances where the speaker forgets what he or she is saying or has word-finding problems so cannot continue, and so on) were coded X (whether they appeared to be responses or initiations) and were not analysed further.
Turntaking was coded in a second pass through the transcript. In contrast to the first pass, where every utterance was coded for exchange structure, turntaking codes were only noted against utterances where gaps or turntaking violations occurred.
Fluent, rapid conversation requires both partners to synchronise the timing of their utterances precisely, predicting accurately when the other will stop talking. This is a skill that develops as children grow older (see McTear, 1985, Chapter 7, for a review). A gap <G> was coded when there was a noticeable interval between the completion of the adult's utterance and the start of the child's: the interval is arbitrarily fixed at --- or longer from the transcript (following transcription conventions of Crystal et al. (1976) who use - to indicate one pulse of the speaker's rhythm), e.g.
36 A: where did you go on holiday?/ 37 C: ----Cornwall/ <G>
A gap was not coded when there was a pause in the middle of an utterance as the child groped for a word: this is indicative of a different sort of problem, i.e. word finding rather than synchronisation with the adult. Even an 'er' is sufficient to mark the start or an utterance, so that the following would not be coded as a gap:
38 A: where did you go on holiday?/ 39 C: er------Cornwall/
A gap was not coded if an interval persists for so long that the adult resumed the conversation, without the child saying anything (not even 'er ---').
Where there was overlap in the interaction this is marked by an asterisk at the start of the overlap in both speakers' utterances. Inadvertent overlap <I> was coded when the child's utterance occurred at a 'transition relevance place' (Levinson, 1983), i.e. at a point when the child had legitimate reason to predict, on syntactic or prosodic grounds, that the adult completed her turn, but she then continued with a tag question or fresh clause, e.g.
40 A: your uncle's got a car/ have you sometimes been somewhere in it?/ 41 C: (nods)/ 42 A: yes?/ where have you been?/ *did you go on holiday in it? 43 C: *MacDonalds/ <I>
If, instead of making a reasonable but wrong prediction about the point at which the adult had finished speaking, the child simply cut across in the middle of the adult's utterance, this was coded as a rule-violating overlap <V>, e.g.
44 A: tell me about a *party that you had/ 45 C: *there's a tractor out there/ <V>
Instances where the adult interrupted the child were coded as adult interrupt <A>, on the assumption that this might be an indication of the child's poor control of synchronisation rather than the adult's. For instance, the adult might assume that the child has finished when he had not, e.g.
46 A: what did you have at your party?/ 47 C: crisps, cakes, pop/ --- I *had a cake with my name on/ 48 A: *did you get any presents?/ <A>
However, no code was given for instances where an adult follow-up (e.g. mmm, really?) overlapped with a child's utterance, as this gave the impression of an entirely normal conversational behaviour of an adult encouraging a child to converse.
Four types of repair behaviour were distinguished. The first is an adequate response to a request for clarification. A request for clarification was identified where the adult appeared not to have understood part of the child's utterance, usually because of unintelligibility, but sometimes when an unusual word or idea was used and the adult required confirmation. A crucial feature of a request for clarification is that it represents a halt in the progress of the conversation while the speaker and listener confirm the preceding exchange. It is not the same as a request for further information. RI was coded where the child responded appropriately, giving the adult the information required, e.g.
49 A: where are you going?/ 50 C: Lake District/ 51 A: where?/ 52 C: Lake District/ <R1>
If, however, the child did not respond appropriately (either producing no response or failing to provide clarification), this was coded as R2, inappropriate response to request for clarification.
53 A: why do you think the doctor has come?/ 54 C: because (three syllables)/ 55 A: because what?/ 56 C: he's feeling his tummy/ <R2>
Note that coding is not always clear-cut. It was assumed that 56 was a new initiation, but a case could be made for coding it as R1, if the child was in fact repeating or elaborating a response for 54. In this instance, the analyst judged that the child was unable or unwilling to give an adequate response to this question and was moving the conversation on by commenting on the picture. Inevitably there will be a degree of subjectivity in such judgements.
Children should not only be able to recognise and respond to requests to clarification from the adult, they should also he able to produce such requests themselves. Indeed, for children with comprehension problems, this may be a useful strategy to teach them. We coded as R3 those instances where the child asked for confirmation or repetition/clarification, in an appropriate context.
57 A: he didn't look ill/ perhaps he was faking it/ 58 C: hey?/ <R3> 59 A: perhaps he was faking it/
We also included in this category those rare instances where the child explicitly corrected something the adult had said:
60 A: well, Peter, what have you been doing?/ 61 C: I'm Graham/ <R3>
The final type of repair we coded was the self-repair, R4, where the child repaired his or her utterance or modified it to make the message clearer.
62 A: that's the first one/ what's happening in that picture?/ 63 C: - the doctor erm - point erm putting that thing in the boy's erm stomach/ <R4>
In 63 he modifies 'point' to 'putting', i.e. he corrects his utterance to include a more appropriate verb.
Cohesive devices, i.e. reference and ellipsis, help to link elements from different utterances in a conversation (Halliday & Hasan, 1976). A major means of achieving cohesion is to use pronouns and demonstratives which are not interpreted semantically in their own right, but make reference to an element from another utterance for their interpretation, e.g.
64 C: My friend bought some new shoes/ When she got home she put them on/
The underlined words are all examples of anaphobic reference, i.e. they are interpreted by referring back to something that has been previously established linguistically: 'she' = my friend, and 'them' = new shoes. Halliday and Hasan (1976) draw a distinction between anaphoric and cataphoric reference, the former term applying to a pro-form referring back to something previously mentioned (as above), and the latter applying when it refers ahead to a later portion of text, as in:
65 C: Before she went home, my mum bought me some shoes/
Detailed analysis of all aspects of cohesion would be far too time consuming to include in a clinical analysis. It was therefore decided to focus on the child's use of pronouns and demonstrative terms, making it possible to use computerised search procedures to identify potential instances of cohesion. Although personal and demonstrative reference do not give an exhaustive picture of a child's mastery of cohesion, there is evidence that they do constitute a major portion of children's cohesive devices. Liles (1985), in a study of children's story-telling narratives, showed that personal and demonstrative reference accounted for around 50% of instances of cohesion in both normal and language-impaired children.
In this analysis, all instances of the following terms were first identified: he, him, she, her, it, we, us, they, them, his, her, hers, its, our, ours, their, theirs, this, these, that, those, here, there, then, one. Instances where these words were used as non-referring terms (e.g. 'that' as relative pronoun, 'one' as cardinal numeral etc.) were excluded. Each occurrence was then coded as an instance of cohesion if it was used for anaphoric or cataphoric reference, i.e. the referent for the term could not be identified within the utterance in which it occurred, but it was recoverable from another utterance in the conversation. For anaphoric reference, the original referent may have been introduced by either the adult or the child. The code <L> (referent recoverable from linguistic context) was coded against all such terms, with more than one instance being possible per utterance.
As well as noting instances of cohesive usage of personal and demonstrative pro-forms, instances of exophoric reference were coded, i.e. where a pro-form could be understood with reference to the situation in which it occurred, which were coded <S>, e.g.
(child interested in adult's tape-recorder) 66 C: put it<S> there<S>/
Finally, uses of pro-forms in which the referent was ambiguous or not recoverable (because it had not been mentioned in earlier conversation and could not be deduced from context) were coded as <U> (Unrecoverable referent).
General Characteristics of Sample
Table 2 shows background linguistic and other data for the two subgroups of language-impaired children and controls. These data confirm that the language-impaired children are well matched on age and non-verbal (Picture Completion) score to the 8- to l2-year-old control group, but are impaired on receptive and
Table 2: Characteristics of sample
______________________________________________________________________________________ Language-Impaired Control ________________________ ________________________ Semantic- Pragmatic Other 4-6 years 8-12 years (n=14) (n=52) (n=36) (n=31) ______________________________________________________________________________________ Mean age in years (s.d.) 9;7 (1;6) 9;10 (1;6) 5;5 (0:9) 10;3 (l;9) z-score: Pic. comp. -0.1 (0.89) -0.2 (0.78) 0.3 (0.98) 0.2 (0.63) Verb. Comp. -2.1 (0.67) -1.4 (0.90) -0.6* (1.08) -0.1 (0.77) TROG -1.3 (0.94) -1.2 (1.12) -0.6 (0.95) -0.4 (1.09) Raw score: Verb,. Comp. 6.0 (3.59) 9.1 (4.62) 3.9 (2.95) 16.6 (4.67) TROG 13.7 (2.79) 14.2 (2.87) 8.9 (3.55) 16.3 (2.80) MLU (story recall) 8.7 (1.46) 8.0 (1.93) 7.9 (1.73) 11.0 (2.61) PCC 97.4 (5.39) 84.4 (15.24) - - ______________________________________________________________________________________*'Could not be computed for those aged under 6 years.
expressive language tests relative to that group. Both language-impaired groups performed significantly below the level of age-matched controls on MLU and the two comprehension tests [Scheffé tests significant at 0.05 level (Scheffé), 1953)] Children coded as having semantic-pragmatic disorder did not have significant phonological problems, whereas many of the other language-impaired children had reduced phonological systems. The younger control group can be regarded as an MLU-matched control, as they did not differ significantly on this variable front either of the language-impaired groups. However, this group was not well matched with language-impaired children in terms of receptive language. Although for their age the language-impaired children were impaired on comprehension, their absolute level of performance on both tests was superior to that of the younger control children (aged 4-6 years), the difference being statistically significant for all comparisons except for that between the semantic-pragmatic disordered group and the younger controls on the raw score of the Verbal Comprehension subtest.
Each section of the analysis will now be discussed after considering in turn reliability, normal developmental trends, and differences between control and language-impaired children.
Exchange Structure Elements
Data on reliability come from phase 2 of the study. Table 3 shows correlations between the numbers of utterances in each exchange structure category for two raters independently rating the same 20 conversations. Table 3 also shows the
Table 3: Correlations obtained in phase 2 of the study
______________________________________________________________________________________ Inter-rater Test-retest Test-retest Agreement agreement over agreement over 2 weeks 9 months (n=20) (n=20) (n=14) ______________________________________________________________________________________ Utterance/turn 0.997+ 0.930+ 0.853+ Percentage all initiations 0.800+ 0.732+ 0.618* IQ 0.983+ 0.553+ 0.161 IS 0.901+ 0.740+ 0.646+ Percentage all mimimal 0.847+ 0.659+ 0.759+ RMv 0.897+ 0.630+ 0.749+ RMn 0.971+ 0.704+ -0.284 Percentage response extended 0.984+ 0.707+ 0.565* Percentage continuation 0.976+ 0.752+ 0.631+ Percentage follow-up 0.767+ 0.473* 0.090 Percentage unintelligible 0.822+ 0.611+ 0.239 ______________________________________________________________________________________*P<0.05
correlations between indices obtained for two conversations obtained on separate occasions within 2 weeks of each other (using different sets of photographs) for 20 children. Finally, Table 3 shows the correlations between indices for 14 children who had conversations with one of us during phase 1 of the study, and then again (with the same photograph context) with a speech therapist from their school about 9 months later in phase 2.
The inter-rater correlation gives an indication of how far it is possible to rate this aspect of behaviour objectively. Highly significant correlations were obtained for all indices, the lowest agreement occurring for the rare category of follow-up. Note that where lack of agreement arose over the category 'unintelligible' this was not disagreement about transaction, as both raters worked from the same transcript: rather it was disagreement about whether the unintelligibility precluded analysis of the utterance.
The test-retest correlations over a 2-week period give an indication of how far the indices obtained in one conversation reflect stable characteristics of the language-impaired child and how far they are influenced by day-to-day variations in circumstances and the particular topic of conversation (which was different for the two conversations rated here). All the categories that showed good inter-rater reliability also gave highly significant test-retest correlations over this period.
The test-retest correlations in the final column indicate how stable conversational behaviour is over a 9-month interval. Given that conversational indices change in normal children as they grow older, plus the fact that these children are receiving full-time special education for language problems, stability of all conversational indices were not be expected. Several of the major indices do not give significant correlations over this interval, but the most common types of exchange structure element, namely, minimal verbal responses, initiating statements, extended responses and continuations, all remain highly significant.
Normal developmental trends
For each subject, the number of turns and number of utterances per turn were calculated. There was a modest trend for the number of turns to decrease with age (r= -0.3, degrees of freedom (d.f.)=65, P <0.05), but no significant developmental trend for utterances per turn (r=0.14, d.f. =65, NS). Each of the elements listed in Table 1 was computed as a percentage of all child utterances plus coded non-verbal responses.
Figure 1:Percentage of exchange structure types for control and specific language-impaired children. The language-impaired children are aged 8-12 years
Figure 1 shows graphically the development of proportion or different utterance types in the control sample. Correlations were computed between proportions and age, first for the whole control sample, and then for the 8 to 12 year olds only. This made it possible to see whether there were any developmental trends for these indices and, if so, whether development has usually stabilised for the age range of the language-impaired sample. Results are shown in Table 4.
Initiating statements showed a significant decrease with age, with the trend continuing beyond 8 years of age. There was a corresponding increase in the use of extended responses and continuations, which was significant over the age range 4-6 years. These categories were strongly negatively correlated with the occurrence of initiating statements (r= -0.71 and -0.63 respectively, d.f. =65, P <0.01).
There was a slight increase in the proportion of minimal verbal responses and a sharp decrease in the proportion of non-verbal responses between 4 and 12 years of age, the two types of minimal response being negatively correlated (r= -0.67, P <0.01). When all classes of minimal response were summed together, there was an overall trend for a slight decrease in such responses with
Table 4:Correlations between proportions of utterance types and age in the control group
______________________________________________________________________________________ All children (4-12 years) 8-12 year olds only (n=67) (n=31) ______________________________________________________________________________________ Initiation Question 0.102 0.497+ Statement -0.483+ -0.413* Continuation 0.238* -0.142 Response Minimal, verbal 0.409+ -0.094 Minimal, non-verbal -0.572+ -0.239 Extended 0.253* 0.345 Follow-up -0.189 0.068 Unintelligible -0.334+ 0.123 Incomplete -0.063 -0.067 ______________________________________________________________________________________*Significant at 0.05 level.
age, which just reached statistical significance when the whole control sample was included, but which showed no relationship with age when only those aged 8 years and over were included.
There was no developmental trend for follow-ups, which were rare in these conversations. The percentage of unintelligible utterances declined significantly with age when the whole age range was considered, but there was no significant age relationship within the 8 to 12 year olds.
Comparison of language-impaired and control children
Data from language-impaired children are shown in Figure 1. These data were analysed with two questions in mind: first, which measures discriminate between language-impaired and normal children; secondly, do the indices distinguish between children with the clinical picture of semantic-pragmatic disorder and other language-impaired children?
For this analysis, control children were pooled into two broad groups: a control group (C) of the same age range as the language-impaired children (8-12 years) and a younger group (Y) of 4 to 6 year olds. Language-impaired children were divided into two subgroups: 14 classified as 'semantic-pragmatic disorder' (S) and 43 'other language-impaired' (O). Each of the proportions described above was transformed logarithmically to stabilise variances. An exploratory analysis was performed using analysis of variance to compare these four groups on all proportions, and specific comparisons (Scheffé test) carried out when F ratios with probabilities of 0.01 or less were obtained. Results from these may be summarised as follows. Language-impaired children in the semantic-pragmatic group produced a significantly higher rate of initiating questions than all other groups, whereas the other language-impaired children did not differ from age-matched controls on this index. For initiating statements, both language-impaired groups produced a higher rate than age-matched controls, but they did not differ from the younger controls in this respect.
Neither language-impaired group differed significantly from age-matched controls in the proportion of verbal minimal responses they gave, whereas there was a substantial difference between older and younger control children in this regard, with the younger children producing fewer responses of this kind. The opposite pattern was found for minimal non-verbal responses, where, again, the language-impaired groups were no different from age-matched controls, whilst the rate was much higher for younger controls.
Neither language-impaired group differed significantly from age-matched controls in the proportions of extended responses or continuations that they used.
Children with semantic-pragmatic disorder did not produce significantly more unintelligible utterances than age-matched controls, but the other language-impaired children did.
Both groups of language-impaired children produced significantly more follow-ups than age-matched controls, being comparable to younger children in this regard.
These analyses are described as exploratory because interpretation of probability levels is complicated by the fact that all proportions sum to one, so the different measures are not independent, and an increase in one index means that there will be a decrease in another. Of particular concern is the possibility that the 'other' language-impaired group might fail to be differentiated from other children simply because they tend to be more unintelligible. However, the absolute number of unintelligible responses was very small in all groups, and recomputation of all indices as a percentage of all intelligible and complete responses had no effect on the pattern of results.
A subsidiary analysis was carried out to investigate how many individual children in each language-impaired group had very abnormal frequencies for a given class of utterance. For this analysis, the mean and standard deviation for each class of utterance (logarithmically transformed) was computed for the age-matched control group C, and used as a basis for computing z-scores for each language-impaired child. The number of children with z-scores below -1.65 or above 1.65 in each group was then calculated. For some measures, only one tail of the distribution could be computed, because the median frequency for controls was zero. Results are shown in Table 5. The Fisher exact test (Hays, 1963) was used to compute the significance of observed differences between each language-impaired group and the controls. On this more stringent analysis, the semantic-pragmatic group again emerge as abnormal in their overuse of initiations and follow-ups. The other language-impaired group is abnormal only in terms of producing a relatively high proportion of unintelligible utterances. Note, however, that this figure is high only in relative terms, and not in absolute terms (see Figure 1).
Despite the subjectivity of many of the definitions of terms, the proportions of different exchange structure categories in a child's conversation can be rated with high levels of inter-rater reliability.
The number of utterances per turn, which may be regarded as an index of how talkative the child was, proved to be a stable characteristic of the individual,
Table 5:Numbers* of children with over- and under-use of exchange structure types
______________________________________________________________________________________________ Semantic-pragmatic(S) Other language Impaired(O) Same age controls(C) Signi- ____________________ __________________________ _______________________ ficant Diffe- Over Normal Under Over Normal Under Over Normal Under rence ______________________________________________________________________________________________ Initiation Question 8(57) 6(43) - 9(17) 43(83) - 4(13) 27(87) - S vs C Statement 4(29) 10(71) 0(0) 6(12) 46(88) 0(0) 0(0) 31(100) 0(0) S vs C Response Minimal Verbal 0(0) 11(79) 3(21) 1(2) 49(94) 2(4) 2(6) 29(94) 0(0) Minimal non- verbal 2(14) 12(86) - 2(4) 50(96) - 2(6) 29(94) - Extended 0(0) 13(93) 1(7) 0(0) 47(90) 5(10) 0(0) 29(94) 2(6) Follow-up 5(36) 9(64) - 4(7) 48(93) - 3(9) 28(91) - S vs C Continuation 0(0) 13(93) 1(7) 0(0) 51(98) 1(2) 0(0) 29(94) 2(6) Unintelligible 3(21) 11(79) 0(0) 19(37) 35(63) - 0(0) 31(100) - O vs C ______________________________________________________________________________________________*Values in parentheses are the percentages.
with high correlations not only over a 2-week interval but even over 9 months. It might have been expected (hat children with semantic-pragmatic disorder would have obtained high scores on this measure, given that they have been described as 'verbose' (Rapin, 1987), but although there was a trend in this direction, it was not statistically significant.
Although results have been mixed, some previous studies have argued that language-impaired children make heavy use of 'back channel responses', defined as utterances that maintain the topic without expanding it or adding new information to it (Fey & Leonard, 1983). In this analysis, there were three categories of utterance relevant to this issue. A minimal response is a classic form of back channel response: it enables the speaker to acknowledge the conversational partner without taking any initiative for further conversation. An extended response does contribute further information, although this may be no more than the minimum required as a response to the preceding question. The clearest evidence that the child is taking the initiative for maintaining conversation and going beyond merely fulfilling minimal obligations to respond in a turn is the continuation. If language-impaired children tend to he passive and non-assertive conversational partners, then we might expect an increase in minimal responses, a slight decrease in extended responses and a sharp decrease in continuations in the language-impaired children. No such trends were found in this study: the proportions of extended responses and continuations used by language-impaired children was just as high as for age-matched controls. In the sample there were one or two children who were extremely passive in conversation and readily responded 'yes', 'no' or 'don't know' whenever they could:. however, these were a minority. The reasons for this discrepancy with the study by Fey and Leonard are not clear, although this result agrees with Edmonds and Haynes (1988) who also found no evidence for increased back channel responding by language-impaired children. One possibility is that the language-impaired child's willingness to engage in conversation with an unfamiliar but friendly stranger is related to educational and social experiences, such as whether the child has suffered teasing by other children or unsympathetic reactions from teachers. Children in special residential schools are spared many of these negative experiences.
Another somewhat surprising result was that whilst there was a steep decline with age in the proportion of non-verbal responses in the control group, there was no evidence that language-impaired children relied heavily on such responses. Thus, although their language development was impaired and their MLU closely resembled that of the younger control group, in two important respects the language-impaired children were different from younger normal children in conversation: they give fewer minimal responses and few non-verbal responses, and in this regard were very like normally developing children of the same age.
The proportion of unintelligible utterances is not of great interest in its own right, and it was not surprising that there was a decline in this variable with age in normal children, and many language-impaired children had relatively high scores. This measure does confirm that those language-impaired children who were coded as cases of semantic-pragmatic disorder were typically the more intelligible. This is to be expected because clinical descriptions of these children typically stress their clear, fluent speech; so this would have been a variable taken into account when classifying children.
The finding that the proportion of initiations decreased with age in control children is almost certainly a consequence of the type of conversational situation we used, in which an unfamiliar adult was in effect interviewing the child. This does not mean that the children were passive respondents to the adult's questions: many gave extended responses to the adult's initiations, continued over several turns. However, they typically did not produce utterances that demanded a response from the adult; Indeed, the results show that in this conversational context this was an immature or even abnormal behaviour if used with any frequency. It is noteworthy that language-impaired children in the semantic-pragmatic group produced a significantly higher proportion of initiations than age-matched controls, whilst other language-impaired children did not. These included statements introducing fresh topics, as well as direct questions to the adult. This appears to be a stable conversational style: the proportion of initiations produced by the child was one of the conversational features that gave a significant test-retest correlation over a 9-month period.
Table 6 shows the agreement in categorisation of turntaking clashes by two independent raters rating 68 instances where overlap had been transcribed. Agreement about the categorisation of overlap was 76%.
Table 6: Agreement between two raters on category of overlap
Rater A Rater A Rater A Rater A | Inadvertent Rule violation Adult Interrupt Adult follow-up ________________________|___________________________________________________________ | Rater B Inadvertent | 7 1 6 1 Rater B Rule violation | 1 16 0 1 Rater B Adult Interrupt | 2 1 16 1 Rater B Adult follow-up | 0 0 2 13
Language-impaired children compared with younger and age-matched controls.
Turntaking problems (i.e. gaps, overlaps and interrupts) were relatively rare events and, to obtain sufficient data for meaningful comparisons, it was necessary again to combine the 4- to 6-year old and 8- to l2-year old control children. For each class of problem, the percentage of children scoring 0, 1, 2, 3, 4, or 5 or more on a turntaking measure is shown in Table 7. Statistical comparisons were made using a Kruscal-Wallis non-parametric test (Hays, 1963). The only result to reach significance was for the category of adult interrupt (H =17.11, d.f.=3, P<0.001), where further tests contrasting two groups showed that children in the semantic-pragmatic group scored significantly higher than age-matched controls (H=14.17, d.f. = 1, P<0.001) and higher than other language-impaired children (H=6.79, d.f. = 1, P<0.05) The difference between the language-impaired children without semantic-pragmatic disorder and age-matched controls also reached significance (H=4.18, d.f.= I, P<0.05).
Table 7: Percentages of children producing instances of turntaking, overlap or gap.
______________________________________________________________________________________ Inadvertent Violating Adult Gap overlap overlap interrupt Number of ___________ ___________ __________ __________ Instances S O Y C S O Y C S O Y C S O Y C ______________________________________________________________________________________ 0 43 55 56 68 36 48 50 65 29 55 75 84 43 56 47 61 1 43 17 28 26 29 21 31 19 36 33 14 13 7 19 22 29 2 7 21 14 6 14 21 6 16 29 10 8 3 14 14 14 3 3 7 7 0 0 0 5 8 0 7 2 0 0 0 5 6 0 4 or more 0 0 3 0 21 5 6 0 0 0 3 0 36 7 11 6 ______________________________________________________________________________________S: Semantic-pragmatic disorder (n=14).
The distinction between inadvertent overlap and rule-violating overlap proved easy to make, and there was good agreement between raters on this. Somewhat more problematic was the distinction between inadvertent overlap and adult overlap. Adults seldom cut right across children as they spoke, but tended to come in at a point where the child could legitimately be assumed to have finished ('transition relevance place'), making coding ambiguous. Consider, for instance:
A: who went on holiday with you?/ C: my dad couldn't come --- *because he was ill/ A: *why was that?/
This particular transcription suggests a coding of adult interrupt, but had the sequence been transcribed differently, a coding of inadvertent overlap would seem more appropriate, e.g.
C: my dad couldn't come --- / A: *why was that?/ C: *because he was ill/
Inadvertent overlap does not indicate poor turntaking skill, and so our failure to find any increase in its frequency in language-impaired children is to be expected. It was, however, expected that there would be evidence of more violating overlaps in the semantic-pragmatic group, and this expectation was not confirmed. Nevertheless, there was a suggestion in our data that, although not characteristic of any one group, a high frequency of violating overlap might be a conversational feature of certain individuals. There were four language-impaired children who generated violating overlaps at a rate far outside those observed in normal children, producing 11, 7, 6 and 5 instances of violation respectively. Two of these children were classified as clear instances of semantic-pragmatic disorder. The third was a child who was not categorised as having semantic-pragmatic disorder by his school, but who had all the features of this disorder except that his speech was not clearly articulated (see Bishop & Adams, p.245). The last of these children had not appeared conversationally abnormal to the adult with whom he conversed.
The relatively high frequency of adult-generated clashes in conversations with language-impaired children needs further investigation. The impression was that such clashes arose for three quite different reasons. In some cases the adult appeared to be trying to regain control of the conversation from a child who was chattering incessantly, and could only do so by butting in. In other cases, the adult had trouble predicting turntaking exchange points because the child failed to use appropriate intonation. In many cases, the clash seemed to arise because the child was so slow in formulation that the adult would assume that the child had completed an utterance, just at the point when the child continued. It would he valuable in a future study to see if it is possible to devise objective criteria for distinguishing these different reasons for adult interruption, and explicitly to vary adult speed of response to see what effect this has on the child's side of the conversation.
The lack of significant findings in analysis of gaps should he interpreted with caution. The data shown in Table 7 indicate a trend for excessive numbers of gaps in the conversations of some children with semantic-pragmatic disorder, and this might become clearer with a larger sample of children and/or a longer stretch of conversation. Another consideration is that gaps were transcribed only if they sounded obvious when listening to the tape-recording, but more fine-grained analysis involving timing of inter-speaker interval with a stopwatch may well be more revealing of differences between language-impaired and normally speaking children. It would be of particular interest to see how pause duration relates to complexity and predictability of the child's response in relation to the preceding utterance (compare Garvey & Berninger, 1981).
Inter-rater agreement for 20 conversations rated by two independent analysts was high and significant for the categories appropriate response to request for clarification (r=0.701, P<0.0l) and self-repair (r=0.840), but non-significant for the categories of inappropriate response to request for clarification (r= -0.05) and child request for clarification (r=0.305). Test-retest correlations for 20 children seen twice over a 2-week interval were significant for the categories of appropriate response to request for clarification (r=0.426) and child request for clarification (r=0.498), but not for inappropriate response to request for clarification (r= -0.09) or self-repair (r = 0.295).
Comparison of normal and language-impaired children
Instances of repairs, like turntaking, were rare, so data were summed for control children in the younger (4-6 years) and older (8-12 years) age group. Results are shown in Table 8. Differences between groups in the numbers of requests for clarification made by the adult (coded <R1> or <R2>) were highly significant (H= 19.9, d.f. =3, P<0.00l). Supplementary Kruskal-Wallis tests indicated that both language-impaired groups differed significantly from age-matched controls on this measure, although the trend for a greater frequency
Table 8: Percentages of children producing instances of repair behaviours.
______________________________________________________________________________________ Adult request Inappropriate Child request Self-repair for clarification response to for clarification (R4) (R1 + R2) clarification (R3) request (R2) Number of __________________ ______________ ________________ ______________ Instances S O Y C S O Y C S O Y C S O Y C ______________________________________________________________________________________ 0 21 47 53 81 64 86 83 100 43 72 83 74 29 35 44 42 1 14 21 19 16 29 9 11 0 29 19 11 16 21 28 14 19 2 14 5 3 3 0 2 1 0 14 5 3 6 21 12 17 26 3 7 12 11 0 0 0 0 0 7 2 3 0 7 7 11 10 4 or more 43 16 14 0 7 0 3 0 7 2 3 0 21 18 14 3 ______________________________________________________________________________________S: Semantic-pragmatic disorder (n=14).
in the semantic-pragmatic group compared with other language-impaired children did not reach significance.
Differences between the groups in frequency of requests for clarification by the child fell just short of significance. The trend was for the semantic-pragmatic group to produce more of these than other children.
There was no difference between groups in the frequency of self-repair.
The high frequency of adult requests for clarification for children in the semantic-pragmatic group confirms the difficulty that the adults experienced in talking with these children. As intelligibility was seldom a problem for this group, we can assume that the content rather than the form of the utterance was usually the problem.
Johnston (1985) noted that ability to reformulate an utterance to repair may depend on linguistic knowledge or flexibility, but in our sample it was not the linguistically least competent who had most difficulty in producing an appropriate repair. The trend, though non-significant, was for the semantic-pragmatic group to be most likely to score an inadequate response to a request for clarification. It would be of interest to study this aspect of conversational behaviour in children with semantic-pragmatic disorder in more detail by using Gallagher and Darnton's (1978) technique in which an adult periodically feigned misunderstanding ('What?').
Donahue, Pearl and Bryan (1980) investigated conversational competence in learning-disabled children using a referential communication task in which an adult gave ambiguous messages. They found that the learning-disabled group could recognise which messages were inadequate, but were less likely than other children to request clarification. In contrast, we found that in conversation language-impaired children were as likely as other children to request clarification, suggesting that we cannot extrapolate from an artificial situation to behaviour in a more natural setting. However, it must be noted that the overall rate of child requests for clarification was very low for most children.
Inter-rater agreement was good for all indices: correlations between two independent analysts rating 20 conversations were 0.978, 0.610 and 791 for indices of cohesion, situational reference and unestablished reference respectively. Test-retest correlations for 20 children over a 2-week interval were 0.573 (P<0.0l), 0.401 (NS) and 0.500 (P<0.05) for the three indices respectively.
Normal developmental trends
An index of cohesion was calculated as the total number of referring terms coded as <L> (referent retrievable from linguistic context) multiplied by 100 and divided by the total number of utterances. The resulting figure is not a percentage, as it is possible for a child to use more than one referring term in an utterance, but it does give a quantitative index of the use of referring terms relative to the length of the conversation.
Indices of exophoric reference and unestablished reference were computed in the same way.
In the control group, there was a modest tendency for the cohesion index to increase with age (r  = - 0.299, P<0.05) over the entire range from 4 to 12 years, but this was no longer significant when only 8 to 12 year olds were considered (r  = -0.17, NS).
The index of exophoric reference did not show any relationship with age (r  = 0.204, NS), whereas the index of unestablished reference showed a slight tendency to decrease with age (r  = -0.275, P<0.05).
Comparison of control and language-impaired children
Mean scores on each reference index are shown for the two language-impaired groups and for younger and older control children in Table 9. Each index was
Table 9: Cohesive and non-cohesive use of pronouns and delimiters.
______________________________________________________________________________________ Rate per utterance x 100* ____________________________________________________________ Cohesive Exophoric Unestablished Use reference reference ______________________________________________________________________________________ Semantic-pragmatic 19.4 (13.44) 8.7 (2.94) 6.0 (6.87) Other SLI 12.7 (8.21) 7.1 (4.01) 2.7 (3.12) Control 4 to 6 year olds 14.4 (11.91) 8.0 (7.81) 3.9 (4.84) Control 8 to 12 year olds 19.0 (10.89) 10.2(5.89) 2.3 (2.45) ______________________________________________________________________________________* Values are mean with s.d. in parenthesis
transformed logarithmically to stabilise variances before conducting an analysis of variance. There was a significant main effect of group for the index of cohesion (F=5.28; d.f.=3, 120; P<0.0l), and Scheffé tests indicated that children with semantic-pragmatic disorder did not differ from age-matched controls, whereas other language-impaired children produced significantly lower indices of cohesion, not differing in this regard from the younger controls.
For the index of exophoric reference, the main effect was of marginal significance (F=2.68, d.f. =3, 120; P<0.05) but no specific comparisons reached significance. The main effect for the index of unestablished reference was non-significant.
Cohesive use of personal and demonstrative terms can be rated with high reliability, and shows stability over a 2-week period. Lack of cohesion is not a feature of semantic-pragmatic disorder, but is characteristic of the conversations of other language-impaired children, whose use of cohesion is like that of younger normal children of equivalent MLU. We had anticipated that use of unestablished reference might characterise the children with semantic-pragmatic disorder, but, although there was a trend in this direction, it was not statistically significant.
Although there were significant differences between groups in use of cohesion, there was very wide variability within the control group, and differences between children of different ages were slight compared to the range of variation within a given age group. Thus a wide range of use of cohesion is normal, and an index of cohesion is unlikely to be or great clinical utility, except where lack of cohesion is extreme.
Despite the subjectivity of many of the categories used, it seems that analysis of a 5-10 minute stretch of conversation can generate reliable and useful data. However, for some categories, especially those under the heading of turntaking and repair, low levels of inter-rater reliability were obtained. It is probable that low reliability was at least in part a consequence of the rarity of these categories, where the range of scores was very small and the median number of occurrences per conversation was close to zero. Longer samples of conversation might give more stable results for these indices.
For those indices that were reliable, several differences between language-impaired and control children emerged, but it was evident that language-impaired children do not form a homogeneous group, and it is not sensible to ask 'do language-impaired children have conversational impairments?'. Results from this study support Fey and Leonard's (1983) view that 'the population of SLI [specific language impairment] children is no more homogeneous with respect to pragmatic skills than it is with respect to semantic, syntactic, or phonological abilities'. Furthermore, the usual practice of comparing language-impaired children with 'language-matched' controls can be shown to be oversimplistic. As can be seen from Table 2, a control group which is matched to a language-impaired group on MLU may be very different in terms of level of comprehension. Rather than treating language as a unidimensional variable that may be impaired to a greater or lesser extent, it is preferable to consider a range of components of language function, to see how far these may be dissociated in development. This study has shown that conversational problems cannot be simply explained as an invariable concomitant of expressive language impairment: in many regards, those children with immature or abnormal conversational skills were not the most severely impaired in terms of expressive abilities.
Whilst most professionals who have worked with language-impaired children agree that they are not all identical, there is little agreement as to how children should be subclassified. The notion that there is a distinct subgroup with 'semantic-pragmatic disorder' has been proposed on the basis of clinical observations, but still needs to be validated. This study may be regarded as first step in that direction. Certain conversational features that have been described as characterising semantic-pragmatic disorder were quantified in conversational analysis. Children identified as matching the description of semantic-pragmatic disorder did not generate an unusually large number of utterances, but they did produce a high rate of initiations in conversation with an adult, and this was shown to be a stable and abnormal conversational characteristic. There was also a trend for these children to produce more interrupts that violated turntaking rules, although this seemed to be a pronounced tendency in a few individuals rather than a general characteristic of the whole group. The adult's difficulties in conversing with these children were apparent in the higher frequency of adult interrupts and requests for clarification. Use of cohesion was normal for children with semantic-pragmatic disorder, although other types of language-impaired child had less cohesive conversations.
This study shows that the conversational features of exchange structure, turn-taking, repair and cohesion can be identified from a short semi-structured conversation with adequate inter-rater reliability, and that these indices do go some way towards identifying those conversational characteristics that lead to a diagnosis of semantic-pragmatic disorder. Our impression, however, was that analysis of these aspects of conversational behaviour did not capture the essence of what was abnormal about the language of children with semantic-pragmatic disorder. A further study was therefore conducted with the aim of identifying those language behaviours that led to an impression of inappropriacy, to see if it was possible to pinpoint those features that typified children with semantic-pragmatic disorder. Results from this study are described in the next paper (see p.241).
We are grateful to the staff and pupils of the following schools for making it possible for us to gather conversational data with children: Dawn House School, Mansfield; John Horniman School, Worthing; Percy Hedley School. Newcastle-upon Tyne; Rosstulla Language Unit, Belfast; Ewing School, Manchester; Matthew Arnold School, Liverpool; Fazakerly School, Liverpool; Lewis St Primary School, Eccles; Clifton Primary School, Swinton; St Kentigern's Primary School, Manchester; St Thomas RC Secondary School, Whalley Range; Brownhill School. Rochdale; Bruntwood Speech and Language Unit, Cheadle Hulme; Leveredge Lane County Primary School Language Unit, Bolton; Crab Lane Primary School, Manchester; Booth Hall Hospital School, Manchester. Especial thinks are due to Mrs Hazel Roddam for assisting with our reliability checking, and to speech therapists at Dawn House School who collected additional conversations for phase 2 of the study.
This work was supported by Medical Research Council project grant no G5517435N.
Received July 20 1988; revised version received March 8 1989.
Dorothy Bishop will be happy to send copies of further research papers on semantic-pragmatic disorder to interested parties.
Click here for an updated list of her publications.
However please note that, as a full time researcher she is not in a position to offer opinions, assessments or consultations on individual children.
Copyright Holders Note:
Permission has been granted to put a copy of this article on this World Wide Web page by Whurr Publishers Limited on behalf of the copyright holders: The Royal College of Speech and Language Therapists. It is not for reproduction without permission.