Memory & Cognition 1980, Vol. 8 (5), 415423 Represeritation of inflected nouns in the internal lexicon G. LUKATELA, B. GLIGORIJEVIC, and A. KOSTIC University ofBelgrade, Belgrade, Yugoslavia and M.T.TURVEY University ofConnecticut, Storrs, Connecticut 06268 and Haskins Laboratories, New Haven, Connecticut 06510 The lexical representation of Serbo-Croatian nouns was investigated in a lexical decision task. Because Serbo-Croatian nouns are declined, a noun may appear in one of several gram­ matical cases distinguished by the inflectional morpheme affixed to the base form. The gram­ matical cases occur with different frequencies, although some are visually and phonetically identical. When the frequencies of identical forms are compounded, the ordering of frequencies is not the same for masculine and feminine genders. These two genders are distinguished further by the fact that the base form for masculine nouns is an actual grammatical case, the nominative singular, whereas the base form for feminine nouns is an abstraction in that it cannot stand alone as an independent word. Exploiting these characteristics of the Serbo­ Croatian language, we contrasted three views of how a noun is represented: (1) the independent­ entries hypothesis, which assumes an independent representation for each grammatical case, reflecting its frequency of occurrence; (2) the derivational hypothesis, which assumes that only the base morpheme is stored, with the individual cases derived from separately stored inflec­ tional morphemes and rules for combination; and (3) the satellite-entries hypothesis, which assumes that all cases are individually represented, with the nominative singular functioning as the nucleus and the embodiment of the noun's frequency and around which the other cases cluster uniformly. The evidence strongly favors the satellite-entries hypothesis. Inflection is the major grammatical device of Serbo­ see that most nouns were masculine and that the nomi­ Croatian, Yugoslavia's principal language. In general, the native singular was the most popular grammatical case. grammatical cases of nouns are formed by adding a The italicized numbers are normalized percentages and suffix to a root morpheme, where the suffix is of the can be read as follows, taking the masculine gender as an vowel, vowel-consonant, or vowel-consonant-vowel type. example. For any given masculine noun that occurs in Less frequently, inflection involves additional processes, the language with frequency f, the nominative singular such as vowel deletion and consonant palatalization. form of that noun occurs with a frequency of approxi­ The grammatical cases of Serbo-Croatian nouns mately .29f, the genitive singular form with a frequency produced by inflection are not equal in their frequency of approximately .19f, the dative singular form with a of occurrence. Table I summarizes the frequency frequency of approximately .02f, and so on. In short, analysis of D. Kostic (1965) on a corpus of approxi­ the normalized percentage for a given grammatical case mately 2 million Serbo-Croatian words appearing in of a given gender is the likelihood that, when a noun of the daily press and contemporary poetry. The non­ that gender appears, it appears in that particular case. italicized numbers are actual percentages. Thus, for all The question of interest to the present paper is how nouns in the corpus, 12.83% were masculine nouns in the inflected Serbo-Croatian nouns are represented in the nominative singular, 7.8% were feminine nouns in lexical memory. Following MacKay (1978) and Manelis the genitive singular, .13% were neuter nouns in the and Tharp (1977), we can distinguish two hypotheses instrumental plural, and so on. Reading the totals, we about the lexical representation of words with common morphological stems. According to the independent­ This research was supported in oart by NICHD Grant HD­ entries hypothesis, the individual grammatical forms of a 08495 to the University of Belgrade and in part by NICHD Serbo-Croatian noun would be represented in the lexicon Grant HD.{) 1994 to Haskins Laboratories. Reprint requests by independent representations, one internal representa­ can be sent either to G. Lukatela, Faculty of Electrical Engi­ neering, Bulevar Revolucije 73, Beograd, Yugoslavia, or to tion for each grammatical form. On the derivational M. T. Turvey, Haskins Laboratories, 270 Crown Street, New hypothesis, rather than all the forms of a given noun Haven, Connecticut 06510. being instantiated in the internal lexicon, there would be Copyright 1980 Psychonomic Society, Inc. 415 0090-502X/80/050415-Q9$O 1.15/0 416 LUKATELA, GLIGORIJEVIC, KOSTIC, and TURVEY Table I Case Frequencies in Percentages Singular Plural Masculine Feminine Neuter Total Masculine. Feminine Neuter Total Nominative 12.83 8.84 2.88 24.55 3.33 3.58 .69 7.60 28.89 22.56 20.44 7.50 9.14 4.30 Genitive 8.56 7.88 3.47 19.91 3.96 3.22 .61 7.79 19.27 20.11 24.63 8.92 8.22 4.33 Dative .87 .38 .31 1.56 .28 .16 .04 .47 1.96 .97 2.20 .63 .41 .28 Accusative 5.49 5.48 2.55 13.52 2.21 2.75 .73 5.69 12.36 13.99 18.10 4.98 7.02 5.18 Instrumental 1.90 1.94 .86 4.70 .60 .80 .13 1.46 4.28 4.95 6.10 1.35 1.86 .92 Locative 3.77 3.42 1.61 8.80 .61 .80 .21 1.62 8.48 8.73 11.43 1.37 2.04 1.48 Total 33.42 27.94 11.68 73.04 10.99 11.24 2.41 23.64 75.25 71.31 82.89 24.75 28.69 17.11 Note-This tableisadopted from D. Kostic (1965). Figures in italics represent the normalizedpercentages, asrelatedto the particular gender. Percen tages do not add to 100% because the rarely occurring vocative casehasbeenomitted. but one instantiation, probably of the noun's root successful (meaning that the letter string is a nonaffixed morpheme. There would also be in memory only a word), then a positive response can be initiated. How­ single instantiation of the set of inflectional morphemes. ever, if the search is unsuccessful, then the letter string Appropriate combinations of the root morpheme and is decomposed and the combination of root and affix inflections would be determined by separately stored is tested for its validity. Obviously, on the decomposition­ syntactic rules. second argument, lexical decision should be slower for There have been relatively few direct contrasts of affixed words. The Manelis and Tharp (1977) investi­ the two hypotheses for English lexical items, and the gation failed to find a difference between affixed and results have been largely equivocal. Manelis and Tharp nonaffixed words in either direction, a result that (1977) compared lexical decision ("Is this letter string favored the independent-entries hypothesis over either a word?") times for pairs of affixed words (words version of the decompositional hypothesis. consisting of two morphemes, a root morpheme and a However, the failure to find evidence for morpho­ suffix) with lexical decision times for pairs of nonaffixed logical decomposition with suffixed words contrasts words (words consisting of a single morpheme). Manelis with the provision of such evidence by Taft and Forster and Tharp (1977) predicted two possible outcomes from (1975) for prefixed words. These investigators reported the derivational or, as they termed it, decompositional that rejecting real roots (for example, SULTS as in hypothesis. For a given letter string, decomposition into INSULTS) as words took longer than rejecting false root and ending could be an obligatory first step, with roots (for example, NINGS as in INNINGS) as words. lexical search for the whole item a contingent later The interpretation given was that real stems. would be step; or lexical search for the whole item could be the found in the lexicon and a subsequent check would be initial obligatory step, with decomposition occurring needed to determine that these lexical entries do not later and dependent upon failure to find the whole item constitute words in the absence of an appropriate in memory. Consider the prediction that follows from prefix. the notion that decomposition occurs first. A word, A further demonstration of morphological decompo­ whether it be affixed or nonaffixed, is partitioned into sition is reported by MacKay (1978), although his exper­ root and ending. A test is then made to determine iment is distinguished from the experiment described the validity of the combination as an affixed word. If above in that it looks at the production of words rather the combination proves valid, a positive response is than at their perception. Subjects heard verbs (e.g., con­ initiated; if it proves invalid (meaning that the word is clude, decide) that they had to nominalize (conclusion, nonaffixed), a search of the lexicon is conducted for the decision) as rapidly as possible (MacKay, 1978). It was nondecomposed letter string. In brief, with everything shown that certain nominalizations took longer than else equal, the decomposition-first argument predicts others; precisely, the more complicated the derivational faster lexical decision for affixed words than for non­ process (the more steps intervening between verb form affixed words. The contrary prediction follows from the and noun form), the slower the nominalizations. decomposition-second argument. If the initial search of The source of the discrepancy between the experi­ the lexicon for the nondecomposed letter string is ments of MacKay (1978) and Manelis and Tharp (1977) INFLECTED NOUNS 417 could be relatively trivial, a matter of differences in the coding of word frequency in lexical memory can be methodology. On the other hand, the discrepancy might referred to as an interentry account, then its popular arise from a deep-seated difference between the kind of alternative can be referred to as an intraentry account, memory structure needed to recognize words and the for here the emphasis is not on an entry's position kind of memory structure needed to produce them.