Carlisle Fleming 2003

SCIENTIFIC STUDIES OF READING, 7(3), 239-253 Copyright © 2003, Lawrence Erlbaum Associates, Inc Lexical Processing of Morphologically Complex Words in the Elementary Years Joanne F. Carlisle University of Michigan Jane Fleming University of Missouri at St. Louis This study explores emerging lexical processes that may be the foundation for children's acquisition of morphological knowledge and the relation of these processes to reading comprehension First and third graders were given two tasks involving lexical analysis of morphologically complex words. Two years later, they were given a measure of processing derived words in sentence contexts and a test of reading comprehension. The results support the view that the development of morphological processing in the elementary years might depend on access to representations of full forms, base forms, and affixes. Further, semantic and syntactic knowledge of morphemes was related to morphological processing of sentences and contributed to reading comprehension in the late elementary years. The vocabulary explosion that begins around fourth grade is in large part attributable to the increase in affixed words children know (particularly derivatives; Anglin, 1993; White, Power, & White, 1989), but children's learning of morphologically complex words begins much earlier. Behavioral data show that preschool children are acquiring morphological knowledge. They give evidence of implicit understanding of useful ways that morphemes can be combined to express particular meanings. For example, Clark’s (1995) children made statements like, “You be the stroyer, Dad” (p. 401). By the late elementary years, children's ability to segment and manipulate morphemes within complex words relates to word reading (Fowler & Liberman, 1995; Singson, Mahony, & Mann, 2000) and reading com- — Requests for reprints should be sent to Joanne F. Carlisle, Professor, Educational Studies, University of Michigan, School of Education, 610 East University Avenue, Ann Arbor, MI 48103. E-mail: jfcarl@umich.edu. 240 CARLISLE AND FLEMING prehension (Carlisle, 2000). However, there are many questions about how children learn to process words with regard to morphemic structure. How can we characterize the emergent knowledge of morphemes and morphological processing in this period of development? Does it have long-term implications for children's language and reading comprehension? The study reported herein was designed to address these questions. LINGUISTIC PROCESSING OF MORPHOLOGICALLY COMPLEX WORDS For adults morphological structure facilitates lexical processing of written words (e.g., Feldman & Andjelkovic, 1992; Marslen-Wilson, Tyler, Waksler, & Older, 1994). Models of morphological processing of words generally posit coactivation and processing of submorphemic components (phonological and orthographic features), morphemic components (base morpheme, affixes, and full form), and concepts or meaning (Chialant & Caramazza, 1995; Schreuder & Baayen, 1995; Taft & Zhu, 1994). Of relevance here is the question of how children develop mental representations of morphemes, particularly those that they encounter embedded in words (i.e., bound morphemes). Schreuder and Baayen (1995) proposed both a model of processing of complex words and an explanation of how children learn morphological processing. Their model posits a mechanism for carrying out symbolic computations on representations that have become available through activation. These computations take place in three stages. The first stage (segmentation) entails mapping of the speech input onto form-based access representations of full as well as bound forms (affixes, bound stems). The second stage (licensing) entails determining whether representations that have become coactive can be integrated on the basis of their subcategorization properties (e.g., syntactic role). The third stage involves the computation of the lexical representation of the complex words from the constituents; this lexical representation entails processing of semantic and syntactic information. According to Schreuder and Baayen, the crucial component of children's morphological learning is their development of mental representations of bound morphemes (e.g., prefixes and suffixes). Through the affix discovery principle, children monitor the mental lexicon for correspondences between form (phonological or orthographic strings that co-occur regularly in the language, given posi tional constraints) and meaning. This is a two-stage process. The first stage involves detecting patterns of coactivation of developing semantic representa tions. If a pattern is detected (e.g., -er in teacher, runner), a concept node is created for this pattern, and a corresponding representation at the access level begins to develop. Thereafter, the representation can be accessed as it is encountered in words, and it becomes invested with semantic and syntactic information (e.g., -er signify- 241 LEXICAL PROCESSING ing “one who—”). Evidence that -er is represented apart from specific words comes from neologisms children spontaneously produce, as in the earlier example of stroyer (Clark, 1995). On the basis of Schreuder and Baayen's model, we should be able to observe certain features of children's morphological learning. First, we should be able to find evidence that children gradually develop representations for bound morphemes, particularly those they encounter frequently. Second, we should find developmental changes in the semantic and syntactic knowledge associated with the representations, and this would allow for integrative processing of form and meaning (Taft & Zhu, 1995). Third, lack of familiarity with the full form or any of the constituent morphemes should result in incomplete morphological processing. To date, studies of morphological processing in the early school years have focused predominantly on recognition of word structure (i.e., decomposition). Jones (1991) investigated the underlying representation of morphophonemic segments among first graders on tasks that involved leaving out a part of words such as grandfather or leaves (i.e., decomposition) and discussion of the meaning of the remaining part. She found that compared to language-delayed first graders, language-advanced children gave stronger evidence that they had representations of morphophonemic segments. Somewhat similar findings come from a study by Rubin (1988). Exploring the relation of morphological development and spelling, Rubin gave kindergartners and first graders one- and two-morpheme words, asking whether each word contained a smaller word that was similar in meaning (e.g., "Is there a little word in pinned that means something like pinned?"). Foils (e.g., wind) were similar in phonological structure but contained one morpheme. The child could not be successful on the task if he or she could access only the meaning and phonological representation of the whole word. In terms of Schreuder and Baayen’s model, the child might respond “no” to both pinned and wind if he or she could not decompose the words or did not have representation of the past tense. Rubin’s results showed that first graders who performed well on this task also tended to have superior implicit morphological knowledge. Derwing and Baker (1979) also investigated understanding of form-meaning relations by asking children and adults whether one word “came from” another (e.g., does bear come from beard?) The results showed that children in Grades 3 through 6 tended to be biased by similarities of sound, without regard for differences in meaning, whereas older students were more likely to base their judgments on both phonological and semantic similarities. However, for the younger children, some of the words were relatively uncommon or complex in meaning, a condition that, given the previous theory, would likely impede morphological processing. These and other studies (e.g., Carlisle, 1995; Fowler & Liberman, 1995) show that early elementary children can decompose phonologically transparent forms but provide little insight into their understanding and use of complex words. In contrast, Anglin (1993) studied access to meanings of complex words. He asked 242 CARLISLE AND FLEMING first, third, and fifth graders to explain the meaning of words and to use them in a sentence, including morphologically simple (e.g., closet) and complex (e.g., still- ness) words. Results showed significant growth in knowledge of morphologically complex words by grade level. Anglin’s interview technique provides a way to examine aspects of the morpho- logical processing model proposed by Schreuder and Baayan. Of particular inter- est is the final stage, called combination, which involves the integrative processing of the semantic and syntactic components of the constituent morphemes. Through scrutiny of the children’s responses, Anglin found that children could sometimes figure out the meaning of a word they had never heard, a process he called morpho- logical problem solving. They did this by recognizing the morphemic constituents of a word and using that knowledge to infer the meaning of the whole word (e.g., inferring the meaning of knotless from knot and -less). He found few instances of morphological problem solving in first grade but significant increases in Grades 3 and 5. Clearly, the process works far better with unfamiliar words that are semanti- cally transparent, like knotless, than those that are semantically opaque, like appli- ance, as has been found in studies of lexical processing of adults (e.g., Marslen-Wilson et al., 1994) According to Schreuder and Baayen, analysis of unfamiliar complex words for meaning also depends on the earlier stages in the process (decomposition and re- censing). These stages could not be accomplished without access representations of affixes, base form, and full form. For example, children might think they do not “know” treelet because they do not recognize the full form. RELEVANCE FOR SENTENCE PROCESSING AND READING COMPREHENSION By the late elementary years, children who are sensitive to morphological structure have a way to infer the meanings of unfamiliar morphologically complex words (Nagy & Anderson, 1984). They are also likely to be good readers (Carlisle, 2000; Fowler & Liberman, 1995; Windsor, 2000). Developmental changes between fourth and eighth grade in understanding the meanings and grammatical roles of af- fixes may be a prerequisite for inferring the meanings of unfamiliar words during reading (Tyler & Nagy, 1989; Wysocki & Jenkins, 1987). Breadth and depth of word knowledge moderately correlate with reading com- prehension (Anderson & Freebody, 1985), and this word knowledge includes mor- phologically complex words. However, gradual development of syntactic and semantic knowledge of morphemes is a potentially important additional contribu- tor to comprehension, as this is likely to provide a basis for inferring the meanings of unfamiliar words and interpreting word meanings in texts (Rego & Bryant, 1993) Thus, we explored the possibility that lexical analysis of morphologically 243 LEXICAL PROCESSING complex words in the early elementary years contributes to morphological pro- cessing in sentences and to reading comprehension in the third and fifth grades. SUMMARY AND RESEARCH QUESTIONS The purpose of this study was to see whether we would find evidence to support Schreuder and Baayan’s explanation of the process by which children come to ac- cess and process morphemes, given tasks of lexical analysis involving morphologi- cally complex words. One of the tasks was an adaptation of the word analysis task developed by Rubin (1988), and the second was Anglin’s (1993) word interview. In addition, we selected three derived words from Anglin’s interview for close analy- sis. These words (knotless, stillness, and treelet) varied in terms of familiarity of the full form and the affix. Analysis of responses to these words was used to study the extent to which decomposition and analysis of form and meaning depended on fa- miliarity with word parts and the extent to which first and third graders differed in the semantic and syntactic knowledge of the morphemes. To explore the relevance of children’s lexical analysis of complex words to lan- guage and literacy acquisition, we administered the Test of Morphological Struc- ture (TMS), an oral task that required decomposition or production of derived forms (Carlisle, 2000), and a standardized test of reading vocabulary and compre- hension. Our primary questions were as follows: (a) Is evidence from first and third graders’ lexical analyses of morphologically complex words compatible with the development of morphological processing proposed in Schreuder and Baayen’s (1995) model? (b) Is performance on the Year 1 lexical analysis tasks significantly related to the Year 3 task involving morphological processing in sentence con- texts? (c) Does performance on the Year 1 lexical analysis measures significantly predict Year 3 reading vocabulary and comprehension? METHOD Participants The participants included 34 third graders and 26 fifth graders. The younger group included 18 boys and 16 girls (mean age = 6.7 years in Grade 1); the older group in- cluded 11 boys and 15 girls (mean age = 8.9 years in Grade 3). The children at- tended a private school in a suburban community outside Chicago. All children were native speakers of English. Receptive vocabulary was assessed at the start of the study (Year 1) with the Peabody Picture Vocabulary Test–Revised (Dunn & 244 CARLISLE AND FLEMING Dunn, 1981). For first graders, the mean standard score (SS) was 113.6 (SD = 11.9) and for third graders, the mean SS was 113.1 (SD = 15.5). Materials In Year 1, the children were given two measures of morphological processing of words: 1. The Word Analysis Test (WAT). Adapted from Rubin (1988), this test was designed to determine whether children could distinguish two-morpheme from one-morpheme words. The children were asked, “Is there a little word in_____that means something like _____?” Successful performance required both pho- nological segmentation and attention to whether the segment extracted was a word and whether it was related in meaning to the presented word. Test items included six words with the suffix -y (e.g., sunny), six agentives and instrumentals (e.g., teacher), and six past tense verbs (e.g., pulled), suffixes that should be familiar to first graders (Berko, 1958; Clark, 1995). Each of these words was matched to a monomorphemic word with the same final sound (e.g., silly for hilly). There were four training items (one modeled by the examiner and three for practice and feed- back). Performance was scored using Rubin’s system of giving credit only when the child responded correctly to both items in a pair (e.g., “yes” to hilly and “no” to silly). 2. Definition. This measure was drawn from children’s performance on the Test of Absolute Vocabulary Knowledge (TAVK; Anglin, 1993). Anglin’s test was designed to estimate the number of dictionary entries children knew. He de- veloped his word list by systematic sampling of words from Webster’s Third New International Dictionary of the English Language. Words types included root words, inflected words, derived words, literal compounds, and idioms (see Anglin, 1993, for a detailed explanation). The task involved an interview in which the child was asked to tell what a given word means. When the child could not explain the meaning and use the word in a sentence, he or she selected a meaning from a set of four possible meanings. For this study testing ended when the child did not know the meaning of seven successive words, five of which also yielded incorrect multi- ple-choice selections. The interview responses were tape-recorded, transcribed, and scored using Anglin’s criteria for what counted as an acceptable definition. However, the defi- nition and sentence conditions were combined to yield one score, hereafter called Definition. The Definition score consisted of instances in which a child defined a base morpheme in a morphologically complex word (inflected and derived forms and compounds) and used the complex word correctly in a sentence. (The multi- ple-choice responses were not included in this measure.) Interscorer reliability, de- termined by comparing Definition credit awarded by the two researchers, was LEXICAL PROCESSING 245 94.9% for the first grade and 94.8% for the third grade. Differences were resolved by discussion. Analysis of three words from the interview (knotless, stillness, and treelet) was carried out to provide further insights into children's processing of complex words. The three words had base morphemes generally familiar to first and third graders. The suffixes differed in this respect: -less is likely to be understood as meaning "without"; -ness is likely to be familiar but is abstract in meaning, and children would not be likely to understand its grammatical role; -let is not gener- ally familiar to first and third graders (see Windsor, 1994). First and third graders' explanations of these words were scored on the following criteria: Decomposition: Does the child mention the base morpheme spontaneously? If so, this would be ev- idence of segmentation of the morphemic structure of the word. Definition of the base morpheme: Does the child define the base word? This would provide evi- dence that the child has a mental representation for the base morpheme. Sentence: Does the child use the word appropriately in a sentence? If so, the child would be likely to have encoded semantic and syntactic information associated with the morphemes. The two researchers scored all responses to the words independently. Interrater reliability was 97%. Differences were resolved by discussion. The following tests were administered in Year 3: 1. TMS. The task assesses relational knowledge of derived words and analysis of morphologically complex words in meaningful sentences. There were two parts. Production required the children to produce a derived form that completed the sentence accurately (e.g., "Farm. My uncle is a ____." ). This task required syntactic and semantic knowledge of suffixes. Decomposition required the chil- dren to find the base word to complete the sentence meaningfully (e.g., "Driver. Children are too young to ____." ). For each part, half of the items were derived words whose base and complex forms were phonologically transparent (e.g., en- joy, enjoyment), and half required a phonological shift (e.g., nature, natural; see Carlisle, 2000, for further details and test items). 2. Gates-MacGinitie Reading Test (MacGinitie & MacGinitie, 1989): Appro- priate forms were administered to the third and fifth graders. The Reading Vocab- ulary subtest assesses ability to read and select the best meaning for a word from an option set; the Reading Comprehension subtest assesses ability to read short pas- sages and select the best answers to multiple-choice questions about the passages. Procedures For both Year I and Year 3, the experimental tests were administered to each child individually during the winter. The TAVK was administered in several sessions (lasting about *h r each). The Gates-MacGinitie Reading Test was administered by the classroom teachers in the spring. 246 CARLISLE AND FLEMING RESULTS Performance on Lexical Analysis Measures Performances on the experimental tests administered at Year 1 and Year 3 are shown in Table 1. Grade-level comparisons were carried out to determine the ex- tent to which first graders differed from the third graders on WAT and Definition. For the Definition measure, third graders were significantly stronger than first graders, t(59) = 6.23, p < .001 (d = 1.6). On the WAT, two of the first graders did not understand the task requirements on practice items and so were not administered the test. Third graders performed significantly better than first graders, t(59) = 3.32, p < .01 (d = .86). Analysis of Definition Items Analysis of the first and third graders' responses to three derived words yielded measures of Decomposition, Definition of the base word, and Sentence; the per- centages of the children who received credit for each aspect are shown in Table 2. Fisher exact probability calculations were used to determine whether differ- ences in first and third graders' responses were significantly different. As Table 2 shows, for knotless and stillness the groups differed only on the use of the word in a sentence. For treelet, however, third graders were significantly more likely than first graders to decompose the word and to use the word correctly in a sentence. Relationship of Lexical Analysis Tasks to TMS Was children's morphological processing of words in Year 1 significantly related to their performance on the TMS in Year 3? For the younger group, the WAT was significantly related to TMS Production (r = .37, p < .05) but not to TMS Decompo- TABLE 1 Performance on Measures in Years 1 and 3 Year 1 ________________ WAT Definition Derivation TMS Production VOC %ile COM %ile Younger 57 2 (14 4) 5 1 (2 9) 75 2 (14 6) 85 1 (11 6) 75 3 (17 9) 69 4 (16 7) group Older 71 0 (17 8) 11 7 (5 3) 75 9 (12 2) 94.4 (0.06) 80 3 (22 0) 78 5 (13.4) group Note Definition is the raw score; WAT and TMS are % correct. WAT = Word Analysis Test, TMS = Test of Morphological Structure, VOC = Reading Vocabulary; COM = Reading Comprehension LEXICAL PROCESSING 247 TABLE 2 Percentage of First and Third Graders Who Earned Credit for Lexical Analysis Variable Grade 1 Grade 3 Fisher Exact Probability Knotless Decomposition 42.5 62 1 087 Definition of base 30.0 37 9 332 Sentence 25.0 51 7 .021 Stillness Decomposition 72.5 62 1 255 Definition of base 85 0 82 8 528 Sentence 10 0 35 7 007 Treelet Decomposition 10.8 55.2 .001 Sentence 0.0 17.2 .047 Note. Children were not asked to define the base of treelet; three first graders' interviews were terminated prior to treelet sition (r = .13). For the older group, WAT was not significantly related to either subtest (for Production, .25; for Decomposition, .37). Definition was significantly related to both TMS subtests for both grade-level groups: for Definition and Pro- duction, younger group, .57, p < .001, and older group, .54, p < .01; for Definition and Decomposition, younger group, .48, p < .01, and older group, .46, p < .05. Prediction of Reading Comprehension Did performance on the lexical analysis tasks in first and third grades predict read- ing vocabulary and comprehension 2 years later? Two standard multiple regression analyses were carried out for each comprehension measure—one for the younger group and one for the older group. Table 3 shows the results of these analyses. For the younger group, the lexical analysis tasks did not significantly predict reading vocabulary, F(2, 29) = 1.86, p = .17; only 11% of the variance was ac- counted for. However, lexical analysis accounted for 23% of the variance in read- ing comprehension, F(2, 29) = 4.33, p < .05. For the older group, the lexical analysis tasks accounted for 41% of the variance in reading vocabulary, F(2, 23) = 7.86, p < .01; they accounted for 27% of the variance in fifth-grade reading com- prehension, F(2, 23) = 4.29, p < .05. In each of these analyses, Definition but not WAT accounted for a significant portion of the variance 248 CARLISLE AND FLEMING TABLE 3 Prediction of Reading Vocabulary and Comprehension From Lexical Analysis Standard Error of B Grade 3 Beta WAT Definition RVCM WAT Definition Grade 5 RVCM WAT Definition RVCM WAT Definition 04 32 14 42 13 57 21 40 3 25 1 09 81 08 40 02 46 004 29 05 181 168 176 176 Error of B 85 2 51 3 25 1 09 t Standard 24 1 19 85 2 51 74 3 25 46 29 p Note RVCM= Reading Vocabulary; WAT = Word Analysis Test; RCOM = Reading Comprehension. DISCUSSION This study was designed to explore the development of morphological processing in the early elementary years. We took as our starting point a developmental model of the acquisition of morphological knowledge proposed by Schreuder and Baayan (1995), examining whether children's performance on tasks of lexical analysis con- formed to expectations based on this model. Further, we explored the relevance of early elementary lexical processing of complex words to language and reading comprehension 2 years later. Our results suggest that third graders, more than first graders, were capable of morphological processing that is meaning driven, but that even for the first graders, emerging lexical analysis of complex words was related to morphological analysis in sentences and contributed to reading comprehension 2 years later. Early Morphological Processing The initial stage of developing representations of morphemes (bound and free) de- pends on the child's recognition of recursive elements within words—those consis- tent in form and meaning. For decomposition of complex words to occur, the indi- vidual must have access to representations of morphemes (i.e., base word and