The Effect of the Amount of Information on Language Processing

Standard models of processing inflected morphology assume that processing latency to inflected word forms is inversely related to suffix frequency. However, number of lexical decision type of experiments done in Serbian indicates that suffix frequency is not correlated with processing latency. An alternative approach has been developed which assumes that cognitive system is sensitive to variation of the amount of information (bits) carried by inflected forms. The amount of information has been derived from the average probability per syntactic function/meaning for a given inflected form. In regression analysis this descriptor accounts for almost all processing variability of inflected Serbian noun forms. Such an outcome suggests that our cognitive system operates within a uniform metrics and that the observed processing variation may not be due to language specific processing mechanisms. Standard models of processing inflected morphology are developed mainly for English, which is an analytic language with few inflections. Two models seem to be of the utmost relevance: one which assumes that affixed words are stored in the lexicon as a whole unit (the Full Listing Hypothesis) (Manelis & Tharp, 1977; Butterworth, 1983), and the other which assumes that affixed words are decomposed into a base form and affix (the Decomposition Model) (Taft & Forster, 1975; 1976; Taft, 1979, 1981, 1984 Taft, 1994; Taft & Zhu, 1995). According to the Full Listing Hypothesis, processing latency to affixed words is related to the frequency of a stem and on this basis no processing difference should be observed for affixed and non-affixed words. In contrast, the Decomposition Model assumes that an affixed word is decomposed into its base form and affix, each independently searched in the lexicon. By the decomposition model account, processing latency for an affixed word should be affected by two processes: a) the process of decomposing the base form from its affix: this process is assumed to be of constant duration, and b) lexical search for the base form and affix, each search being influenced by the respective form frequencies. Since the base form of a word is the same for all its affixed forms, the Decomposition Model predicts that processing-time variation to different forms of an affixed word should be related to affix frequency. Likewise, the model predicts that affixed words should be processed slower than non-affixed ones due to the fact that non-affixed words need not be decomposed. The lexical decision task is a standard paradigm in this kind of research. Participants are presented with a string of letters and their task is to answer (by pressing yes/no key) whether the presented string is a word or not. The dependent variable is reaction time (expressed in milliseconds [ms]) measured from the stimulus onset. Unlike English, Serbian is a highly inflected language where each open class word is characterized by an inflectional suffix that specifies the particular grammatical property of the word. Thus for example, Serbian nouns, adjectives and pronouns have an inflectional suffix that specifies case, grammatical number and grammatical gender. In the case of verbs an inflectional suffix specifies person, grammatical number, tense and sometimes grammatical gender. However, an inspection of the declension of Serbian feminine nouns (Table 1) indicates that there are several homomorphs, i.e. in some cases, both singular and plural share the same inflectional suffix. As a consequence, in a lexical decision task a word presented in isolation may be equivocal with respect to its case. Thus, what can be experimentally manipulated is an inflected form that may contain several cases (Table 2). Each grammatical form appears with some probability. By the same token, probabilities of inflected forms also vary (Kostić, Đ. 1965a). Table 1: Declension of Serbian feminine nouns Case Singular Plural nominative trav-a* trav-e genitive trav-e trav-a dative trav-i trav-ama accusative trav-u trav-e instrumental trav-om trav-ama locative trav-i trav-ama