The Acquisition of English Derivational Morphology

Three paper-and-pencil measures were administered to students in fourth, sixth, and eighth grades to assess different aspects of their knowledge of English derivational suffixes. Children appear to develop a rudimentary knowledge of derivational morphology--the, ability to recognize a familiar stem in a derivative--before fourth grade. Knowledge of the syntactic properties of derivational suffixes appears to increase through eighth grade. Knowledge of the distributional properties of suffixes also increases, with sixth-grade students showing an increase in overgeneralization errors parallel to that found for inflectional suffixes in much younger children. Derivational Morphology 2 THE ACQUISITION OF ENGLISH DERIVATIONAL MORPHOLOGY English derivational morphology, that is, the basic units of word formation and the principles governing their combination, has drawn increasing attention from researchers in linguistics, psychology, and reading over the past ten years. The findings indicate that knowledge of derivational morphology may be important in language processing in several ways: Knowledge of the internal structure of words may play a role in lexical access (Fowler, Napps, & Feldman, 1985; Stanners, Neiser, Herndon, & Hall, 1979; Taft & Forster, 1975). Because derivational suffixes mark words for part-of-speech, they may be useful in helping speakers establish the syntactic structure of sentences (Clark & Clark, 1977). Finally, knowledge of morphology appears to be helpful in assigning meaning to unfamiliar derivatives (Dowty, 1978; Jackendoff, 1975; Nagy & Anderson, 1984), thus facilitating vocabulary growth. In spite of the increased attention given derivational morphology, we have only fragmentary and inconsistent information about its acquisition. Previous research on acquisition does not clearly identify when children acquire knowledge about the internal structure of words, just what knowledge they acquire, or how well they are able to utilize such knowledge. This paper reports research aimed at establishing a fuller picture of children's acquisition of English derivational morphology by distinguishing different aspects of knowledge about morphology, and by examining the effects of various task demands on children's ability to demonstrate their implicit knowledge. Some research indicates that the acquisition of derivational morphology begins as early as the preschool years. Clark and Cohen (1984) found evidence of some knowledge of the agentive suffixes -er and -ist even for fourand five-year-olds, and Condry (1979) found that second graders had already begun to learn the relationship between stems and derived forms with common suffixes, such as argue and argument. But even if some derivational suffixes such as -er are acquired fairly early, several studies suggest that, in general, students in the middle grades do not have much knowledge of morphology, nor make much use of what knowledge they may have. Freyd and Baron (1982) compared above-average fifth graders with average eighth graders in their use of suffixes when learning morphologically-related nonce words. Children were taught a list of nonce words, half of which were related by real English suffixes (e.g., prok meant "high" and prokness meant "top"). For the other half of the words, the suffixed and nonsuffixed forms had totally unrelated meanings. The students' knowledge of derivational morphology should be reflected in a better score for the former group of words than for the latter. Results indicated that bright fifth graders evidenced some knowledge of morphological relations while average eighth graders did not demonstrate this knowledge. Furthermore, neither group incorporated the part-of-speech information inherent in the suffixes into their definitions. Wysocki and Jenkins (1987) taught fourth, sixth, and eighth grade students the meanings of infrequent words such as sapient, and then tested their knowledge of suffixed derivatives, e.g., sapience. Although eighth graders were usually able to recognize the relationship between the suffixed derivative and the word they had been taught, they were able to demonstrate knowledge of the syntactic contribution of the suffix for only a third of the suffixed derivatives, even though the words were presented in a context that made the part of speech of the word apparent. Sternberg and Powell (1983), investigating students' use of common Latin prefixes and stems to infer the meanings of unfamiliar words such as exsect, found that college students, but not high school students, use the internal morphological structure to infer the meaning of a word. At first glance, the available research presents a somewhat contradictory picture. But some order can be introduced by recognizing differences in the types of morphological knowledge that were being tested and the tasks used to test the knowledge. More specifically, it is necessary to take into account distinctions between classes of word formation processes, different types of knowledge about Tyler & Nagy Derivational Morphology 3 derivational morphology, different degrees of knowledge of stems, and different types of tasks subjects have been asked to perform. Categories of Suffixes On linguistic grounds, it is useful to distinguish two classes of English derivational suffixes, Neutral and Non-Neutral suffixes. Neutral suffixes, such as -ness, -er, -ize, and -ment, have several properties which should make them relatively easy to learn. They attach to independent words; so, for example, when the suffix -er is removed from owner, the result is an independent word, own. Neutral suffixes do not cause changes of stress or vowel quality in the word to which they are added. Usually, although not always, the meaning of a word formed from Neutral suffixes is transparently related to that of the stem. Non-neutral suffixes, such as -ity, ify, -ian, -ous, -ic, or -ive, differ from Neutral suffixes in several respects. They often attach to bound morphemes (stems that are not words in their own right); hence, taking off the -ify in gratify or quantify fails to produce an independent word. Non-neutral suffixes tend to cause changes of stress and vowel quality in the stem to which they attach, as exemplified by the difference in the pronunciation of the a in profane and profanity. Finally, the meaning of words originally formed with Non-neutral suffixes is often not transparently related, as in the words formed from the bound morpheme cam 'meat' such as carnival, carnivore, carnation. The two types of suffixes also vary in their applicability. Neutral suffixes have a wide range of applicability. The primary restriction on these suffixes is their subcategorization for the part-ofspeech of the morpheme to which they can attach. So, for instance, the Neutral suffix -er can attach to virtually any verb in order to form an agentive. Non-neutral suffixes, on the other hand, do not have the same broad range of applicability. As Kiparsky (1982) points out, words formed from particular root morphemes take particular Non-neutral suffixes. For example, words containing the root ceive take -tion, as in receive/reception, deceive/deception, perceive/perception. The root fer, on the other hand, takes the suffix -ence as in prefer/preference, refer/reference. Moreover, within these paradigms there are often idiosyncratic exceptions; compare the triplets arrive/arrival/*arrivation and derive/*derival/derivation. From the properties of Neutral and Non-neutral suffixes, one would predict that the former are acquired earlier, and more easily. The available data bear this out. Dale and O'Rourke (1974) studied sixth, eighth, tenth, and twelfth graders' knowledge of the meaning of common English suffixes. The subjects were asked to match suffixes with a list of definitions. While there are some problems with the study, somewhat odd definitions and the artificiality of the task, the results show a definite trend of subjects at all ages having significantly more knowledge of Neutral suffixes than Non-neutral suffixes. Similarly, Sterling's (1982) study on 20 sixth graders' ability to use suffixes to form context-appropriate, novel forms shows a marked contrast in sixth grade children's knowledge of Neutral and Non-neutral suffixes. Condry (1979) and Clark and Cohen (1984), who found that young children use morphology to aid in learning new words, were looking at Neutral suffixes such as agentive -er. Taking the type of suffix into account, then, already introduces more order into our picture of the acquisition of derivational morphology. Neutral suffixes are being acquired by some children even in the preschool years; in contrast, Non-neutral suffixes appear to be in the process of acquisition even as late as high school. Another possible source of the contradictions among earlier findings is the difference between these studies in the degree of familiarity subjects had with the target words. Condry (1979) found evidence of third graders consistently seeing the relationship between familiar words such as argue and argument. In contrast, Freyd and Baron (1982) and Wysocki and Jenkins (1987), who found far less Tyler & Nagy Derivational Morphology 4 evidence of children seeing lexical semantic relationships, tested subjects on newly-acquired stems. It might be that speakers tend to use morphological knowledge only with derivatives formed from familiar stems. Some apparent discrepancies between previous studies might also be attributed to differences in the types of tasks that subjects were asked to perform. For example, the eighth graders in Freyd and Baron's (1982) study did not make use of suffixes in learning the meanings of nonce words. However, younger subjects in other studies did show some evidence of knowledge of suffixes in different types of tasks. Both Freyd and Baron (1982) and Wysocki and Jenkins (1987) found their subjects unable to convey the syntactic contribution of derivational suffixes in a definitional task; that is, some subjects would incorrectly define sapience as "wise," rather than as "wisdom." However, Sterling (1982) found that for some Neutral suffixes, at least, sixth graders were sometimes able to produce novel forms that were contextually appropriate, that is, which reflected knowledge of the syntactic contribution of the suffix. This suggests that a definitional task imposes extraneous metalinguistic demands that prevent some subjects from evidencing knowledge of suffixes that they do, in fact, possess. Finally, knowledge of morphology appears to be multifaceted. Studies designed to test one aspect of a subject's knowledge of morphology, as was the case with most previous studies, may miss part of a subject's knowledge. Full knowledge of derivational morphology involves at least three aspects, which we label lexical semantic, syntactic, and distributional knowledge. Lexical semantic knowledge is recognizing that words have complex internal structure and that two or more words may share a common morpheme. It is the knowledge that create is related to creator in a way that me is not related to meter. Syntactic knowledge is knowing that derivational suffixes mark words for syntactic category. For example, syntactic knowledge is the tacit knowledge that regularize is a verb by virtue of being suffixed with -ize and regulation is a noun by virtue of being suffixed with -ion. Distributional knowledge has to do with the constraints on the concatenation of stems and suffixes. For instance, -ness attaches to adjectives but not to verbs, so quietness is a fine word in English while *playness is not. It is not logically necessary, nor even likely, that children acquire all three aspects of word formation simultaneously. One could conceivably come to see that there was a regulate in regulation without knowing what part of speech the latter word was or without assigning any systematic part-of-speech characterization to -ion or -ate. Conversely, one might recognize on the basis of the suffix that basement was a noun without any perception of its relationship to the word base. We hypothesize that children acquire knowledge of distributional constraints after acquiring knowledge of lexical semantics and syntax since distributional constraints are in part determined by whether or not the stem morpheme is bound, as well as syntactic factors. Thus, one might expect to find an age at which children in the process of acquiring morphology had already acquired lexical semantic and/or syntactic knowledge of suffixes but had still not acquired knowledge of distributional constraints. The present study attempts to provide the basis for a more coherent picture of the acquisition of suffixes by extending previous studies in several respects. Both Neutral and Non-neutral suffixes were examined, so that differences in their acquisition could be explored. Furthermore, three different types of experimental tasks were used, aimed at measuring the three different aspects of knowledge about suffixes. For clarity of presentation, the three tasks, though all administered to the same subjects in a single session, are reported here as three distinct experiments. Tyler & Nagy Derivational Morphology 5 Experiment 1 Knowledge of Lexical Semantic Relationships We consider the ability to recognize that two words share a common morpheme a basic level of knowledge of morphology, probably the first to be acquired. It is therefore surprising that the subjects in Freyd and Baron's (1982) and Wysocki and Jenkins' (1987) studies showed relatively little evidence of such knowledge. We hypothesized that in these studies, extraneous task demands may have lead to an underestimation of students' actual knowledge of morphological relationships. Specifically, it may be difficult for students to utilize their knowledge of suffixes when the stems involved are unfamiliar or only superficially known; also, producing definitions is more demanding than recognizing correct definitions. A pilot experiment was therefore conducted in which students were asked to answer multiple choice questions about sentences containing low-frequency (and hence presumably unfamiliar) derivatives of high frequency stems. For example, we assumed that most fourth-grade students would already know the words explosive and astronaut, but that the derivatives explosivity and astronautic are infrequent enough that even eighth graders would not be likely to have encountered them before. A seven-item multiple choice test was constructed in which choosing the correct answer depended on knowing the meaning of a low-frequency derivative of a high-frequency stem such as explosivity. All seven derivatives involved Non-neutral suffixes. This multiple choice test was administered to 21 students in fourth grade, 50 in sixth grade, 38 in eighth grade, and also to 12 college students who had been identified as less skilled readers. Fourth graders correctly answered an average of 39% of the items (corrected for guessing), and the mean score increased over the grades tested, with the college students answering 97% correctly. This pilot study indicated that even fourth graders were able to demonstrate lexical semantic knowledge of derived forms. However, the test may still have underestimated students' knowledge of morphology. All the items in the pilot test involved Non-neutral suffixes, which are hypothesized to be more difficult to learn than Neutral suffixes. The pilot test also gave us no way to determine whehter failure to answer an item correctly reflected lack of morphological knowledge, or more general limitations on reading ability or test-taking skills. An experiment that attempted a more thorough measure of students' lexical semantic knowledge of suffixes was therefore conducted. Method Subjects Subjects in this study were students in a medium-sized midwestern town, in schools different from those in which the pilot study had been conducted. There were 40 students in fourth grade, 30 in sixth grade, and 30 in eighth grade. Procedures Subjects took three paper and pencil tasks, of which the test constituting Experiment 1 was the second. Instructions for each task were read aloud by the experimenter as subjects followed along in their individual test booklets. Sufficient time was allowed for all subjects to complete each of the tasks. Tyler & Nagy Derivational Morphology 6 Materials Twenty-four multiple choice items were developed, for which choosing the correct answer depended crucially on the meaning of a target word. For each item, two versions were constructed. In one,the target word was a low-frequency derivative of a high-frequency word. The other version of the item was identical, except that the high-frequency stem itself was used. (Minor changes in wording were made to adjust for differences in part of speech between the stem and the derivative). An example of the two versions of an item is given in Table 1. Two versions of the test were constructed, so that each subject saw only one version of each item. [Insert Table 1 about here.] The low frequency target derivatives all had frequencies of less than 0.3 per million words of text (frequencies based on Carroll, Davies, & Richman, 1971). All the base words had frequencies ranging from 10 to 100 per million words of text, and hence were likely to be familiar, even to the fourth-grade students. The low frequency derivatives fell into three categories, on the basis of their suffix: Eight had high frequency Non-neutral suffixes, eight had high frequency Neutral suffixes, and eight low-frequency Neutral suffixes. The forms used are listed in Table 2. [Insert Table 2 about here.] Design and Analysis A mixed factorial design was used, with item type (suffixed or nonsuffixed target word) and suffix type (high-frequency neutral, low-frequency neutral, and non-neutral) as within-subject factors, and grade (4, 6, & 8) and test version as between-subject factors. The order of items within the test was randomized, but the same order was used in both versions. Results and Discussion Table 3 presents subject means. For each grade and category of derivative, three values are given. The first is the proportion of correct answers for suffixed items, corrected for guessing. The second is the proportion of correct answers for the items with the base words, also corrected for guessing. The third number is the ratio of the first to the second--an informal indication of what proportion of suffixed items subjects would have gotten correct of those items for which they could answer the nonsuffixed version correctly. Means in this table are collapsed across test version. [Insert Table 3 about here.] Results of a MANOVA show a main effect of grade (F(2,94) = 39.6, p < .001), a main effect of item type (suffixed versus nonsuffixed items; F(1,94) = 19.0, p < .001), and a main effect of suffix type (F(2,94) = 120.2,p < .001). The main effect of test version was not significant (F(1,94) = 1.9, p = .173). There was a significant version x suffix type interaction (F(2,94) = 4.5, p = .012), and also a grade x item type x suffix type interaction (F(4,94 = 4.0,p = .004). The main effect of item type shows that suffixed items were more difficult than nonsuffixed items. The fact that at all grade levels, students scored higher on non-suffixed items than on suffixed items shows that novel derived forms continue to cause readers some difficulty through eighth grade. The main effect of grade indicates that older subjects were better at answering both types of items. The non-significance of the grade x item type interaction (F(2,94) = 0.2, p = .793) shows that additional difficulty offered by suffixed items did not change across grades. This result suggests that Tyler & Nagy Derivational Morphology 7 children do not substantially increase in their lexical semantic knowledge of suffixes between grades four and eight. The ability to see lexical semantic relationships between stems and their derivatives is presumably the most basic level of morphological knowledge. It appears that fourth graders have already attained a substantial grasp of this basic level of morphological knowledge. The interaction of item type with suffix type was not significant (F(2,94 = .09, p = .910. This indicates that at the grade levels tested, children evidenced no difference between Neutral and Non-neutral derivatives in terms of lexical semantic knowledge. The grade x suffixation x suffix type interaction does not allow a straightforward interpretation. Comparison of the proportions in Table 3 shows that in fourth grade, Non-neutral suffixes were most difficult, followed in order by low frequency Neutral suffixes and then high frequency Neutral suffixes. This order is what would be predicted, but tests of simple effects indicate that this difference (represented by a suffixation x suffix type interaction) is not significant within the fourth grade (F(2,94 = 2.0, p = .138). However, this order is significantly, although inexplicably, reversed in sixth grade (F(2,94 = 4.4,p = .013. Our results also illustrate some of the methodological problems involved in assessing children's knowledge of morphology. The main effect of suffix type indicates that attempts to make all items equally easy were not successful; inspection of the means shows that items for Non-neutral suffixes were consistenely harder than items for Neutral suffixes, whether the suffixed word or its familiar stem was being tested. The version x suffix type interaction similarly shows that division of items within each suffix type into two groups for constructing the two versions of the test did not result in groups of equal overall difficulty. Non-suffixed items were included as a baseline, and were not intended to pose difficulty for even the youngest subjects. We were surprised to find that even at eighth grade, subjects scored well below 100% on these items. Subjects' relatively low scores on non-suffixed items reflect general limitations on reading and test-taking ability. Therefore, failure to answer the suffixed items correctly cannot be attributed solely to lack of morphological knowledge. Results of Experiment 1 indicate that fourth graders have already attained basic knowledge of both Neutral and Non-neutral derivatives in terms of their ability to recognize that a novel suffixed word is related to its stem. There is an overall increase in performance across grades, but this appears to be due to factors such as increased reading ability, vocabulary, and test taking skills, rather than any specific increase in the ability to see lexical semantic relationships between known words and unfamiliar derivatives. Experiment 2 Knowledge of Syntactic Properties of Suffixes The second aspect of suffixal knowledge under consideration is knowledge of the syntactic properties of suffixes. A primary function of derivational suffixes is to change the part of speech of the stem. Even if one does not know the stem of a word, the derivational suffix can often give unambiguous information about its syntactic category. For instance, one should be able to gather from their endings that aggression is a noun and aggressive is an adjective without knowing the "stem" aggress. In order to measure students' knowledge of the syntactic properties of suffixes with a minimum of extraneous task demands, a multiple choice format was chosen in which the student was asked to choose which of four words fit best into a blank in a sentence. The four words differed only in their suffix, so that a correct choice would reflect knowledge of the syntactic contribution of the suffix. The fact that the four choices differ only by suffixes should also focus the students' attention on the Tyler & Nagy Derivational Morphology 8 suffixes, and hence maximize the likelihood that they would apply to this task whatever knowledge of suffixes they possess. A potential weakness in this design, however, is that children could conceivably learn the part of speech of derived words without attributing any particular syntactic properties to the suffix. For example, one could learn that aggression is a noun without analyzing the word into stem and suffix, in the same way that one learns that joy is a noun, even though it has no overt marking of its part of speech. Therefore, a more stringent test of knowledge of the syntactic properties of suffixes would be a multiple choice item such as just described, but with the four choices consisting of nonce stem plus a variety of suffixes. Answering such an item correctly, aside from chance, is unmistakable evidence that the student knows the syntactic properties of the suffix. Method Subjects Subjects were the same 40 fourth-grade, 30 sixth-grade, and 30 eighth-grade students who took part in Experiment 1. The test described here was the third part of a three-part paper-and-pencil task. Materials Real word items. Sixteen items, such as the example in Table 4, were constructed. Each item consists of a sentence with a blank. Beneath the sentence are four words which differ only in their suffixes. The task is to circle the word that fits best into the sentence. [Insert Table 4 about here.] In half of the items, the correct answers had Neutral suffixes; in the other half, they had Non-neutral suffixes. Most of the suffixed words appeared in the Ginn Word Book (1983), and hence were likely to be familiar to many fourth graders, and most sixth and eighth graders. Nonce word items. Sixteen additional items parallel to those exemplified in Table 4 were constructed, but using nonce words with suffixes in place of real words. An example of this type of item is given in Table 5. [Insert Table 5 about here.] Two versions of each item were created, one in which the nonce words had Neutral suffixes, and one in which the nonce words had Non-neutral suffixes. Each subject saw each item only once, half in the Neutral suffix version and half in the Non-neutral suffix version. Neutral and Non-neutral items were randomly ordered. In the complete test, the 16 real word items came first, followed by the 16 nonce word items. Design and Analysis A mixed factorial design was used, with item type (real word items or nonce word items) and suffix type (Neutral or Non-netural) as within-subject factors, and grade (4, 6, and 8) and test version as between-subject factors. Tyler & Nagy Derivational Morphology 9 Results and Discussion Table 6 gives the means, corrected for guessing. Figures are collapsed over test version. Results of a MANOVA show a main effect for grade (F(2,94 = 38.3, p < .001), and a main effect of item type (F(1,94) = 222.5, p < .001). No other main effect or interaction was significant. Post-hoc tests revealed that all cell means were significantly greater than zero, except for fourth graders' scores for Neutral nonce word items. [Insert Table 6 about here.] Experiment 2 was replicated with a second group of students consisting of 21 fourth graders, 50 sixth graders, and 38 eighth graders. Results, given in Table 7, were almost identical to those of the first group of students, with the exception that fourth grade means for both Neutral and Non-neutral nonce words were higher, and both were significantly greater than zero. [Insert Table 7 about here.] Our results show that students at all grade levels tested have at least some knowledge of the syntactic properties of suffixes. In the main experiment, students at fourth grade did not perform significantly above chance for Neutral nonce word items; in the replication, fourth graders performed above chance on both types of nonce word items. Fourth grade, therefore, appears to be the point at which students are becoming able to apply their knowledge of syntactic properties of suffixes to derivatives formed from unfamiliar stems. Fourth graders are already performing significantly above chance on the real word items. This clearly shows that they are able to distinguish among derived words that differ only in their suffixes. The performance on real word items in itself does not demonstrate knowledge of the syntactic properties of the suffixes, since the students could be treating these words as unanalyzed units. However, the fact that fourth graders are beginning to successfully answer nonce word items demonstrates that they have begun to recognize the syntactic properties of suffixes. This knowledge can only have come from analyzing known derivatives. The main effect of grade, and the lack of significant interactions with grade, indicates that older subjects did better on all types of items to a roughly equal extent. We believe that the substantial increase in scores, especially between fourth and sixth grades, represents some growth in knowledge of the syntactic contribution of suffixes. However, at least some of this increase might also be due to gains in vocabulary and test-taking skills. The main effect of item type shows that subjects did better with real word items than with nonce word items. The presence of an unfamiliar or unknown stem appears to constitute an extraneous source of difficulty, which hinders at least some students from demonstrating what knowledge of suffixes they do possess. This sheds some light on the differences in performance between the subjects in Condry's (1979) and Freyd and Baron's (1982) studies. In the former study, children as young as third grade evidenced knowledge of morphological relationships between familiar words (e.g., argue and argument); in the latter, eighth graders failed to utilize morphological relationships when learning nonce stems and their derivatives. The lack of any effect of suffix type shows that students at all three grade levels knew the syntactic properties of Neutral and Non-neutral suffixes equally well. Experiment 3 A third aspect of knowledge about suffixes tested in this study is distributional knowledge. All suffixes are constrained by the syntactic category of the base they attach to, e.g., -ness attaches to adjectives (happy/happiness), -ize attaches to nouns (critic/criticize). Many suffixes also have Tyler & Nagy Derivational Morphology 10 restrictions such as [+Latinate]; for example, -ity attaches to Latinate adjectives (as in nude/nudity) but not Germanic adjectives (bad/*badity). This aspect of knowledge about suffixes should be the last to be acquired; children would have to recognize a suffix as such, and have at least some idea of its syntactic contribution, before they could figure out in exactly which ways its distribution is restricted. Furthermore, we expect that in the process of acquiring distributional knowledge about suffixes, learners will go through a period of incomplete distributional knowledge marked by overgeneralizations, i.e., the use or acceptance of words (such as *badity or *repeatize) in which the presence of a suffix violates constraints on the distribution of that suffix. Overgeneralizations occur when the child has recognized the existence of a linguistic regularity, but has not yet precisely identified its range of application (Bowerman, 1982). This could happen because the child has not learned specific exceptions to otherwise productive rules, e.g., that the plural of mouse is mice and not *mouses. Or the child might misanalyze words, producing doubly marked forms such as *feets and *sheeps, or failing to add a plural morpheme to words such as house and rose which already end in alveolar fricatives. As with inflections, the person learning derivational morphology must learn lexical exceptions to otherwise productive rules (e.g., that the agentive form of to spy is spy and not *spyer), and restrictions on the distribution of suffixes. Overgeneralizations provide the clearest type of evidence that acquisition of morphological processes as a productive rule is taking place. MacWhinney (1978) has argued that overgeneralizaation is the major indicator of a productive, rule-governed process, and that the failure to overgeneralize is indicative of memorization or analogical processes. Therefore, it is crucial to include some way of looking for overgeneralizations in any attempt to arrive at a systematic picture of the acquisition of derivational morphology. Assessing Overgeneralization with Nontimed Lexical Decision Tasks Overgeneralizations of inflectional suffixes, such as *foots or *eated, are very frequent in the speech of young children. Researchers of child language acquisition have compiled long lists of inflectional overgeneralizations in young children's speech (Berko-Gleason, 1971). Evidence for derivational overgeneralization, on the other hand, although it does occur, is harder to come by. Speakers are generally less likely to make up new words using derivational suffixes than to apply inflectional suffixes to new forms. However, it is possible to tap overgeneralization by measuring subjects' willingness to accept novel forms that violate distributional constraints. Anderson and Freebody (1983), using a type of nontimed lexical decision task (similar to the task used by Aronoff & Schvaneveldt (1978) to determine adult subjects' knowledge of the distributional constraints of derivational suffixes), found that high-ability fifth graders were more likely than lowability fifth graders to say they knew the meanings of novel derived forms such as *loyalment. Saying that one knows the meaning of a novel derived form such as *loyalment, which violates distributional constraints of the suffix -ment, can be taken as a type of overgeneralization, hence as evidence that subjects are in the process of learning about derivational suffixation as a productive, rule-governed process. Nagy, Anderson, and Herman (1987) used a similar task with third, fifth, and seventh graders, and found that the tendency to accept novel derived forms appeared to peak between 5th and 7th grade, after which the tendency declined, again suggesting a period of overgeneralization. However, Nagy, Anderson, and Herman did not differentiate between well-formed novel derivatives such as dogless and ill-formed derivatives such as *earthous. The purpose of Experiment 3 was to determine if this finding could be replicated, and to extend it in two ways. First, a distinction was made between well-formed and ill-formed novel derivatives, and second, a systematic distinction was made between Neutral suffixes and Non-neutral suffixes. Tyler & Nagy Derivational Morphology 11 We hypothesized that at a certain stage in their acquisition of derivational morphology, children would accept an increased number of ill-formed derivatives as words of English, indicating that they were formulating general morphological rules that did not yet incorporate all the necessary constraints on distribution of suffixes. Later this tendency to accept ill-formed derivatives would decline, indicating that they had acquired more adultlike formulations of the rules. Because of the increased complexity of Non-neutral suffixation, we hypothesized that the overgeneralization peak for Non-neutral suffixes would occur at a later age than the peak for Neutral suffixes, indicating that children acquire Non-neutral suffixes after Neutral suffixes. Method Subjects Subjects included the same 40 fourth-grade, 30 sixth-grade, and 30 eighth-grade students who took part in Experiments 1 and 2. Because pilot studies had indicated that important changes in performance on this task might occur after eighth grade on this type of task, data from 29 undergraduate students at a large midwestern university tested in a separate session were also included in the analyses. Materials A 171-item nontimed lexical decision test was developed. Each item consisted of a word or nonce word followed by the words Yes and No. Subjects were instructed to circle Yes if they knew the meaning of the word and No if they did not. The list was comprised of items of several categories: 1. Real word of English words that ranged from low-frequency (e.g., emir) to high-frequency (e.g., ocean). Subjects' scores on these words provided a general measure of vocabulary knowledge. 2. Phonologically possible but non-occurring words (e.g., werpet). 3. Experimental words which were suffixed words, 20 of which were well-formed (i.e., obey distributional constraints) and 20 of which were ill-formed (i.e., violate distributional constraints). Half of both the well-formed and the ill-formed derivatives were suffixed with Neutral suffixes and half with Non-neutral suffixes. The stems from which derivatives were formed were balanced for length and frequency. A list of these forms appears in Table 8. [Insert Table 8 about here.] All the derivatives were intended to be words that subjects were unlikely to have seen before. Wellformed derivatives were intended to be possible, although infrequent, words in English. In fact, all of the well-formed derivatives except for interiorist and centrate appear in Webster's Third Unabridged Dictionary. Those that did appear in Carroll, Davies, and Richman's (1971) Word Frequency Book were low in frequency--all occurred less than three times in 10 million words of text, and all but three occurred less than once in 10 million words of text. Thus, it is very likely that our subjects had seen or heard very few, if any, of the derivatives in this lexical decision task. Three versions of the test were constructed, each with a different randomized order of items. Tyler & Nagy Derivational Morphology 12