Journalof Speech and Hearing Research, Volume 35, 1303-1315, December 1992 Spoken and Written Language Relationships in Language/Learning.Impaired and Normally Achieving School.Age Children Ronald B. Gillam* University of Missouri-Columbia Judith R. Johnston University of British Columbia Vancouver, Canada Students with language/learning impairment (LLI) and three groups of normally achieving children matched for chronological age, spoken language, and reading abilities wrote and told stories that were analyzed according to a three-dimensional language analysis system. Spoken narratives were linguistically superior to written narratives in many respects. The content of written narratives, however, was organized differently than the content of spoken narratives. Spoken narratives contained more local interconnections than global interconnections; the opposite was true for written narratives. LLI and reading-matched children evidenced speakingwriting relationships that differed from those of the age- and language-matched children in the way language form was organized. Further, LLI children produced more grammatically unacceptable complex T-units intheir spoken and written stories than students from any of the three matched groups. The discussion focuses on mechanisms underlying the development of speaking-writing differences and ramifications of spoken-language impairment for spoken and written-language relationships. KEY WORDS: speaking, writing, narratives, school-age children, language Impaired The linguistic interconnections between speaking and reading and the ramifications of linguistic deficits for literacy learning have received considerable attention in the recent linguistics, psychology, special education, and communication disorders literature. Children who are language impaired frequently experience difficulty learning to read (Gillam & Johnston, 1985; Maxwell & Wallach, 1984; Stark & Tallal, 1988). Conversely, poor readers present performance limitations on a variety of speaking tasks (Kamhi, Catts, & Mauer, 1990; Norris & Bruning, 1988; Scarborough, 1990; Snyder & Downey, 1991). Like reading, writing is also a language-based academic skill (Naremore & Hopper, 1990; Smith, 1982). Because language-learning-impaired children evidence speaking and reading difficulties, it is likely that they also experience difficulties with writing. Writing has historically been viewed as secondary to and parasitic upon spoken language (Barthes, 1967; Saussure, 1959). However, recent empirical (Beaugrande, 1984; Chafe, 1985; Chafe & Danielewicz, 1987; Rowe & Harste, 1986) and theoretical advances (Culler, 1975; Derrida, 1981; Vygotsky, 1978) have produced compelling 'Currently affiliated with the University of Texas at Austin. © 1992, American Speech-Language-Hearing Association 1303 0022-4685/92/3506- 1303$01.00/0 1304 Journal of Speech and Hearing Research reasons to consider spoken and written language as complex symbol systems with their own distinctive characters. During the preschool and early school-age years, children investigate the equivalencies of speaking and writing as they interweave playing, writing, talking, and drawing experiences (Baghban, 1984; Dyson, 1983, 1989; Ferreiro & Teberosky, 1982; Harste, Woodward, & Burke, 1984). As they gain control of processing constraints associated with mechanics and spelling, children begin to explore the symbolic potentials that are unique to each mode (Dyson, 1991; Shatz, 1984). Between the ages of 9 and 12 years, children's speaking and writing become increasingly differentiated (Kroll, 1981). Written texts become longer and more cohesive, contain more subordinated structures, and are better organized than spoken texts (Perera, 1984; Rubin, 1982). Writing strategies also change. For example, children start to plan before they write, and begin to self-evaluate sentential and textual form (Burtis, Bereiter, Scardamalia, & Tetroe, 1983; Calkins, 1983; Daiute, 1989; Erftmier & Dyson, 1986). Although it is not altogether clear how knowledge of specific differences between spoken and written language emerges, it is clear that children apply their realizations about modality differences relatively late in the language development process. Differences between spoken and written texts may occur because written language poses unique conceptual, linguistic, and mechanical constraints. Conceptually, writers create communicative context, project the information needs of an imagined audience, reflect and rereflect on intended meaning, elaborate their messages, and shape their texts in accordance with genre-specific organizational schemata. Linguistically, they make lexical choices that convey intended meanings and mood, and formulate sentences that best convey the interconnections between propositions. Mechanically, they deal with the perceptual and motor requirements of handwriting whie they adhere to spelling, capitalization, and punctuation conventions. Written products represent an orchestration of the interconnected demands associated with a variety of later-developing processes. Given the difficulty of the writing process and the fact that important discoveries about written language occur late in development, it is possible that writing might be an especially difficult task for LLI children. Although much is currently known about the listening, speaking, and reading deficiencies presented by LLI children, relatively little is known about their writing. Purpose The present study was designed to investigate potential differences between the cognitive and linguistic properties of spoken and written language. To accomplish this, spoken and written narratives produced by school-age children were broadly examined for structure and complexity properties at sentential and textual discourse levels. A second purpose was to determine whether the relationship between spoken and written language was different for LLI children as compared to normally achieving children. To this end, the spoken and written language performance of LLI children was compared to that of three groups of normally achieving children: a chronological/nonverbal mental age matched group, a 35 1303-1315 December 992 spoken-language-ability matched group, and a reading-ability-matched group. Two research questions were addressed: 1. Are the conceptual and linguistic characteristics of written narratives produced by 9- to 12-year-old school children the same as those of their spoken narratives? 2. Is the relationship between the conceptual and linguistic properties of spoken and written narratives the same for LLI and NL children matched for age, spoken-language ability, and reading ability? Method Subjects Forty school-age children participated in this study. The core group contained 10 language/learning-impaired (LLI) children between the ages of 9 and 12 years. Each LLI student was matched with 3 same-sex, normally achieving students: a student matched for age, a student matched for spoken language ability, and a student matched for reading ability. All four groups consisted of 7 boys and 3 girls. Children in the four groups had normal hearing and vision and came from monolingual, English-speaking homes. The language/learning-Impaired(LLI) group. LLI subjects had average or above average nonverbal intelligence together with significant deficits in spoken language and reading. All LLI students had been diagnosed as learning disabled by school district special education personnel according to Wyoming state guidelines. Table 1 presents profile information for the LLI subjects. The chronological age range for the 10 subjects was 9:0 to 11:7 with a mean of 10:7 (years:months). The LLI students' nonverbal cognitive abilities, as indicated by their performance on the Test of Nonverbal Intelligence (TONI) (Brown, Sherbenou, & Johnsen, 1982), were well within normal limits (TONI quotient X = 103.6; range = 93-129). Their verbal abilities, as indicated by performance on the verbal aptitude composite of the Detroit Test of Learning Aptitude-2 (DTLA-2) (Hammill, 1985), were well below age expectations (Verbal Cluster quotient, X = 75.4; range = 72-80). Discrepancies between verbal (DTLA-2) and nonverbal (TONI) quotients ranged between 16 and 52 points with a mean of 28.2 points. Age percentile scores on the Reading Recognition subtest of the Peabody Individual Achievement Test (PIAT) (Dunn & Markwardt, 1970) ranged from the 2nd to the 12th percentile with a mean percentile value of 6.3. Jeff, a typical LLI student, was a fourth grader who had repeated first grade. His nonverbal cognitive abilities were at the fifth grade level, his spoken language abilities were at the early third grade level, and his reading abilities were at the middle second grade level. The age-matched (AGE) group. Same-sex students whose birth dates fell 3 months from an LLI subject's and whose teachers estimated were functioning within the middle two quartiles of their class were asked to take home an explanation of the research project and a parent permission form. Upon return of this form, the TONI was administered. For each LLI student, the first same-sex, age-appropriate student earning a TONI score in the normal range was enrolled as an AGE group match. After selection, age- Gillam & Johnston: Spoken and Written Language 1305 TABLE 1. Comparisons between the LLI, AGE, LANG, and READ subjects. Group Subject CA 1 2 3 4 5 6 7 8 9 10 11:1 9:11 11:6 9:0 11:7 11:3 9:9 11:2 10:5 10:1 TONI 101 97 103 93 99 129 109 96 99 110 SI 10 11 8 11 11 12 14 15 9 5 PIAT 26 26 30 22 32 31 27 26 22 24 CA TONI 10:10 10:0 11:6 9:1 11:7 11:5 9:10 11:2 10:5 10:1 93 101 103 97 97 113 93 95 102 101 READ LANG AGE LLI SI 22 16 18 15 17 17 15 21 17 13 PIAT CA TONI 44 55 47 46 51 52 46 68 53 36 8:1 7:10 8:9 6:8 8:3 7:11 7:11 8:9 7:7 7:8 96 105 101 101 105 98 108 97 107 96 SI 10 11 8 11 12 14 13 15 11 6 PIAT CA TONI SI PIAT 36 33 39 41 38 41 34 47 35 33 7:3 7:6 8:9 7:4 9:0 7:10 8:2 7:11 6:8 6:9 96 98 93 91 105 98 98 108 98 113 10 8 16 8 16 12 12 12 7 14 26 26 31 20 33 31 27 26 22 25 X 10:7 103.6 10.6 26.6 10:7 99.5 17.1 49.8 7:11 101.4 11.1 37.7 7:9 99 11.5 26.7 SD 0:10 10.4 2.9 3.5 0:9 5.6 2.7 8.4 0:7 4.6 2.68 4.5 0:9 6.8 3.2 4.1 Note. CA = chronological age (Yr:Mos); TONI = Test of Nonverbal Intelligence quotient; SI = Detroit Test of Learning Aptitude-2 sentence imitation subtest; PIAT = Peabody Individual Achievement Test word identification subtest raw score. matched control subjects were given the reading recognition subtest of the PIAT and the sentence imitation subtest of the DTLA-2 for descriptive purposes (Table 1). The spoken language-matched (LANG) group. A language age was computed for each LLI child by multiplying the verbal quotient obtained on the DTLA-2 and chronological age (in months). This product was converted to an age:months value. Same-sex children whose chronological age was +3 months from the language age of an LLI child and whose teachers estimated were functioning within the middle two quartiles of their class were asked to take home an explanation of the research project and a parent permission form. The language-matching process was further refined by adding performance on the DTLA-2 sentence imitation subtest to the selection criteria. Sentence imitation tasks have proven to be powerful for discriminating between language-impaired and reading-impaired children, and between reading-impaired and normally achieving children matched for reading ability (Kamhi & Catts, 1986; Stahl & Erickson, 1986). The DTLA-2 sentence imitation subtest and the TONI were administered upon return of the permission form. Each spoken-language-matched control had a chronological age that was 3 months from an LLI subject's language age, obtained a Sentence Imitation score within ±2 points of the LLI match's, and earned a TONI quotient above 90. In cases in which more than one child fit this criteria, the child whose sentence imitation score was the closest to the LLI child's was selected. After selection, language-matched control subjects were given the PIAT reading recognition subtest for descriptive purposes (Table 1). The reading-matched (READ) group. LLI subjects were matched with reading-level control subjects on the basis of word recognition ability as measured by the reading recognition subtest of the PIAT. This measure was selected over a reading comprehension measure because evidence suggests that word recognition skill substantially determines the extent to which contextual information is used during reading (LaBerge & Samuels, 1974; Perfetti, 1985; Stanovich, 1986). The LLI childrens' PIAT reading recognition raw scores were first con- verted to grade scores. The matching scheme involved equating the PIAT reading recognition grade level score eamed by each LLI student with the actual grade level of a potential control group member. Same-sex children in the appropriate grades were asked to take home an explanation of the research project and a parent permission form. The PIAT reading recognition subtest and the TONI were administered upon return of the permission form. Selected reading-matched controls obtained TONI quotients above 90 and PIAT raw scores that were ±2 points from scores already earned by an LLI subject. After selection, reading-matched controls were given the DTLA-2 sentence imitation subtest for descriptive purposes (Table 1). Group Comparisons Preliminary ANOVA tested whether the a priori-designed targets of subject selection were met. The LLI and AGE groups were significantly older and presented higher nonverbal mental ages than the LANG and READ groups. The mean Reading Recognition score of the AGE group was significantly higher than that of the LANG group which, in turn, was significantly higher than that of both the LLI and READ groups. The mean Sentence Imitation score for the AGE group was significantly higher than that of the LLI, LANG, and READ groups with no significant differences between the values for the latter three groups. The LLI and AGE groups equated for chronological and nonverbal mental age and the LLI and READ groups equated for reading recognition ability. The mean language age of the LLI group was equivalent to the mean chronological age of the LANG group. The LLI and LANG groups also equated for sentence imitation ability. Procedures Each subject produced two spoken and two written stories. Participants were shown sets of three 7" x 10"' color pictures 6.5 1306 Joumal of Speech and Hearing Research December 19, 1990mounted to 10" x 13" sheets of construction paper. Each set of three pictures contained an outdoor action picture, a nature picture, and a portrait. Picture sets were changed for each narrative to prevent participants from telling or writing the same story more than once. Students were asked to create a story about any one of the three presented pictures. After selecting a picture, they were told to take a couple of minutes to think of a story. When they indicated they were ready, they were asked to tell their story. Spoken narratives were tape-recorded. In the writing conditions, participants were told, "This time I want you to write your story. I can't help you with the writing. Just write your story so that you can read it back to me." After writing their stories on paper provided by the investigator, participants were asked to read their stories aloud. Readings were tape-recorded for use during data analysis in the event that a child's handwriting or spelling made it difficult to determine what had been written. Two presentation sequences were used to control for possible task order effects. Each LLI student and her/his three matches received the same sequence, with sequences counterbalanced across the four groups. Sequence A children produced the spoken narrative before the written during the first session, and the written narrative before the spoken in the second session. The order was reversed for Sequence B children. Spoken and written language samples were segmented into T-units (Hunt, 1970) and were transcribed according to Systematic Analysis of Language Transcripts (SALT) (Miller & Chapman, 1984) conventions. Only the scores from each participant's "longest" spoken and written narrative were used for the present analysis. Length in this case was determined by number of story constituents. Thus, 80 (40 spoken and 40 written stories) of the 160 narratives that were collected were subjected to a complete language analysis. Students selected nature pictures (35% spoken, 52.5% written) and outdoor action pictures (37.5% spoken, 40% written) over portraits (27.5% spoken, 7.5% written). Selection trends were similar for all four groups. Note also that, for both modalities, "active event" pictures were selected approximately 40% of the time, and "static scene" pictures, approximately 60%. As a further check for possible effects of picture selection on the narrative data, preliminary ANOVA tested for potential differences in story length (measured as number of story constituents) as a function of group and picture selection. Group x Picture Type interactions were insignificant for both spoken [F(6, 28) = .713, p = .642] and written [F(6, 28) = .896, p = .496] modes. Language Analysis Two types of complexity were calculated for each narrative. One set focused on complexity of linguistic form, the other on complexity of underlying content. Complexity was defined in two ways, by amount and by degree/nature of organization. Within this framework, analogous measures were developed at sentence and text levels of discourse. The resulting analysis scheme consisted of eight measures for 35 1303-1315 December 1992 TABLE 2. Measurements used for the analysis of spoken and written narratives. Level Domain Form Amount Organization Content Amount Organization Sentence Text MLT-u Percent of complex T-units T-units per story Connectives per T-unit Propositions per T-unit Predicate types per T-unit Constituents per story Percent of dyadic constituents each narrative (Table 2), four measures of form and four measures of content, as described below. Measures of form complexity 1. Amount (sentence ievel)-Morphems per T-unit (MLT-u). The sentence-level measure of amount of language was mean length of T-unit. Brown's (1973) conventions for morpheme segmentation were followed during transcription. Actual values were generated by the SALT program as reported in the SALT "Word and Morpheme Summary." 2. Amount (text level)-Number of T-units per story (Tunits per story). As a textual analog to MLT-u, this component of the analysis system measured the total number of T-units contained in each story. 3. Organization (sentence level)-Percentof complexcorrect T-units (complex T-units). Syntactic devices such as complementation and subordination indicate the relationships between major constituents of a sentence. Greater use of such devices implies a higher degree of hierarchical organization and/or greater success in explicating structural links. T-units were judged to be complex if they were grammatically complete and correct, and contained a main clause together with one or more additional coordinating, subordinating, complementing, or relative clauses. The measure consisted of the number of complex, grammatically correct T-units divided by the total number of T-units. 4. Organization (text level)-Number of connectives per T-unit (connectives per T-unit). Speakers and writers use connectives to explicate the links between adjacent and nonadjacent phrases and clauses in a text. The forms of interest were as follows: * Causal connectors. These forms indicate that one proposition expresses the reason, motive, or ground for another proposition. Because, since, to (when it conveys the meaning "in order to" as in "He went to Casper to buy a car") and then (when it conveys the meaning "because of that" as in "He stole some cars. Then he got put in jail") are causal connectors. · Conditional connectors. These forms indicate that one proposition modifies or restricts the nature or occurrence of another. But, so, instead, anyway, except, though, although, in case, in that case, if, however, except for, and or else are conditional connectors. a Temporal connectors. These forms indicate that one proposition chronologically precedes or follows another. Then, Gillam & Johnston: Spoken and Written Language after, when, until, next, before, finally, later, soon, while, and whenever are temporal connectors. Following Thompson (1984), the connective and was not counted due to the difficulty of distinguishing between its use as a discourse adverbial and its use as a connective. With this exception, the measure consisted of the total number of connectives in a story divided by the total number of T-units. The use of T-units as a denominator controlled for potential differences that might have resulted solely from story length. Measures of content complexity 1. Amount (sentence level)-Number of propositions per T-unit (propositions per T-unit). Utterances express ideas about states of objects and/or the relationships and events that link them. Logically, idea units, or propositions, consist of a predicate (a judgment about a state, relation or event) together with its affiliated arguments (the "objects" that play a role in the predicated state, relation, or event). For example, one may judge that a bike is RED; that Billy is a beneficiary who OWNs something (the bike); that Billy is an agent who is RIDING something (the bike); and/or that an event takes place IN a location (a backyard). Each of these judgments would constitute a coherent, separate unit of thought. Depending upon the context, one or more elements of this set of ideas could motivate a variety of T-units, with each T-unit expressing one or more underlying propositions ("The bike is red." "Billy has a red bike." "Billy is riding his bike." "Billy is out in the backyard riding his red bike."). For the present analysis, sentence content complexity was indexed by calculating the average number of underlying propositions per T-unit. Further technical details can be found in Antinucci and Parisi (1976), Johnston and Kamhi (1984), and Kamhi and Johnston (in press). 2. Amount (text level)--Number of constituents per story (constituents per story). Sutton-Smith, Botvin, and Mahony (1976) and Botvin and Sutton-Smith (1977) detailed an analytic approach to children's narratives in which stories were segmented into a series of actions or events called "plot units." Plot units were symbolized as verbal nouns (e.g., escape, rescue, departure, capture, threat, interdiction, etc.) that indicated action or potential action. Within this system, separate plot units combined to form "dyads" when both the action or potential action and the resolution were expressed. For this investigation, a modification of the "plot unit" system was devised that incorporated contextual elements together with action and potential action elements. To clearly differentiate between the morphology of Botvin and SuttonSmith's system and this adaptation, the fundamental content unit was referred to as a "constituent." Three types of constituents were distinguished: contextual constituents, dyadic constituents, and single process constituents. Contextual constituents included the participants in the story and the circumstances surrounding the story events. Dyadic constituents had logical conjugates that, when paired, formed problem-resolution units. Single-process constituents, by their very nature, did not have logical conjugates. These various types of constituents are further defined in the Appendix. 3. Organization (sentence level)---Number of predicate types per T-unit (predicate types per T-unit). As explained by 1307 Antinucci and Parisi (1976), there are three ways in which propositions can be added to a nuclear, or focused, proposition within an utterance boundary. In the first type of linkage, an adverbial predicate takes the entire nuclear proposition as an argument. In the second type, an embedded proposition serves as an argument of the nuclear proposition. Inthe third type, an associated proposition shares an argument with the nuclear predicate. All three types of predicate links appear together in the sentence, "Yesterday I taught the little girl to put a reed in the clarinet" (Johnston, 1986). The nuclear predicate TEACH has as its arguments an agent, I, an experiencer, girl, and a patient ("that which was taught") that is expressed by an embedded proposition centered around PUT IN. The embedded predicate PUT IN takes an agent argument, girl, a patient argument, reed, and a locative argument, (in) the clarinet. The superordinate adverbial predicate YESTERDAY takes the entire nuclear proposition as its patient argument. Finally, LITTLE is an associated predicate that shares the argument girl with the nuclear predicate TEACH. In the idea network underlying this sentence, the three types of linkages occur at different levels of hierarchicalization. The average number of different types of predicates (nuclear, adverbial, embedded, and associated) expressed in a T-unit is tantamount to the average degree of hierarchicalization, and provides a metric for the overall complexity of content organization. 4. Organization (text level)--Percent of dyadic constituents. Recall that certain constituents can be linked with logical conjugates to form problem-resolution units called dyads. At the text level of discourse, organizational complexity increases as dyads are combined into hierarchical levels of plots and subplots. Botvin and Sutton-Smith (1977) found that increased hierarchicalization of plot structures was indirectly related to narrative length. With development, children told longer stories. However, longer stories were not always the most structurally complex. The most advanced storytellers were able to weave their constituents into three or more levels of interrelated dyads. Given two stories with the same number of constituents, the most complex story was the one containing more interconnected problem-resolution pairs. For this reason, the percent of constituents that were members of dyads was selected as the metric for organizational complexity. Reliability The first author calculated the eight language indices for the entire set of 80 narratives. To establish the reliability of the initial analysis, spoken and written narratives produced by two participants from each subject group were selected for re-scoring through a process of random selection within stratified sampling. The 16 selected narratives (20% of the total) were presented to two independent raters in randomized order and in blind fashion. The second author, a developmental psychologist with the Certificate of Clinical Competence in speech-language pathology, independently analyzed the reliability narratives for Propositions per T-unit, Predicate Types per T-unit, and Percent of Complex T-units. No training sessions were 1308 Journal of Speech and Hearing Research necessary because this rater had developed the propositional analysis system and had extensive experience with the grammatical analysis of child language transcripts. Reliability scoring yielded point-by-point interrater agreements of 96.7% for Propositions per T-unit (2,610 agreements out of 2,699 judgments), 92.9% for Predicate Types per T-unit (1,078 agreements out of 1,160 judgments), and 94.9% for Percent of Complex T-units (281 agreements out of 296 judgments). The second rater, a school counselor with no training in speech-language pathology, analyzed three textual properties of the narratives: Constituents per Story, Percent of Dyadic Constituents, and Connectives per T-unit. Six, 1-hour training and practice scoring sessions were completed before the second rater performed the independent analyses. Reliability scoring yielded point-by-point interrater agreements of 88.7% for Constituents per Story (479 agreements out of 540 judgments), 91.11% for Percent of Dyadic Constituents (492 agreements out of 540 judgments), and 91.45% for Connectives per T-unit (182 agreements out of 199 judgments). Reliability ratings were not deemed necessary for MLT-u and T-units per Story because these values were computergenerated by SALT. Reliability percentages fell well within acceptable ranges and indicate that the first author's scoring was not significantly biased by his knowledge of or involvement in the study and, further, that the three-dimensional narrative analysis system developed for this research can be used by independent raters in a consistent manner. Results Separate two-way (Group x Mode) repeated measures ANOVA's could have been computed for each of the eight language measures. Performing a separate analysis on each measure, however, would have prevented examination of potential interactions between Group, Mode, and the three dimensions of the analysis system (form vs. content, amount vs. organization, sentence vs. text). A five-way repeated measures analysis of variance (Gagnon, Roth, Carroll, Hofmann, & Spector, 1989) enabled examination of the relevant interactions while accounting for correlations among measures. The between-subjects factor was Group (four levels; LLI, AGE, LANG, READ). The within-subjects factors were Mode (two levels; SPOKEN, WRITTEN), Domain (two levels; FORM, CONTENT), Complexity (two levels; AMOUNT, ORGANIZATION), and Discourse (two levels; SENTENCE, TEXT). The family-wise alpha level was set at .05. Probability levels reflect adjustments for Greenhouse-Geisser estimates of epsilon. Standard score transformations were performed to enable statistical comparison because the scalar properties of the eight language indices differed. Spoken and written scores for each measure were combined across all four groups for the transformation process, and then separated to build a hierarchical database. Data transformation procedures precluded the possibility of significant main effects for, or interactions between, Domain, Complexity and Discourse. These 35 1303-1315 December 992 variables were free to interact with any of the levels of Group and Mode. Two four-way interactions, Mode x Domain x Complexity x Discourse [F(1, 36) = 12.99, p < .01] and Group x Mode x Domain x Complexity [F(3, 36) = 4.81, p < .01], subsumed both significant main effects, Group [F(3, 36) = 6.74, p < .01] and Mode [F(1, 36) = 10.21, p < .01], and seven significant lower-order interactions. Nonorthogonal tests of simple interactions (Bruning & Kintz, 1987) were computed to further examine the four-way interactions. Critical values for all a posteriori tests were established according to the simultaneous test procedure (STP) (Kirk, 1982) in order to hold the error rate for each collection of contrasts to that allotted to the overall error rate of the families involved in the significant omnibus interaction. Mode Differences The first question of interest concerned complexity differences between spoken and written stories. Recall that two types of complexity (form and content) were measured. Measures of language form included MLT-u, T-units per Story, Percent of Complex T-units, and Connectives per T-unit. Language content measures included Propositions per T-unit, Constituents per Story, Predicate Types per T-unit, and Percent of Dyadic Constituents. The simultaneous test procedure yielded a critical F(1, 36) of 6.8 (p < .02) for eight a posteriori tests of the Mode x Domain x Complexity x Discourse interaction. The data were initially organized to examine the relationships between Mode and Domain for all combinations of Discourse and Complexity levels. Mode differences between form and content were significant for textual organization (Connectives per T-unit, Percent of Dyadic Constituents) [F(1, 36) = 9.84] (Figure 1), but not textual amount (T-units per Story, Constituents per Story), sentential organization (Percent of Complex T-units, Predicate Types per T-unit), or sentential amount (MLT-u, Propositions per T-unit). In spoken narratives, values for the organization of textual form (Connectives per T-unit) were greater than values for the organization of textual content (Percent of Dyadic Constituents). The converse was true for written narratives. Links between propositions were marked more often in spoken stories than in written stories (23.2% vs. 19.7%). At the same time, written story constituents were more likely to be members of process dyads than spoken story constituents (19.5% vs. 12.3%). Although spoken narratives were superior to written narratives in the organization of textual form, written narratives were superior to spoken narratives in the organization of textual content. Next, data were organized to examine the relationships between Mode and Complexity for each combination of Discourse and Domain level. Mode differences between amount and organization were significant for textual content (Constituents per Story, Percent of Dyadic Constituents) [F(1, 36) = 16.18], but not textual form (T-units per Story, Connectives per T-unit), sentential content (Propositions per T-unit, Predicate Types per T-unit), or sentential form (MLT-u, Percent of Complex T-units). As depicted in Figure 2, values for amount of Gillam & Johnston: Spoken and Wntten Language -.1 each spoken story; but 2.4 dyads in each written story. Thus, in comparison to written stories, spoken stories contained a relatively high number of constituents that were not members of a dyad. Table 3 summarizes the characteristics of spoken and written narratives for each of the eight language measures.' Tukey honestly significant difference tests (df = 32, STP p < .006) were computed to analyze simple main effects. Spoken narrative values were significantly higher than written narrative values for all four amount measures (MLT-u, T-units per Story, Propositions per T-unit, and Constituents per Story). Written narrative values were significantly higher than spoken narrative values for only one measure, Percent of Dyadic Constituents. -,2 Group x Mode Interactions O Connectives per T-unit O % of Dyadic Constituents .3 W ;) ,., 0 0 o a) 1309 .2 .1 rA 5-. 0 E_ 5. 0 (U a) -.3 Spoken Written Mode FIGURE 1. Text level organization differences between form and content as a function of modality. textual content (Constituents per Story) were greater than values for organization of textual content (Percent of Dyadic Constituents) for spoken narratives. Just the opposite relationship obtained for written narratives. Spoken narratives were comprised of more constituents than written narratives (32 vs. 24.4). However, there were, on average, only 1.92 dyads in 0 # of Constituents 0 a o o 0 r. G) w ~0 W .3 - .2 - .1 - o of Dyadic Constiutents 0 - 0 rA (U (, kV 1/ :X -,1 - -.2 - The second research question concerned potential group differences for the relationship between spoken and written stories. The critical STP value for this collection of a posteriori contrasts was F(3, 36) = 9.72 (p < .001). The data were organized to examine Group and Mode differences for each combination of Domain, Discourse, and Complexity levels. Group by mode interactions were significant for form organization (Percent of Complex T-units, Connectives per T-unit) [F(3, 36) = 11.48], but not for form amount (MLT-u, T-units per Story), content organization (Predicate Types per T-unit, Percent of Dyadic Constituents), or content amount (Propositions per T-unit, Constituents per Story). Scheffe complex comparisons revealed that, across discourse levels, the LLI and READ groups significantly differed from the LANG and AGE groups in the way they used complex linguistic forms in written as compared to spoken stories [F(3, 32) = 11.61, p < .001]. Children in the LLI and READ groups used more complex linguistic forms (Percent of Complex T-units and Connectives per T-unit) in their spoken stories than in their written stories. Just the opposite mode relationship obtained for the LANG and AGE groups; these children used more complex linguistic forms in their written stories. Percent of Complex T-units accounted for 71% of the variance in the Group x Mode x Form organization interaction. Untransformed values for this measure are depicted in Figure 3. In the spoken mode, the AGE and LANG groups performed somewhat better than the LLI and READ groups. Differences were more pronounced in the written mode. In terms of speaking-writing relationships, the AGE and LANG groups used a greater percentage of complex correct T-units in their written stories as compared to their spoken stories. The READ and LLI groups evidenced the opposite trend. Their performance decreased in the written mode as compared to the spoken mode. -.3 I I Spoken Written Mode FIGURE 2. Text level content differences between amount and organization as a function of modality. 1 The polar symmetry of the transformed mean values is a direct consequence of the transformation process. Eighty scores for each measure (spoken and written scores for all four groups) were combined for standard score transformation, then were separated to build the hierarchical database. Spoken and written mean transformed values for each measure necessarily had to polarize for two reasons: (a) transformation set the mean of each measure at , (b) one half of the scores for each measure were spoken and the other half were written. 1310 Journalof Speech and Hearing Research 35 1303-1315 December 992 TABLE 3. Mean values for the eight spoken and written narrative meaures collapsed over groups. Written narratives Spoken narratives Measure Untransformed Transformed Untransformed Transformed MLT-u T-units per story % Complex T-units Connectives per T-unit Propositions per T-unit Constituents per story Predicate Types per T-unit % Dyadic constituents 10.7 24.5 35.3 23.2 3.5 32.0 2.2 12.3 .319 .386 .076 .159 .212 .274 .215 -.251 9.2 15.5 32.5 19.7 3.0 24.4 2.1 19.1 -.319 -.386 -.076 -.159 -.212 -.274 -.215 .251 To determine whether a particular type of complexity was responsible for these group differences, complex T-units were categorized into Coordinated Clauses, Subordinated Clauses, Clausal Complements, Relative Clauses, and Combined Structures. In both spoken and written modes, the AGE group evidenced a higher percentage of T-units that combined the various sorts of complex forms (spoken = 28.1%, written = 18.1%) than the LLI (spoken = 1.4%, written = 1.9%), LANG (spoken = 6.7%, written = 6.5%), and READ (spoken = 2.3%, written = 1.5%) groups. Values for the other three types of complex T-units were remarkably similar for both modalities across the four groups. Recall that T-units were counted as complex only if they were grammatically correct. The complex T-unit findings could have reflected grammatical accuracy more than syntactic complexity. That is, a high proportion of the incorrect T-units produced by the LLI and READ groups could have occurred on T-units that would have been complex. To o LLI * O[ LANG READ * AGE 60 55 .i 1 50 45 Discussion x 0 40 . 35 3 0 0 a a. examine this possibility, the percent of complex T-units with grammatical errors was compared to the percent of simple T-units with errors (Table 4). The LLI group differed from their age, language, and reading peers in three ways. First, for both spoken and written narratives, the LLI children evidenced more grammatical errors on both simple and complex T-units than children in any of the control groups. Second, complex T-units were more vulnerable to error, especially for the LLI group. Thus, discounting for error affected the LLI group more than the control groups, and complex T-units more than simple T-units. Third, the LLI group was the only group to evidence large differences between the percentage of errors in their spoken and written narratives. These differences were especially wide for complex T-units, which were in error 19.1% of the time in spoken narratives, but 78.3% of the time in written narratives. Consideration of the LLI group's propensity for using incorrect complex T-units may inform the interpretation of the Group x Mode x Form organization results. Figure 4 depicts the situation that would have occurred if the narratives had been produced in complete and correct form. If none of the subjects had evidenced grammatical errors, the distribution of the LLI group's complex T-units by mode would have resembled that of the AGE and LANG groups. Unlike their reading-ability matches, the LLI children would have produced more complex T-units in their written narratives than in their spoken narratives. One objective of this project was to investigate the differences between spoken and written narratives created by elementary school-age children. The two modes were compared by analyzing amount, organization, form, and content 30 25 20 TABLE 4. Percent of simple and complex T-units In error. 15 10 5 Simple T-units . Group Spoken narratives Written narratives Spoken narratives Written narratives LLI READ LANG AGE 12.3 8.7 3.7 2.0 19.1 10.4 7.3 5.2 46.1 13.4 11.7 5.4 78.3 12 10.4 7.2 . . Written Spoken Mode FIGURE 3. Group means for Percent of Complex T-units In spoken and written narratives. Complex T-units Gillam & Johnston: Spoken and Wnrtten Language O O LANG LLI * READ * AGE 60 55 'Z 50 4d X 45 e 40 E 0 35 0 30 X 25 10 5 - Spoken Written Mode FIGURE 4. Group means for Corrected Percent of Complex T-unlts In spoken and written narratives. properties of narratives at sentence and textual discourse levels. A picture of clear differences between spoken and written texts emerged from these analyses. Spoken narratives contained more sentences that were longer, but not necessarily more complex, than the written sentences. The key differences, however, concerned the way texts were organized. The significant textual organization interaction between language form and content indicated that, in comparison to written stories, spoken stories were more likely to contain explicit linguistic forms that indicated the relationships within and between contiguous T-units. These same stories, however, contained numerous process constituents that were left unconnected, leading to diffusely organized textual content. Written narratives contained a higher proportion of linked constituents and thus were denser texts. These findings suggest that the organizational strengths of spoken stories occurred on a local level, whereas the organizational strengths of written stories occurred on a global level. Could these results be artifacts of the testing or scoring methods that were employed? Recall, first, that and was not included in the Connectives per T-unit analysis. Perhaps the exclusion of and connectives was responsible for the Mode differences that were observed. To explore this possibility, ANOVA techniques were used to test for associations between Mode and Group and the percentage of clauses beginning with and. There were no significant Group main effects or Group x Mode interactions. The Mode main effect was significant. Consistent with Thompson's (1984) data on other kinds of connectives, and conjoined clauses occurred with greater frequency in spoken as compared to written narratives. If anything, inclusion of and would have height- 1311 ened the observed differences between spoken and written narratives on the Connectives per T-unit variable. Recall also that children were allowed to generate stories about pictures of their own choosing, and that static scene pictures were chosen more often than outdoor activity pictures. Perhaps stories that were created about static pictures of places and persons contained more scene-setting constituents, thereby adversely affecting the ratio of constituents to dyads. Note, however, that static and active pictures were chosen in equal proportions for spoken and written narratives, diminishing the possibility that picture selection accounted for the diffuse organization of spoken stories. Rather than serving as methodological artifacts, our findings do indeed imply that the organizational strengths of spoken stories occurred on a "local" level, whereas the organizational strengths of written stories occurred on a "global" level. Why might this be so? One explanation for this phenomenon centers around listener/reader needs. Elaborating the ideas of Slobin (1977), we might argue that writers and speakers contend with different audience constraints as they create texts. Readers and listeners experience certain processing limitations that are a function of communicative mode. The temporary nature of the speech signal places demands on listeners' working memory. Listeners must process the speech signal, relate what they are hearing to what they have already heard, and organize the whole, all while new information continues to arrive. These demands are not present for readers. Due to the permanent nature of written signs, readers may choose to stop receiving information and reflect on what they have just read. They can even choose to receive the same information over again as they construct their own interpretations of a text. Faced with the memory constraints of their listeners, speakers may need to make the relationships between successive ideas immediately clear. Linguistic connectives would serve this purpose well, providing local cues to the speaker's train of thought. Writers, on the other hand, knowing that readers can stop to think about the way content is organized, may not need to provide these local cues. They can rely on text structure to convey their organizational scheme. Perhaps the children in this study were either consciously or unconsciously aware of the relationships between mode of expression and audience processing needs, and adjusted their text production strategies accordingly. There are at least two difficulties with this explanation. First, it is rather hard to imagine that children between the ages of 6 and 12 would have enough reading experience and a sufficiently developed theory of mind to arrive at such a complex vision of audience needs. Secondly, this explanation ignores our finding that speakers included more contextual information and/or more single-process constituents that had little to do with the nexus of their plots, a fact that certainly would not aid listener comprehension. An alternative explanation may be found by focusing on language production processes themselves, rather than on audience needs. Consider the on-line processing demands of spoken narration. As proposed by Beaugrande (1984), speakers begin with purposes for communicating that take 1312 Journalof Speech and Hearing Research the form of discourse goals. Text production ideas spring from these purposes. As ideas are organized into conceptual configurations, they are further elaborated to fit the speaker's original goals. At the point of expression, content is mapped onto linguistic forms that best serve the semantic and pragmatic needs of the moment. Although much remains to be understood about these processes, it seems certain that these phases of thought and linguistic decision making are not conducted in a linear fashion, but co-occur and interact. This being so, the speaker's challenge is to both talk and organize future utterances at the same time. The on-line demands of written narration are considerably weaker once writers have a reasonable amount of control of the spelling and handwriting components of writing. The slower expression time, the ability to reread what has already been written, and the lack of pressure for uninterrupted flow make it possible for a writer to selectively focus on one or another aspect of text production. Less needs to be accomplished at once. This view of processing differences allows a coherent account of both the form and content findings. By this account, the precise organization of language form in spoken narratives is a consequence of the same conceptual and processing problems that lead to the diffuse organization of language content. Thinking about the linguistic expression phase of narration, global conceptual configurations are organized into networks of content relationships that are realized linguistically as sequences of propositions. Speakers may try to keep track of these conceptual relationships by overtly marking them as they create clausal structures that best express their intended meanings. In a sense, speakers use connectives to placehold their intentions. The connective implies what has gone before as well as what is yet to come. The scope of the connective, however, is often limited to nearby propositions. Use of this strategy would have little influence on monitoring the organization of the larger text. Our data suggest that elementary-grade speakers, even those who have discovered the processing value of connectives, have difficulty creating and/or implementing a coherent overall narrative plan. Writing, in contrast, affords children some processing advantages. Even if distracted by the mechanics of a new task (Shatz, 1984), writers experience less need for using placeholders as organizational aids since they can review their evolving text at will. In our data, these advantages were manifested by a lower proportion of connectives and a higher proportion of linked constituents in written stories as compared to spoken stories. The second main objective of this project was to compare the characteristics of spoken and written narratives produced by LLI children, and to determine whether their modality relationships were similar to those of normally developing children. By using three comparison groups, we hoped to attain some insight into the nature of any group differences that were found. Group differences in the relationship between spoken and written narratives were not significant for indices measuring the amount of language form, the organization of language content, or the amount of language content. Group differences did emerge for measures relating to the organization of language form, particularly for the syntactic measure 35 1303-1315 December 1992 Percent of Complex T-units. 2 For the AGE and LANG groups, written narratives contained a higher percentage of correct complex T-units than did spoken narratives. For the LLI and READ groups, it was the spoken narratives that had the higher values on this measure. These group differences did not reflect unique abilities with a particular type of complex T-unit. Children in all four groups produced essentially similar proportions of each complexity type. Consistent with the results of numerous prior studies (Johnston, 1988; Johnston & Kamhi, 1984; Johnston & Schery, 1976; Ludlow, 1980; Merritt & Liles, 1987), these findings indicate that language form does indeed pose special problems for school-age language-impaired children. Two aspects of the current data highlight this fact. First, when matched for spoken language abilities with a group of normally achieving children, the LLI group performed appreciably worse on a measure of complex sentence usage, but looked similar on measures of language content. Second, the LLI group produced a uniquely large percentage of grammatically unacceptable sentences, especially in their written narratives. Their formal errors were not limited to attempts to use difficult organizational structures. The LLI children made numerous grammatical errors on both simple and complex sentences. Although some recent reports have identified young children whose language problems seem to resolve (Paul, 1991; Rescorla, 1990; Thai, 1989), these findings clearly demonstrate that specific difficulty with language form can persist well into later childhood. The findings on group differences also pertain to mechanisms underlying the relationships between spoken language, reading, and written language in LLI children. Initially, our findings seemed to suggest a relationship between reading level and the relative complexity of spoken versus written texts. Although the LLI children had nonverbal reasoning skills that were comparable to those of the agematched children, and could imitate sentences as well as the language-matched children, it was not these abilities that predicted differences in their spoken and written sentences. These differences were best predicted by their reading level. Consistent with Stanovich (1986) and Wells (1986), these results seem to suggest that reading serves as a valuable input for sentence-level language acquisition during the school years. Children who read at higher levels read books that contain a greater proportion of complex sentences (Dawkins, 1975). Certainly, the experience of reading complex sentences should provide useful information about the relative complexity of formal, written language. When we reanalyzed the complex T-unit data, ignoring grammatical errors, the importance of reading level was called into question. Recall that the LLI group showed a 2 Because of the size of the subject sample in this study, some degree of caution should be exercised in the interpretation of these results. The most obvious problem associated with studying four groups of 10 children is the loss of power, which results in a larger probability of making a type II error. Such may be the case here, specifically in relation to the insignificant Group x Mode x Domain x Complexity x Discourse interaction. In this case, very little power remained for rejecting the null hypothesis that the LLI group did not differ from one or more of the control groups on mode, domain, complexity, and discourse level when, in fact, this hypothesis may have been false. Larger sample sizes or a more restricted number of variables might have led to a different result. Gillam & Johnston: Spoken and Written Language unique propensity for producing ill-formed T-units, and this trend was especially dramatic for written complex forms. If we credit their intentions instead of their accomplishments, the LLI children resembled the age- and language-matched children rather than the reading-matched children in terms of the directionality and slope of the relationship between the spoken and written language sentential complexity. For the AGE and LANG groups, as for adults (Beaugrande, 1984; Chafe, 1985; Johnson, 1991; Ochs, 1979), written sentence complexity exceeded spoken sentence complexity. If we forgive their grammatical errors, the LLI children also evidenced this "mature" profile; their reading-matched peers did not. The key to understanding these findings seems to lie in the variables that define the groups. The LLI students were fourth graders who spoke like third graders and read like second graders. Their grammatical errors reveal a fundamental weakness with linguistic form, perhaps exacerbated by the relatively new demands of the writing task (Shatz, 1984). But note first that even though they were functioning at literacy and spoken language levels that would be appropriate for 7- and 8-year-olds, in terms of amount of formal educational experience, they were 10-year-olds. Despite their difficulty controlling written language form and their impoverished language repertoire, 4 years of schooling may have provided the LLI children with important information about the relative complexity of written language. Note also that the three groups with the more mature knowledge of written language are similar in having at least the spoken language level of an 8-year-old. Together, these facts suggest that spoken language development and/or amount of language experience can play an important role in the development of mature realizations about speaking-writing differences. We can imagine the following scenario for such learning. As children listen to and tell stories, they create literacy scripts that contribute to reading and writing development. With sufficient exposure to spoken and written narratives, even "poor readers" and children struggling with the mechanics of writing may come to know how particular linguistic features are conveyed in speech as compared to writing, and how written texts can be organized differently from spoken texts. It appears that listening and speaking experiences do serve as important avenues to information about literacy. The performance of the LLI children in this study makes it clear that reading, by itself, is not a necessary means to at least one aspect of literacy knowledge. Conclusion It is well documented that language-impaired children have difficulty learning to read (Aram, Ekelman, & Nation, 1984; Gillam & Johnston, 1985; Kamhi & Catts, 1989). Do their language-leaming difficulties place them at a special disadvantage for learning about the nature of written language? The findings from this study indicate that the relationship between spoken and written language performance for the LLI children was similar to that of their age and languagelevel peers for all but one of the eight language measures. 1313 While LLI children may indeed develop some aspects of literacy at a slower than normal rate, the results reported herein suggest that the relationships between spoken and written language in this population may not be quantitatively or qualitatively unusual to any educationally significant degree. The most significant impact of language impairment on written language development may be primarily confined to grammatical acceptability. The LLI children produced a much higher percentage of grammatically unacceptable T-units than any of their matched peers, a difference that was especially dramatic for written narratives. These formal errors certainly reflect pervasive difficulties with syntax in general. More interestingly, they may also reflect an emergent appreciation of the nature of the written mode. As they experience the weaker on-line requirements of the writing process, LLI children may recognize the potential for creating coherent text and focus on that aspect of narrative, slighting grammaticality. The challenge for educators and other readers is to recognize the sophistication of the resultant text despite its formal deficiencies. Acknowledgments Preparation of this article was supported in part by a grant from NIDCD to the first author. Thanks to the following individuals at Converse County School District #1, Douglas, Wyoming, who assisted with data collection: Margaret Boersma, Dave Fetter, Ramona Gazewood, Bob Pesicka, Liz Groff, and Wayne Porter. The contributions that Jerry Harste, Bill McKay, Rita Naremore, and Marylou Gelfer made to this project are especially appreciated. A. Lynn Williams, Barbara Fazio, Shirley Patterson, Marilyn Nippold, and two anonymous reviewers made insightful editorial suggestions about earlier versions of the paper. Partial reports of this research were presented at the Symposium for Research in Child Language Disorders (1989) and the Annual Convention of the American Speech-Language-Hearing Association (1989). References Antinucci, F., & Parlsl, D. (1976). Essentials of grammar. New York: Academic Press. Aram, D., Ekelman, B., & Nation, J. (1984). Preschoolers with language disorders: 10 years later. Journal of Speech and Hearing Research, 27, 232-244. Baghban, M.(1984). Our daughter learns to read and write. Newark, NJ: International Reading Association. Barthes, R. (1967). Writing degree zero (A. Lavers & C. Smith, Trans.). London: Cape. Beaugrande, R. (1984). Text production: Toward a science of composition. Norwood, NJ: Ablex. Botvln, G., & Sutton-Smith, B. (1977). The development of structural complexity in children's fantasy narratives. Developmental Psychology, 13, 377-388. Brown, R. (1973). A first language. Cambridge, MA: Harvard University Press. Brown, L., Sherbenou, R.J., & Johnseon, S. K. (1982). Test of Nonverbal Intelligence. Austin, TX: Pro-ed. Bruning, J. L., & Klntz, B. L. (1987). Computational handbook of statistics (3rd ed.). Glenview, IL: Scott, Foresmann. Burtis, P.J., Berelter, C., Scardamalla, M., & Tetroe, J. (1983). The development of planning in writing. In C. Wells & B. Kroll (Eds.), Exploration of children's development in writing (pp. 153174). Chichester, England: John Wiley. Calkins, L. M. (1983). Lessons from a child: On the teaching and learning of writing. Portsmouth, NH: Heinemann. 1314 Journalof Speech and Hearing Research Chafe, W. (1985). Linguistic differences produced by differences between speaking and writing. In D. Olson, N. Torrance, & A. Hildyard (Eds.), Literacy, language and learning: The nature and consequences of reading and writing (pp. 105-123). New York: Cambridge University Press. Chafe, W. L., & Danlefewicz, J. (1987). Properties of spoken and written language. In R. Horowitz & S. J. Samuels (Eds.), Comprehending oral and written language (pp. 61-93). Orlando, FL: Academic Press. Culler, J. (1975). Structuralist poetics: Structuralism, linguistics and the study of literature. London: Routledge and Kegan Paul. Dalute, C. (1989). Play as thought: Thinking strategies of young writers. Harvard Educational Review, 59, 1-23. Dawkins, J. (1975). Syntax and readability. Newark, DE: International Reading Association. Derrida, J. (1981). Positions (A. Bass, Trans.). Chicago: University of Chicago Press. Dunn, L. M., & Markwardt, F.C. (1970). Peabody Individual Achievement Test. Circle Pines, MN: American Guidance Service. Dyson, A. H. (1983). The role of oral language in early writing processes. Research in the Teaching of English, 17, 1-30. Dyson, A. H. (1989). Multiple worlds of child writers: Friends learning to write. New York: Teachers College Press. Dyson, A. H. (1991). Viewpoints: The word and the world-reconceptualizing written language development or do rainbows mean a lot to little girls? Research in the Teaching of English, 25, 97-124. Erftmler, T., & Dyson, A. (1986). "Oh, ppbbt!": Differences between the oral and written persuasive strategies of school-aged children. Discourse Processes, 9, 91-114. Ferrelro, E., & Teberosky, A. (1982). Literacy before schooling. Portsmouth, NH: Heinemann. Gagnon, J., Roth, J., Carroll, M., Hofmann, R., & Spector, P. (1989). SuperANOVA: Accessible general linear modeling. Berkeley, CA: Abacus Concepts. Gllam, R.B., & Johnston, J.R. (1985). Development of print awareness in language-disordered preschoolers. Journal of Speech and Hearing Research, 28, 521-526. Hammill, D. D. (1985). Detroit Tests of Learning Aptitude-2. Austin, TX: Pro-ed. Harste, J., Woodward, J., & Burke, C. (1984). Language stories & literacy lessons. Portsmouth, NH: Heinemann. Hunt, K. (1970). Syntactic maturity in school children and adults. Monographs of the Society for Research in Child Development, 35 (Serial No. 134). Johnson, E. D. (1991). The handbook of good English. New York: Washington Square. Johnston, J. (1986, November). Language sample analysis as an assessment tool. Clinical workshop presented at the annual meeting of the American Speech-Language-Hearing Association, Detroit, MI. Johnston, J. (1988). Specific language disorders in the child. In N. Lass, L. McReynolds, J. Northern & D.Yoder (Eds.), Handbook of speech-pathology and audiology. Philadelphia: B. C. Decker. Johnston, J., & Kamhl, A. (1984). Syntactic and semantic aspects of the utterances of language-impaired children: The same can be less. Merrill-Palmer Quarterly, 30, 65-86. Johnston, J., & Schery, T. (1976). The use of grammatical morphemes by children with communication disorders. In D. Morehead & A. Morehead (Eds.), Normal and deficient child language (pp. 239-258). Baltimore: University Park Press. Kamhl, A. G., & Catts, H. W. (1986). Toward an understanding of developmental language and reading disorders. Journal of Speech and Hearing Disorders, 51, 337-347. Kamhl, A. G., & Catt, H. W. (1989). Reading disabilities: Terminology, definitions, and subtyping issues. In A. Kamhi & H. Catts (Eds.). Reading disabilities: A developmental language perspective (pp. 35-66). Boston, MA: College-Hill Press. Kamhl, A. G., Catta, H. W., &Mauer, D. (1990). Explaining speech production deficits in poor readers. Journal of Learning Disabilities, 23, 632-636. Kamhl, A. G., & Johnston, J. R. (in press). Semantic assessment: determining propositional complexity. Best Practices in School Speech/Language Pathology. 35 1303-1315 Dece-mLer 99 Kirk, R. E. (1982). Experimental design (2nd ed.). Monterey, CA: Brooks/Cole Publishing. Kroll, B.(1981). Developmental relationships between speaking and writing. In B. Kroll and R. Vann (Eds.), Exploring speaking-writing relationships: Connections and contrasts (pp. 32-54). Champaign, IL: National Council of Teachers of English. LaBerge, D., & Samuels, J. (1974). Toward a theory of automatic information processing in reading. Cognitive Psychology, 6, 293323. Ludlow, C. (1980). Children's language disorders: Recent research advances. Annals of Neurology, 7, 497-507. Maxwell, S. E., & Wallach, G. P. (1984). The language-leaming disabilities connection: Symptoms of early language disability change over time. In G. Wallach and K. Butler (Eds.), Language learning disabilities in school-age children (pp. 15-34). Baltimore: Williams & Wilkins. Merritt, D. D., & Liles, B. Z. (1987). Story grammar ability in children with and without language disorder: Story generation, story retelling, and story comprehension. Journal of Speech and Hearing Research, 30, 539-552. Miller, J., & Chapman, R. (1984). Systematic analysis of language transcripts. Madison: University of Wisconsin. Naremore, R. C., & Hopper, R. (1990). Children learning language: A practical introduction to communication development (3rd ed.). New York: Harper &Row. Norris, J. A., & Bruning, R. H.(1988). Cohesion in the narratives of good and poor readers. Journal of Speech and Hearing Disorders, 53, 416-424. Ocha, E. (1979). Planned and unplanned discourse. In T. Govan (Ed.), Discourse and syntax (pp. 51-80). New York: Academic Press. Paul, R. (1991). Profiles of toddlers with slow expressive language development. Topics in Language Disorders, 11 (4), 1-13. Perera, K. (1984). Children's writing and reading. London: Basil Blackwell. Perfstti, C. (1985). Reading ability. New York: Oxford University Press. Rescorla, L. (1990). Outcomes of expressive language delay. Paper presented at the Symposium for Research in Child Language Disorders, Madison, WI. Rowe, D.W., & Harste, J.C. (1986). Reading and writing in a system of knowing: Curricular implications. In M. Sampson (Ed.), The pursuit of literacy: Early reading and writing (pp. 126-144). Dubuque, IA: Kendall/Hunt. Rubin, D. (1982). Adapting syntax in writing to varying audiences as a function of age and social cognitive ability. Journal of Child Language, 9, 497-510. Saussure, F. (1959). Course in general linguistics. In C. Bally &A. Sechehaye (Eds.), (W.Baskin, Trans.). New York: The Philosophical Library. Scarborough, H. S. (1990). Very early language deficits in dyslexic children. Child Development, 61, 1728-1743. Shatz, M. (1984). A song without music and other stories. In D. Schiffrin (Ed.), Georgetown University Round Table on Language and Linguistics 1984 (pp. 313-324). Washington, DC: Georgetown University Press. Slobin, D. (1977). Language change in childhood and in history. In J. Macnamara (Ed.), Language learning and thought (pp. 185214). New York: Academic Press. Smith, F. (1982). Writing and the writer. New York: Hold, Rinehart and Winston. Snyder, L.S., & Downey, D.M. (1991). The language-reading relationship in normal and reading-disabled children. Journal of Speech and Hearing Research, 34, 129-140. Stahl, S., & Erickson, L. (1986). The performance of third grade leaming-disabled boys on tasks at different levels of language: A model-based exploration. Journal of Learning Disabilities, 19, 170-180. Stanovlch, K. E. (1986). Matthew effects in reading: Some consequences of individual differences in the acquisition of literacy. Reading Research Quarterly, 21, 360-407. Stark, R. E., & Tallal, P. (1988). Language, speech, and reading disorders in children: Neuropsychological studies. Boston: Col- Gillam & Johnston: Spoken and Written Language lege-Hill. Sutton-Smith, B., Botvin, G., & Mahony, D. (1976). Developmental structures in fantasy narratives. Human Development, 19, 1-13. Thai, D. (1989). Language and gesture in late talkers. Paper presented at the biennial meeting of the Society for Research in Child Development, Kansas City, MO. Thompson, S.A. (1984). 'Subordination' in formal and informal discourse. In D. Schiffrin (Ed.), Georgetown University Round Table on Language and Linguistics 1984 (pp. 85-94). Washington, DC: Georgetown University Press. Vygotsky, L. (1978). Mind in society: The development of higher cognitive processes (M. Cole, V. John-Steiner, S. Scribner, & E. 1315 Souberman, Eds.). Cambridge, MA: Harvard University Press. Wells, G. (1986). The meaning makers: Children learning language and using language to leam. Portsmouth, NH: Heinemann. Received September 16, 1991 Accepted March 6, 1992 Contact author: Ronald B. Gillam, PhD, Department of Speech Communication, University of Texas at Austin, Austin, TX 787121089. Appendix Contextual Constituent Labels and Their Definitions Contextual Event The occurrence or event that predominates subsequent character action. For example, a story about the adventures associated with finding a shipwreck began, "John and Fred went diving . . ." Participants Animals or humans that behave, sense, or exist in a story. Experiencers identified only by pronouns are not counted as participants. Location Specific reference to the site of a contextual event, dyadic process, or single process. Time Specific temporal reference to the period of existence or duration of a contextual event, dyadic process, or single process. References such as once, and then, and after that were not counted since they did not include a specific temporal term (o'clock, day, hour, year, month). Additional Contextual Information Information about features of the environment including distances, surroundings, accompanying circumstances, and roles that add to the listener/reader's understanding of the context of the story. Dyadic Constituent Labels and Their Definitions Problem--Result Villainy-Villainy nullifiedNillainy not nullified Evil or wicked act committed on a protagonist. Lack-Lack liquidated/Lack not liquidated A problematic shortage or deficiency. Deception-Deception revealed/Deception not revealed A misleading act or statement that is intended to cause a participant to think or act wrongfully. Threat-Threat nullified/Threat not nullified A potential source of danger, harm, or distress. Plan-Plan carried out/Plan not carried out A scheme (thought, intention) for doing something. Attack-Counterattack/Wound/Kill/Flee To strike or shoot at a participant with intent to harm. Injury-Recovery/Death Physical harm or damage to a participant. Pursue-Capture/Escape/Release To chase in order to capture or harm. Search-Found/Not found To look for something that is needed or that has been lost. Difficulty-Difficulty rectified/Difficulty not rectified A problem that is hard for a participant to contend with that does not correspond to one of the above categories. Single-Process Constituent Labels and Their Definitions Announcement Inquiry Response A verbal statement. A question intended to seek information. A mental or verbal acknowledgement of a previous event, announcement, or inquiry. Reaction A physical activity that has been influenced by a previous event, announcement, or inquiry. Consequence The outcome or effect of a previous process or set of processes. Fortuitous Occurrence An accidental event that is advantageous to a participant. Departure When a participant leaves a particular location. Return When a participant comes back to a former location. Material Activity A 'doing' process that does not fit into any of the previous categories. Narrator Elaboration An explanation of a participant's behavior or circumstance that lies outside the story. Conclusion Narrator's statement which brings the story to a logical end.