The Mental Representation of Infiected Words: An Experimental Study of Adjectives and Verbs in German

Sonja  Eisenbeiss

Pipelines are an economical means of transporting oil and gas. A commonly encountered performance issue with such structures, however, is their susceptibility to buckling under hydrostatic and hydrodynamic pressure loads. In such a failure scenario, the pipe will suddenly collapse under the action of the loads and can render the transport structure as ineffective. The engineering design of pipes must therefore account for the action and magnitudes of such anticipated service loading, referred to as buckle propagation pressure, in achieving an adequate and sound performance. As for the notion of economy, since buckle propagation pressure is closely related to the pipe wall thickness, it has a direct bearing on overall project costs. In this study, to simulate in situ conditions of subsea pipelines, only uniform hydrostatic pressure was taken into consideration as the source for loading. A finite element method (FEM) model was then used to examine the buckling modes on pipes of two different diameters varying at four diameter-to-thickness ratios. With the pipes modeled as having clamped supports, predicted values of stress, strain, reaction force, and displacement at the ends and at midspan, were obtained. The results found show an inverse relationship between the diameterto-thickness ratios and buckling capacities of the pipes.

The Mental Representation of Inflected Words: An Experimental Study of Adjectives and Verbs in German* * Harald Clahsen Department of Linguistics University of Essex Sonja Eisenbeiss Max Planck Institute for Psycholinguistics Meike Hadler & Ingrid Sonnenstuhl Department of Linguistics University of Düsseldorf In: Language 77(3), pp.510-543 (2001) Address for Correspondence: Harald Clahsen Department of Linguistics University of Essex Colchester, C04 3SQ, UK email: [email protected] * The research reported in this paper is supported by the Deutsche Forschungsgemeinschaft (DFG) 'German Research Council' (grant SFB 282/C7). We thank the members of our research group, Joana Cholin, Kerstin Mauth, and Axel Huth for assistance in administering the experiments. We are also grateful to Martin Atkinson for a close reading of the article. 2 ABSTRACT This paper investigates how morphological relationships between inflected word forms are represented in the mental lexicon focusing on paradigmatic relations between regularly inflected word forms and relationships between different stem forms of the same lexeme. We present results from a series of psycholinguistic experiments investigating German adjectives (which are inflected for case, number, and gender) and the so-called strong verbs of German which have different stem forms when inflected for person, number, tense or mood. Evidence from three lexical decision experiments indicates that regular affixes are stripped off from their stems for processing purposes. It will be shown that this holds for both unmarked and marked stem forms. In another set of experiments, we found priming effects between different paradigmatically related affixes and between different stem forms of the same lexeme. We will show that associative models of inflection do not capture these findings. Instead, we will explain our results in terms of combinatorial models of inflection in which regular affixes are represented in inflectional paradigms and stem variants are represented in structured lexical entries. We will also argue that the morpho-syntactic features of stems and affixes form abstract underspecified entries. The experimental results indicate that the human language processor makes use of these representations.* 3 1. MORPHOLOGY AND THE MENTAL LEXICON. Much psycholinguistic research has been devoted to the question of whether there is any correspondence between the linguistic structure of a morphologically complex word and the way it is segmented by the speakerhearer during on-line production and comprehension. Are morphologically complex words that have stem+affix representations (e.g. derived words such as govern-ment or regularly inflected words such as walk-ed) computed via their constituent morphemes? Are irregularly inflected words that cannot be formed through affixation stored unanalyzed in the mental lexicon? Experimental studies have produced conflicting answers to these questions. Two broad views can be distinguished. Associative models of morphological processing claim that the morphological structure of words plays no role in the way they are produced or perceived and that words are listed as full forms in memory. The key idea is that all morphological patterns including those that can be decomposed into stems, roots, and affixes are derived from a network of associative relations. Connectionist networks of inflection (Rumelhart & McClelland 1986, MacWhinney & Leinbach 1991, Plunkett & Marchman 1993, among others) can be seen as modern implementations of associative models of language. In contrast to this, several other researchers have argued that the mental lexicon encodes morphological structure and that this information plays a role in comprehension and production. Specifically, the language processor is said to make use of morphological decomposition for dealing with morphologically complex words, in addition to full-form representations (e.g. Laudanna & Burani 1985, 1995, Frauenfelder & Schreuder 1992, Schreuder & Baayen 1995, Pinker & Prince 1991). A related question that has received much less attention from psycholinguists is how morphological relationships between inflected word forms are represented in the mental lexicon. Consider, for example, the inflected adjectives from German in (1a), which are 4 marked for case, number, and gender, and the inflected verb forms in (1b), which are marked for person, number, and tense. (1) a. wild-es 'wild-nom. neut. sg.' b. (ich) werf-e st '(I) throw-1 sg. pres.' wild-em wild-er 'wild-dat. masc. sg.' 'wild-nom. masc. sg.' (du) (sie) '(you) wirf-st throw-2 nd sg. pres.' warf-en '(they) throw-3rd pl. pret.' Both adjectival and verbal agreement affixation are highly regular: -s, -m, and -r can be attached to any adjective and -e, -st, and -n to any verb. Exceptions are a few high frequency forms, such as the suppletive form bin ('am' 1st sg. pres.) and 1st sg. and 3rd sg. present tense forms of verbs such as müssen 'to be able to', dürfen 'to be allowed to', sollen 'to be about to', which (like their corresponding preterite forms) do not have a person/number agreement suffix: ich/er muss, darf, soll 'I/he must, may shall'. Thus, decompositional accounts of morphological processing predict that all the inflected forms in (1) are segmented into stems and affixes, yielding two sets of affixes (one for adjectives, one for verbs) and two sets of stems (one invariant adjective stem, wild, and three different stems of the lexeme werfen). This raises the question of how the morphological relationships between affixes such as those in (1a) and the relationships between different stem variants such as those in (1b) are represented in the mental lexicon and how they are processed. With respect to these questions, we can distinguish between two principal views. In associative models of inflection, in which all inflected word forms are represented in terms of associative networks, morphological relationships such as those in (1) are not directly encoded in the mental lexicon. Rather, they are treated as epiphenomena of associations between contiguous or otherwise similar linguistic elements (see e.g. Bybee 1991, 1995a, MacWhinney et al. 1989, Taraban et al. 1989, Elman et al. 1996). For example, the final segments -s, -m and -r of German adjectives may form an associative pattern in that they 5 share sets of phonological and semantic connections that are repeated across multiple sets of words. Under this account, the affixal properties of these linguistic elements as well as the paradigmatic relations between them are said to emerge from a network of associative patterns without encoding any morphological structure. Similarly, the different stem variants illustrated in (1b) might be separately stored and associatively related, with respect to phonological and semantic properties. If this is correct, there would be no need for morphological representations and operations in the mental lexicon. Different variants of associative models of inflection have been suggested in the literature; see section 2 for some discussion. Here, our focus will be on the questions of how associative models deal with morphological relationships, specifically with related regular affixes and with stem variants of the same lexeme. The alternative view holds that morphological structure and morphological relationships are encoded in the mental lexicon and are important for understanding on-line morphological processing (see e.g. Marslen-Wilson et al. 1994, Pinker 1999, Clahsen 1999 for review). For many morphologists, regular affixes such as those in (1a) and (1b) constitute inflectional paradigms. In psycholinguistic terms, one can think of a paradigm as a matrix or access system for mapping grammatical information, i.e. morpho-syntactic features, to their exponents or affixes. The question, then, is whether there is any empirical evidence that the mental lexicon is organized into paradigms or whether paradigms are epiphenomena that reflect similarities between word forms in meaning and form. A related question concerns the representation of properties that define paradigms: Are affixes fully specified for their morpho-syntactic features, or are they underspecified for redundant features? Linguists have also proposed mechanisms to account for the relationships between stem variants such as those in (1b), e.g. lexical redundancy rules (Chomsky 1970, Jackendoff 1975) and default inheritance representations (Corbett & Fraser 1993, Wunderlich 1996). The purpose of these 6 mechanisms is to capture relationships between lexical items and at the same time to permit the lexicon to avoid listing redundant parts of the word. The question we will address here is whether there is any psycholinguistic evidence that lexical representations for stem variants (as well as for affixes) are ‘economical’, in the sense that they are underspecified for redundant features. In what follows, we will first discuss how different models of the mental lexicon deal with morphological relationships between regular affixes and relationships between stem variants of the same lexeme. We will then report results from two experiments on German adjective inflection examining the processing of regular affixes, and three experiments examining the so-called strong verbs of German to determine how stem variants are represented in the mental lexicon. 2. ASSOCIATIVE MODELS OF INFLECTION. In associative models of inflection, a word's morphological structure plays no direct role in the way it is produced or perceived. Inflected words are all represented in the same way, i.e., by storing them in associative networks and by creating connections among them. Through repeated exposure to multiple sets of inflected words, an associative network will form patterns that range over sets of connections. In this way, morphological relationships are claimed to be secondary, as they are derivable from associations between words. A well-known example are the so-called satellite models in which word nodes are posited for each morphological variant of a given lexeme, and bidirectional connections between each word node and a corresponding base form, most typically the stem (Lukatela et al. 1978, 1980, 1987, Fowler et al. 1985, Feldman & Fowler 1987). Applying this approach to the inflected word forms in (1) would yield lexical entries 7 such as those in (2) which are composed of a nucleus, the stem in these cases, and satellites representing all other forms: (2) (a) (b) wirfst wildem wilder werft warft werf wild wildes warfst wilde werfe wirfst werfen warf This model posits a special status to the base form or nucleus. Some experimental evidence supports this view. Lukatela et al. (1978, 1980) showed, for example, that Serbo-Croatian nouns in oblique case forms (instrumentals and datives) elicit longer response times in visual lexical decision tasks than nominative forms. In their model, the latter are regarded as nucleus forms and the oblique case forms as satellites, i.e. as full-form representations connected to the nucleus. Shorter lexical decision times for base forms have also been obtained for other categories and for other languages (see Günther 1988). These results have been taken to indicate that the lexical identity of a word is tied to the nucleus, and that if access is made via a satellite form, extra time is required. Moreover, the inflectional variants of a lexeme are connected to the nucleus, but not necessarily connected to each other. Thus, if this model is correct, we would expect corresponding experimental effects, e.g. frequency effects for inflected word forms, rather than for affixes and stems. Furthermore, this model does not predict any experimental differences between the various satellites of a nucleus. In more recent associative models of inflection, the distinction between a designated base form and inflectional variants has been given up in favour of the view that all word forms are stored in associative memory irrespective of their morphological constituency. This view has 8 been taken in most connectionist models of inflection (see Elman et al. 1996) as well as in schema-based models of inflection (Bybee 1995a, Köpcke 1993, 1998). Inflectional paradigms are claimed to 'emerge on the basis of associations between cues' (MacWhinney et al. 1989:274). Along the same lines, Bybee (1995b:242f.) states that paradigms should be represented as 'clusters of highly connected words'. Under this view (see also Taraban et al. 1989), a paradigm is a set of word forms connected more or less strongly in various ways (phonologically, semantically) to other word forms that need not necessarily be members of the same paradigm, in traditional terms. The strength of the connections is largely determined by frequency and similarity. Hence, the nature of the connections between two phonologically related word forms, e.g. car and card, does not differ in any fundamental way from the connections between morphologically related word forms such as those in (1a) and (1b). For illustration, consider the associative representation of inflected word forms of the verb werfen in (3). (3) 9 As illustrated in (3), this model does not attribute any special status to base forms or nuclei, and all inflectional variants (including the base form) have word-level status. Hence we would expect to find word form frequency effects for all inflectional variants of a given word. For example, if there are processing differences between inflected forms, e.g. between wildes and wildem, these should be due to different word form frequencies. Moreover, we should find similarity effects for all inflected word forms. That is, pairs of word forms which exhibit a high degree of formal (= orthographic and/or phonological) or semantic similarity are more closely connected and should therefore exhibit stronger priming effects than pairs of word forms which are less similar with respect to their formal and semantic properties. 3. COMBINATORIAL APPROACHES TO INFLECTION. In contrast to associative models of the mental lexicon, several other researchers have argued that morphologically complex words are decomposed or parsed, i.e. segmented into smaller morphological units, e.g. stems, roots, affixes. This view has been supported by a rich body of psycholinguistic literature demonstrating experimental effects of morphological decomposition for derivational morphology and regular inflection (see Marslen-Wilson et al. 1994, Pinker 1999, Clahsen 1999 for review). Consider, for example, results from morphological priming studies. For (semantically and phonologically transparent) derived words, Marslen-Wilson et al. (1994) found, that nouns such as governor and government both prime their corresponding form govern, but that the two derived nouns do not prime each other. To account for these findings, they proposed decomposed, i.e. morphemically-based, representations of derived nouns in terms of stems and affixes, such as in (4): (4) -ment govern -or 10 The arrows in (4) are meant to indicate inhibitory links between the two suffixes to capture the lack of priming between -or and -ment forms of the same stem. Thus, govern+ment activates the stem govern, but at the same time inhibits other derived forms of the same stem. Marslen-Wilson et al. (1994) attribute this to lexical competition: -ment and -or produce lexical items with distinct meanings that are incompatible with each other, and hence they inhibit each other in priming tasks. With respect to regular inflection, similar issues arise. Suppose that regularly inflected words are indeed decomposed into stems and affixes (see Stanners et al. 1979, Kempley & Morton 1982, Fowler et al. 1985, Napps 1989, Marslen-Wilson et al. 1993, Sonnenstuhl et al. 1999). This then raises the question of how morphologically related affixes of the same stem are to be represented in the mental lexicon. In most morphological frameworks the relationships between regular affixes such as those in (1) are described in terms of paradigms1. An inflectional paradigm is a multi-dimensional, potentially recursive matrix which is defined by the morpho-syntactic features of word forms or affixes2 (see e.g. Aronoff 1976, 1994, Anderson 1982, 1992, Zwicky 1985, Carstairs 1987, Stump 1993). A paradigm contains a set of slots defined in terms of morpho-syntactic feature values and shows how each slot is to be 1 2 Note, however, that the status of inflectional paradigms is controversial among linguists. There are theories that are based on paradigms (e.g. Stump (to appear), Wunderlich 1996), others that countenance paradigms (e.g. Anderson 1992), and yet others that deny their existence altogether (e.g. Bybee 1985, Lieber 1992, Halle & Marantz 1993, and associative models in general). A related linguistic controversy concerns the status of affixes. Wunderlich (1996) and Jackendoff (1997), for example, believe that regular affixes represent lexical entries, and consequently, paradigms are considered to be affix-driven, i.e. directly constituted by the 'combinatory force of the inflectional affixes' (Wunderlich 1996:96). Other morphologists do not assume lexical entries for inflectional affixes. Instead, they posit what Anderson (1982, 1992) calls 'morpholexical rules' and Stump (1993) 'realization rules', i.e. rules that specify how a given set of features is to be spelled out. These rules effectively determine a set of slots in a paradigm and show how each slot is to be filled. Thus, irrespective of whether affixes form lexical entries or are better treated as exponents of morpho-syntactic features derived from realization rules, under both views 11 filled. The result is that any lexeme that belongs to a particular syntactic category has inflected word forms defined by the paradigm. The formation of paradigms is constrained by general principles, such as Blocking and Specificity (Kiparsky 1982, 1998), and by paradigm structure constraints, such as Completeness and Uniqueness (Pinker 1984, Wunderlich 1996). Completeness requires that every cell of a paradigm must be occupied, and Uniqueness that every cell is uniquely occupied. Blocking and Specificity require that if two rules or affixes are in competition for one paradigm slot, the rule that is more specific in its application is preferred over the more general one. If this is how regular affixes are organized in the mental lexicon, we would expect to find corresponding experimental effects, for example, priming differences between specific and less specific affixes (or rules). Moreover, the idea that regular affixes are represented in inflectional paradigms means that affixes such as those in (1) (or equivalently corresponding realization rules) are stored independently of the stems on which they occur. Hence inflected word forms containing regular affixes should be decomposed into stem and affix, irrespective of the kind of stem on which they occur. Finally, if regular affixes are stored in the mental lexicon, we would expect that they produce experimental effects that are characteristic of lexically stored items, e.g. frequency effects. We have examined these predictions in the experiments to be reported in section 4. With respect to the representation of stem variants such as those in (1b), linguists have developed the notion of lexical redundancy rules (Chomsky 1970) or equivalent mechanisms, e.g. 'lexical rules' (Jackendoff 1975, 1997) or default inheritance hierarchies (Corbett & Fraser 1993, Wunderlich 1996), to capture the morphological relationships between them. The key idea that is common to these approaches is that stem variants of the same lexeme are paradigms play an important role in organizing related inflected forms. 12 not separately and completely listed, but that some of them have an impoverished (= underspecified) entry (e.g. wirf and warf) and that when these items are used, their full interpretation and form is filled in from the base entry (= werf). In other words, stem variants may form subnodes within hierarchically structured lexical entries. If this is how stem forms are stored in the mental lexicon, we would expect corresponding experimental effects, for example, stem frequency effects in lexical decision experiments and priming differences between the various stem variants. We have examined these predictions in the experiments to be reported in section 5. 4. EXPERIMENTS ON ADJECTIVE INFLECTION. In this section, we will investigate evidence from lexical decision and priming tasks on German adjective inflection in the light of the theoretical controversy between associative and combinatorial models of inflection. Before turning to the experiments, we will provide a brief description of adjective inflection in German. 4.1. ADJECTIVE INFLECTION IN GERMAN. German attributive adjectives carry a Portmanteauaffix that expresses the grammatical features gender, number, and case. With respect to the morphological expression of these features, two declension classes are commonly distinguished, weak and strong declension as shown in (5); see e.g. Wunderlich (1987). Adjectives which are used without a determiner or a demonstrative are combined with an uninflected determiner carrying a 'strong' affix, e.g. (ein) kalter Wein '(a) cold wine', while adjectives that are combined with a strongly inflected determiner take an affix from the weak paradigm, e.g. der kalte Wein 'the cold wine', mit dem kalten Wein 'with the cold wine', mit einem kalten Wein 'with a cold wine'. 13 (5) a. Strong adjective declension Singular Plural Masculine Neuter Feminine Nominative -r -s -e -e Accusative -n -s -e -e Dative -m -m -r -n Genitive -n -n -r -r b. Weak adjective declension Singular Plural Masculine Neuter Feminine Nominative -e -e -e -n Accusative -n -e -e -n Dative -n -n -n -n Genitive -n -n -n -n As is clear from (5), the affixes differ with respect to the degree of homonymity. Both, -e and -n occur in the strong and weak declension, while -s, -r, and -m only occur in the strong declension. Several linguists (Bierwisch 1967, Zwicky 1986, Blevins 1995, 2000, Wunderlich 1997) have analyzed this system using morphological paradigms. In these accounts, morpho-syntactic properties are represented in terms of binary features with marked (= positive) and unmarked (= negative) values. The distribution of these feature values is governed by a morphological Blocking or Specificity condition according to which forms with positive feature values take precedence over unmarked forms. Moreover in order to reduce lexical redundancy, the distinct forms of a paradigm are proposed as having underspecified entries, i.e. minimally specified analyses. A common assumption is that only positive feature values are directly specified, while negative feature values are introduced into paradigms in opposition to a marked positive value on the same feature (see e.g. Wunderlich 1997, Blevins 2000). For example, plural forms of nouns are said to be positively 14 specified for number, i.e. [+PL], whereas singular nouns are not directly specified; instead, a noun receives the unmarked singular interpretation, i.e. [-PL], by its paradigmatic opposition to the positively specified forms. In Bierwisch's (1967) and Blevins' (2000) accounts of German adjective inflection, the forms -e and -s have only negative feature specifications, while -m has two positive features, as shown in (6). Thus, in these analyses the -m affix comes out as the most specific form. Moreover, as is clear from (6), -s is specified for more features than -e, indicating that -s is more specific in terms of its feature content than -e (see also Cahill & Gazdar 1997). (6) -e -s -m [-OBL] - [-PL] [-FEM] [-MASC] [-OBL] - [-PL] [-FEM] [+OBL] [+DAT] Wunderlich's (1997) account relies on specifications of the structural cases in terms of a hierarchy of Theta-roles as shown in (7) for the four cases of German. Here [+hr] stands for 'there is a higher role' indicating that this case links to a lower Theta-role, and [+lr] stands for the reverse; [+nominal] means that this case links to the Theta-role of (relational) nouns. (7) Nominative: [ ] Accusative: [+hr] Dative: [+hr, +lr] Genitive: [+hr, +nominal] Using this system, a comparison of the adjective affixes in (6) reveals that in contrast to all other forms, the -m affix of the strong declension paradigm is restricted to just one case, dative, whereas, for example, -s occurs in nominatives and accusatives. Moreover, in terms of (7), dative requires two features for its specification, whereas nominative is completely 15 unspecified and accusative has just one feature. Thus, like in Bierwisch's and Blevins' analyses, Wunderlich's morpheme-based account yields -m as a more specific adjective ending than e.g. -e and -s. Summarizing, despite various differences between the linguistic treatments of German adjective inflection mentioned above, there seems to be agreement that the -m affix is paradigmatically more specified than any of the other adjective forms, and that the feature content of -s is more specific than that of -e. Note that this also fits in with frequency differences between the various adjective forms. Schriefers et al. (1992:376) observed that '-e is the suffix with the highest frequency in the inflectional system of German adjectives', and this corresponds to its low paradigmatic specificity. As far as the adjective affixes -m and -s are concerned, we compared their type and token frequencies in the CELEX database. On both counts, -s comes out considerably more frequent than -m (-m, types/tokens: 1,264/5,965; -s, types/tokens: 2,286/12,828)3. Hence, the lower paradigmatic specificity of -s relative to -m corresponds to higher frequencies of use. The empirical question we will address here is whether the linguistic differences between the various adjective forms have any effect on morphological processing. If this is the case, we would expect to find two experimental effects: specific affixes should produce longer lexical decision times and should be more difficult to prime than less specific affixes. Consequently, in a word/non-word (lexical) decision task, adjectives inflected with -m should exhibit longer response times than the same adjectives inflected, for example, with -s. This is because -m is the more specific form, and the mapping of the form to its corresponding feature bundle is likely to cause a longer lexical search. Moreover, we would expect to find that in a priming 3 These and all subsequent frequency counts are based on the CELEX corpora of written and spoken German (Baayen et al. 1993). 16 task adjectives inflected with -m should be less effectively primable than adjectives inflected with -s, and adjectives with -s less effectively than those with -e. This is because forms affixed with -m require the processing of specific, i.e. unprimed features which should lead to longer response times. Similarly, -s adjective forms contain features that cannot be primed by -e adjectives, whereas -e adjective forms do not have any features that could not be primed by -s. Therefore, if the morphological feature content of these adjective forms matters for processing, we would expect to find corresponding asymmetries between -s and -e adjective forms in priming. 4.2. EXPERIMENT 1: A VISUAL LEXICAL DECISION TASK ON INFLECTED ADJECTIVES. The experimental paradigm we employed here was a word/non-word discrimination task with reaction time (RT) as the dependent variable. Lexical decision times on non-inflected simplex words have consistently been shown to be affected by word frequency: subjects take less time to decide that high-frequency items are existing words than they do for lowfrequency items (see Balota 1994 for review). This is conceived of as a memory effect: as memory traces get stronger with additional exposures, high-frequency entries can be more readily accessed than low-frequency ones. We have used this task to examine the processing of inflected adjectives and to test for potential differences between specific and less specific affixes in their lexical decision times. We examined adjectives inflected with -m and -s. Recall that the latter form is approximately twice as frequent as the former. However, despite the overall low frequency of the -m affix compared to -s, some adjectives appear more often with -m than with -s; this is, for example, the case for ruhig 'quiet', as shown in (8). Compare this with an -s dominant adjective such as rein 'pure', which appears more often with -s than with -m, also shown in (8): (8) 17 -m dominant adjectives ruhig -s dominant adjectives stem form -m -s 838 51 13 stem form -m -s 783 14 38 rein By using -m dominant and -s dominant adjectives with both -m and -s in a lexical decision task, the potential effects of affix frequency and word form frequency can be teased apart. If inflected adjectives are stored as wholes, we would expect to find word form frequency effects. For example, ruhigem should produce shorter lexical decision times than ruhiges, whereas reines should produce shorter response times than reinem. If, on the other hand, adjectives are decomposed into stems and affixes, word form frequencies should be irrelevant, and we would instead expect to find affix frequency effects, i.e., adjective forms with -s should produce shorter response times than -m adjective forms (where stem frequency is held constant), because -s is the more frequent affix. The design of this experiment involves four conditions shown in (9) that allow us to compare response times to -m and -s forms of adjectives and to tease apart word form from affix frequency effects. (9) Low Word Form Frequency High Word Form Frequency Low Affix Frequency -s dominant adjective with -m -m dominant adjective with -m High Affix Frequency -m dominant adjective with -s -s dominant adjective with -s We predict that -s forms of adjectives will produce shorter lexical decision times than -m forms, because -m is the most specific adjective form in the paradigm and should therefore require a longer lexical search than the less specific -s form. Associative models of inflection, on the other hand, would predict word form frequency effects, but no affix frequency effects. 18 Hence, adjective forms with high word form frequencies should produce shorter response times than the same adjectives with low word form frequencies. Differences in affix frequency, however, should not affect lexical decision times. 4.2.1. MATERIALS. We selected 18 -m dominant and 18 -s dominant adjectives, which were matched for stem frequency. Each of these adjectives was presented with -s and with -m. Because no participant should see the same adjective more than once, two experimental versions were constructed. The 18 -m dominant and 18 -s dominant adjectives were divided into two groups, matched for mean stem and word form frequency (see Appendix A1). The two inflectional variants of the two types of adjectives were then distributed over two experimental versions in a Latin square design (Winer 1971). Each version included 18 -m dominant adjectives (9 of them inflected with -m and 9 of them with -s) and 18 -s dominant adjectives (9 of them combined with -m and 9 of them with -s). In addition to the 36 experimental items, we constructed 140 filler items (52 words and 88 pseudo-words), which should obscure the regularities in the test items. The list of word fillers consisted of adjectives inflected with the affixes -s, -m, -e, and -r so that each affix appeared equally often within the total list of 88 word stimuli. The pseudo-word fillers were created from real words by replacing two adjacent letters in varying positions. The real words used as a basis for the pseudo-words showed the same distribution of inflectional affixes as the 36 test items and 52 word fillers. The entire stimulus set for each experimental version consisted of 176 items. The two lists were pseudo-randomized, with the same order of test and filler items in both lists. 19 4.2.2. METHOD. 30 native speakers of German, all of them students at the University of Düsseldorf (mean age: 28) were paid for their participation in the experiment. None of them participated in more than one experimental version. Each trial consisted of the presentation of a fixation point in the middle of a computer screen in front of the participant, followed after 1,000 ms by the stimulus at the same position. The stimuli were presented on a 17'' computer monitor in white letters (Arial 24pt) on a dark background and remained on the screen for 200 ms. The participants reacted by pressing a green button (for a word) or a red button (for a pseudo-word) on a dual box. The green button was on the right side for right-handed and on the left side for left-handed participants. The next trial was initiated 1,200 ms after the response4. A written instruction with a detailed description of the task and some examples for inflected adjectives and pseudo-adjectives were given to the participants before the experiment. The experiment itself started with a short practice phase, after which the participants were given the opportunity to ask any remaining questions about the procedure. There were no further breaks during the experiment. The overall duration of an experimental session was about 10 minutes. Errors, i.e. non-word responses to existing words and word responses to non-words, as well as extreme reaction time values were removed from the data set before any further analysis of the reaction times5. Incorrect responses and outliers made up less than 8.3 % of the data set. Errors were analyzed separately. For the reaction time analysis we computed means for each subject and each item in each condition. These means were entered into two separate 4 5 For all experiments reported in this paper, the presentation of the stimuli and the measuring of the reaction times were controlled by the NESU software package (Baumann et al. 1993). In all our experiments, we used the same criterion for outliers. For each condition, extreme reaction times that exceeded two standard deviations of the subject's mean in this condition were removed prior to any further analysis of the reaction time data. 20 ANOVAs for subjects (F1) and for items (F2), with the factors 'Dominance' (-m dominant vs. s dominant) and 'Affix' (-s vs. -m). 4.2.3. RESULTS. The mean reaction times for the four experimental conditions are shown in Fig.1. Fig.1 Mean reaction times 630 614 mean RT (in ms) 609 600 568 570 564 540 510 -m -m dominant -s -s dominant Fig.1 shows that adjectives inflected with -s consistently produced shorter lexical decision times than adjectives co-occurring with -m, irrespective of whether word form frequencies were -m dominant or -s dominant. This was also confirmed statistically. There was a significant main effect of 'Affix' for subjects and for items (F1(1,29) = 29.55, p < .001, F2 (1,17) = 14.55, p = .001). Pairwise statistical comparisons using matched t-tests confirmed that reaction times for adjectives affixed with -s were significantly shorter than reaction times for adjectives with -m, both for -m dominant adjectives and for -s dominant adjectives6. The 6 In this and all subsequent experiments, t-values were determined separately for subjects and for items. RT differences are identified as 'significant', if p < 0.05 for subjects and for items. The t-values for all experiments reported in this article can be made available 21 results indicate that both -s and -m access their own representations in the mental lexicon and that (as predicted) accessing of -m produces longer response times than accessing of -s. We attribute this difference to the fact that -m is the more specific and less frequent form. Furthermore, there were no differences in the response times between -m dominant adjectives on the one hand and -s dominant adjectives on the other hand. Statistically, there was no significant effect of 'Dominance' (F1(1,29) < .01, p = .980, F2(1,17) = .02, p = .880) and no significant 'Dominance' x 'Affix' interaction (F1(1,29) = .42, p = .524, F2(1,17) = .02, p = .897). This means that -m dominant adjectives such as ruhig did not produce significantly shorter reaction times when they were presented with -m than when they were presented with -s. The error analysis revealed that adjectives co-occurring with -m produced significantly more incorrect responses (7.9 %) than adjectives affixed with -s (2.0 %), i.e. there was a significant main effect of 'Affix' (F1(1,29) = 7.18, p = .012; F2(1,17) = 29.21, p < .001). However, error rates did not differ significantly with respect to 'Dominance', that is -m dominant adjectives produced almost the same error rate (5.0 %) as -s dominant adjectives (5.3 %). This pattern is consistent with the observed reaction times. Thus, taken together, we found that word form frequency did not affect lexical decision times or error rates. This finding provides evidence against whole-word based representations for inflected word forms such as those posited in associative models of the inflection (see section 2). Instead, reaction times and error rates were influenced by the form of the inflectional ending. The more specific and less frequent affix -m produced longer reaction times and higher error rates than the less specific and more frequent -s ending. upon request. 22 4.3. EXPERIMENT 2: A CROSS-MODAL PRIMING TASK ON INFLECTED ADJECTIVES. While the results of the lexical decision experiment provide evidence against an associative full-listing representation of inflected adjectives, they do not allow us to decide whether the processing advantage of the -s affix (over -m) is due to its higher frequency or whether it is caused by its lower paradigmatic specificity. To further investigate the mental representation of inflected adjectives, we have used the cross-modal priming paradigm. As will become clear in what follows, results from priming experiments are less directly affected by frequency differences than results from the lexical decision task. In a priming experiment, subjects are asked to perform a word/non-word discrimination, just as in a lexical decision task, but the context in which the target stimuli are presented is manipulated. Consider, for example, the items in (10). In the experimental condition, a word is presented that is morphologically related to the target, e.g. happiness, followed by the presentation of the target word (happy) to which subjects have to make a word/non-word lexical decision. The response times to the target under this primed condition are then compared with those to an unprimed control condition, i.e. happy preceded by e.g. careful, and/or with those to an identical repetition of happy. The effect may be facilitatory or inhibitory. (10) Primes Target Identity Morph. Related Control happy happiness careful happy We adopted the cross-modal immediate repetition priming paradigm (Marslen-Wilson et al. 1993, 1994) in which subjects hear a spoken prime immediately followed by a visually presented target form to which they make a word/non-word decision. This technique has three main advantages. Firstly, since the task is cross-modal, any priming effects are likely to be due to the lexical representations themselves, rather than to effects of modality-specific 23 access procedures. Secondly, since all targets are presented immediately at the offset of the prime, the task is likely to tap on-line processes of morphological priming, while unwanted effects of episodic memory are reduced. Thirdly, results from priming experiments are less directly affected by frequency than results from lexical decision tasks. This is because in a priming experiment we are not directly comparing two different inflected word forms. Instead, morphological priming effects are measured within target sets, that is by comparing response times to the same targets after the presentation of different primes. Thus, in contrast to a lexical decision task in which response times to -m and -s adjectives have to be compared directly, we can examine priming effects separately for -m and for -s adjective forms. In this way, potential effects of the morphological differences between -m and -s forms can be studied independently of frequency differences. The priming conditions are illustrated in (11) for the adjective kariert 'chequered'. Each adjective occurred in three morphological variants (-e, -m, -s), both as a prime and as a target. Priming condition I provided the baseline for each of the three target forms, identical repetition. In priming conditions II and III, different paradigmatically related adjective forms were presented, with condition III containing more specific forms than those of condition II. Compare, for example, priming conditions II-e and III-e, with -e forms as targets and -s versus -m forms as primes and recall from section 4.1 that -m is a paradigmatically more specific form than -s. 24 (11) Conditions Auditory Primes I-e (Identity) kariert-e II-e kariert-es III-e kariert-em I-s (Identity) kariert-es II-s kariert-e III-s kariert-em I-m (Identity) kariert-em II-m kariert-e III-m kariert-es Visual Targets kariert-e kariert-es kariert-em We expect the priming patterns to correspond to the feature specifications of the affixes involved. Given the highly specific feature content of -m, we predict a 'Target Type' effect, i.e. adjectives inflected with -m should be less effectively primable than adjectives inflected with -s or -e. Adjectives inflected with -m are specified for the positive features [+OBL] and [+DAT] which are not available from any of the primes. The presentation of -m adjective forms as targets therefore requires the processing of unprimed features, and this should lead to longer response times. We also expect to find an interaction between prime type and target type. Priming conditions II and III should produce longer response times for the different targets than priming condition I. This is because in conditions II and III, the feature specifications of the affixes in the primes differ from those of the targets, whereas in condition I they are identical. 4.3.1. MATERIALS. 81 monomorphemic adjective stems (see Appendix A2) were chosen, all of which had low stem frequencies of less than 100 in the CELEX database (Baayen et al. 1993). For each of these 81 stem forms, nine prime-target pairs were constructed as shown in (11), resulting in 729 experimental prime-target pairs. To ensure that no participant sees the 25 same target-stem more than once, nine experimental versions were constructed in a Latin Square design. The mean stem frequencies and the mean number of syllables were held constant for each of these nine versions. Thus each version included 81 different prime-target pairs (9 from each of the 9 conditions shown in (11)). No target appeared more than once in any version. In order to keep the proportion of related pairs low and to deter the participants from developing strategies based on expectations about likely relations between primes and targets, 669 filler items were included in the experiment. We constructed 294 prime-target pairs in which primes and targets had different adjective stems, but were otherwise parallel to the experimental items, e.g. abstrakt-es → festlich-e 'abstract → festive'. These items were included to counter-balance the experimental items in which both primes and targets had the same adjective stem. We also constructed 375 prime-target pairs in which the target adjective was a pseudo-word. The pseudo-adjectives were constructed by changing two or three letters of an existing adjective. In 50 of these distractor items, the stems used as primes and targets were formally similar, e.g. anonymen → anonystes, populäres → spopulärem. The distractors were constructed in such a way that the different adjective endings (-e, -s, -m, -r, and -n) appeared equally often in each combination and that all inflectional affixes appeared equally often as primes and targets. In total, the stimulus set of each experimental version consisted of 750 prime-target pairs. In order to eliminate undesired priming effects across items, all prime-target pairs were pseudorandomized making sure that no semantic associations of any kind existed between consecutive items and that not more than four items of the same type occurred consecutively. Each of the nine versions exhibited the same order of test and filler items. 26 4.3.2. METHOD. 63 native speakers of German (mean age: 25) participated in the experiment for payment. None of them participated in more than one experimental version. The primes were spoken by a female native speaker of German and compiled into separate audio wav-files. The auditory stimuli were presented over headphones. The sequence of stimulus events within each trial was as follows: A short attention tone (250 ms) preceded the presentation of a fixation point (800 ms), followed by the auditory prime word. Immediately at the offset of the (spoken) prime, the visual target was presented and remained on the computer screen for 200 ms. Reaction times were measured from the presentation of the target onwards. During the experiment three breaks were provided. During each break and at the end of the experiment, the participants were asked to read a list of 15 words, and to mark those words they had heard during the experiment. For each of these lists, 9 words had been presented as auditory primes in the preceding experimental phase. The remaining 6 words did not occur in the experiment at all. The purpose of this task was to ensure that the participants paid attention to the auditory stimuli. The overall duration of the experiment was approximately 1 hour per subject. All other procedures were parallel to those of experiment 1. Erroneous lexical decisions as well as extreme reaction times were excluded from the database. The data excluded made up 4.9 % of the total responses. Mean response times for each subject and each condition were entered into two separate ANOVAs for subjects and for items with the factors 'Prime Type' and 'Target Type'. 4.3.3. RESULTS. Fig.2 presents the mean lexical decision times on the three visual target forms -e, -s, and -m and the three auditory prime forms -e, -s, and -m; see (11) for examples. 27 Fig.2 Mean reaction times to visual targets 560 549 550 543 mean RT (in ms) 540 533 527 530 516 520 510 533 509 508 513 500 490 480 Target-Affix -e Prime-Affix -e Target-Affix -s Prime-Affix -s Target-Affix -m Prime-Affix -m As expected, the shortest RTs were produced when both prime and target forms were identical. Fig.2 shows that this was the case for all the target forms tested. Moreover, Fig.2 shows that the target affix -m elicited longer RTs than the two other adjective forms. 28 Both ANOVAs produced significant main effects of 'Target Type' (F1(2,124) = 4.95; p = .009, F2(2,160) = 5.58; p = .005) and a significant 'Prime Type' × 'Target Type' interaction (F1(4,248) = 10.96; p < .001, F2(4,320) = 13.18; p < .001). There was no significant main effect of 'Prime Type' (F1(2,124) = 1.27; p = .283, F2(2,160) = .51; p = .306). In order to examine these effects further, we compared the overall mean RTs shown in Fig.2 using matched t-tests. Pairwise statistical comparisons of the RTs to the Identity condition of each of the three target affixes revealed no significant differences (-e: 509 ms, -s: 508 ms, -m: 513 ms); see footnote 6. This means that the 'Target Type' effect cannot be due to different response times to the Identity condition. Significant differences were, however, found when comparing the target affix -m to the two other target affixes: -m adjective forms produced significantly longer RTs when primed by -e than visual targets with -s primed by -e (549 ms vs. 527 ms). The same holds for -m as a target affix primed by -s, as opposed to the target affix -e primed by -s (543 ms vs. 516 ms). On the other hand, visual targets with -e or -s produced the same mean response time when primed by -m (= 533 ms). These comparisons show that the 'Target Type' effect is due to the longer response times for adjective forms with -m, confirming the prediction that adjectives with -m should be more difficult to prime than the other adjective forms. Furthermore, the lack of a main effect of 'Prime Type' indicates that none of the three adjective forms is overall a better or weaker prime than any of the other adjective forms tested. Pairwise comparisons revealed no significant differences between -e and -m forms used as primes for visual targets with -s (527 ms vs. 533 ms), and likewise no differences between -e and -s forms used as primes for targets with -m (549 ms vs. 543 ms). A significant difference was observed only for target forms with -e, where primes with -m elicited longer RTs than primes with -s (533 ms vs. 516 ms). 29 The results of this experiment indicate that the priming patterns are determined by the feature specifications of the affixes involved. In the Identity condition, the feature specifications of the affixes in the visual targets are fully primed. Reduced priming (relative to the Identity condition) occurs in cases in which a target form contains unprimed features that are not available from the prime. To see this more clearly, (12) shows the unprimed features of the target forms for each of the prime-target pairs tested in our experiment; the feature specifications are taken from the analysis in (6) above. Moreover, (12) shows for each primetarget pair the difference in response times to the corresponding Identity condition. These differences were also tested statistically using t-tests; * indicates a significant difference to the Identity condition. (12) Targets -e -s -m -e - [-PL] [-FEM] [-MASC] 19 ms* [-PL] [-FEM] [+OBL] [+DAT] 36 ms* -s 7 ms n.s. - [+OBL] [+DAT] 30 ms* -m [-OBL] 24 ms* [-MASC] [-OBL] 25 ms* - Primes Consider the prime-target pairs -e→-m and -s→-m which produced the most significant reduction in priming compared to the Identity condition. The targets in these two conditions 30 have unprimed positively specified case features that do not match the unmarked [-OBL] feature of the primes, hence the reduced priming effect. Consider next the prime-target pairs -e→-s and -m→-s, which also yielded reduced priming. In the case of -e→-s, the targets have unprimed gender and number features, and for -m→-s there is an unprimed gender feature [-MASC] in the target, in addition to a case feature mismatch between prime and target. In the prime-target pair -m→-e, there is a case feature mismatch ([+OBL] vs. [-OBL]), and hence a significant reduction in priming. For -s→-e, however, there are no unprimed features in the target form, and correspondingly, -s adjective forms do not have any inhibitory effect on priming -e target forms. Note additionally that prime-target pairs that involved -m target forms produced larger differences to the corresponding Identity condition than prime-target pairs with -e or -s target forms, suggesting that positively specified features are harder to prime than negatively specified ones. 4.4. PRELIMINARY SUMMARY. The purpose of the two experiments reported in this section was to examine how regularly inflected and paradigmatically related word forms are represented in the mental lexicon. Investigating inflected adjectives in German, we found that these word forms are decomposed into stems and affixes and that observed priming patterns depend upon the morpho-syntactic feature content of the affixes involved. We argued that a morphological analysis of German adjective inflection along the lines of (6) can account for the results of both experiments. 5. EXPERIMENTS ON STRONG VERB INFLECTION. In this section, we will investigate the processing of inflected forms of the so-called 'strong' verbs of German. While weak (= regular) verbs employ the same stem in all their inflected forms, strong verbs exhibit stem changes in the preterite and/or in the participle, e.g. werfen-warf-geworfen 'to throw-threw- 31 thrown'. This makes them parallel to irregular verbs in English, in that in both languages the marked stem forms are largely unpredictable from their corresponding infinitive forms. In contrast to English, however, the marked stems of strong verbs in German can be combined with fully regular person and number affixes in much the same way as weak stems, yielding forms such as warf-st 'threw-2nd sg.' and warft-t 'threw-2nd pl'. The first question we will examine in the experiments to be reported below is whether inflected forms of strong verbs are processed via their constituent morphemes or whether they are stored as wholes. It will be shown that the experimental results provide further evidence for morphological decomposition of regularly inflected word forms. The second question we will address concerns the mental representations of different stem forms of the same lexeme. We present results from lexical decision and priming experiments to show that stem forms are stored separately from the affixes with which they combine and that the various stem forms of a lexeme constitute a hierarchically structured lexical entry. Before turning to the experiments, we will provide a brief description of (strong) verb inflection in German. 5.1. STRONG VERB INFLECTION IN GERMAN. There are about 160 simplex verbs in German that belong to the strong class. These verbs have marked stems in present tense, preterite or participle forms. Most of them (= 155) fall into three minor classes, illustrated in (13); see Wunderlich & Fabri (1995) for a detailed classification. While A-B-A and A-B-C verbs have differently marked stem forms in preterite and participle forms, A-B-B verbs exhibit the same vowel change for both preterites and participles. Moreover, a large number of strong verbs have subjunctive forms with umlauted preterite stems, as shown in (13) for gab- / gäb- 'gave', flog- / flög- 'flew' and sang- / säng- 'sang'. There is also a small number of strong verbs in which 2nd sg. and 3rd sg. present tense forms as well as imperatives have fronted vowels (e.g. 32 werfen vs. er wirft 'to throw' vs. 'he throws'). Strong verbs also differ from weak verbs in that their participle forms have the ending -n, rather than the regular participle suffix -t. This holds for all strong verbs and none of the participle forms of weak verbs. Hence, strong verbs have a class-specific participle ending. (13) Infinitive Preterite Participle Subjunctive A-B-A geben gab- gegeben gäb- 'to give' A-B-C singen sang- gesungen säng- 'to sing' A-B-B fliegen flog- geflogen flög- 'to fly' Except for suppletive forms such as those of sein 'to be' and the forms of modals like dürfen 'to be allowed to', the marked stems of strong verbs are regularly inflected for person and number in the same way as the unmarked stems of weak verbs. Wunderlich (1996) has suggested a linguistic analysis of German verb inflection in which the morphological relationships between the various stem variants of strong verbs are represented in terms of non-monotonous default inheritance hierarchies. In this account, stem variants constitute subnodes of hierarchically structured lexical entries in which each subnode is defined in terms of a phonological string and a morphological feature, e.g. [..I..] vs. [..a..] and [±PRET] for wirf vs. warf. Consider, for illustration, the lexical entry for the German verb werfen 'to throw' from Wunderlich (1996: 96): 33 (14) [vεrf]+V [.I..]-1 [.a..]+PRET [...]+IMP [.y..X]+SUBJ [.o..n]+PART Each node in a structured lexical entry represents a pair (<phonological string, morphological feature value>), and each subnode inherits all information of its mother, except for the features it replaces or adds; for example, the subnode [..a..]+PRET inherits the onset w--, the coda -rf and the categorial feature [+V] from the higher node. The base or default stem form is werf-, the form that occurs in most present tense forms and the infinitive. The other stem variants occur under specific circumstances, e.g. wirf- for 2nd and 3rd sg. present tense forms and in imperatives, warf- in preterite forms, (ge)worfen in participles and würf- in subjunctives. Note that most stem variants have impoverished entries (to avoid unnecessary redundancies) and that the various stem forms are hierarchically structured. According to this account, stem forms are represented separately from the (regular) affixes with which they combine. We would therefore expect stem frequency effects in lexical decision and priming differences between the various stem variants corresponding to the structure of the lexical entries. On the other hand, associative models of inflection in which all inflected variants of a lexeme have full-form representations lead us to expect word form frequency effects for inflected forms of strong verbs and word form similarity effects. We will examine these predictions in two lexical decision experiments. 5.2. EXPERIMENT 3: A VISUAL LEXICAL DECISION TASK ON PRETERITE FORMS. In order to examine whether inflected forms of strong verbs are stored as wholes or whether they are 34 morphologically decomposed for processing purposes, we examined preterite forms of strong A-B-A, A-B-B, and A-B-C verbs in a visual lexical decision task. Recall from section 4.2 that lexical decision times are sensitive to frequency. Thus, if forms such as warf-en 'throw-1st/3rd pl.' are stored as wholes, we would expect lexical decision times to be affected by the frequencies of the word forms, i.e. high-frequency word forms should produce shorter response times than low-frequency word forms. If, on the other hand, regularly inflected forms of strong verbs are decomposed into stems and affixes, word form frequencies should be irrelevant, and the response times should be affected by the frequencies of the stem (where affix frequencies are held constant). 5.2.1. MATERIALS. 52 items from the three subclasses of strong verbs were selected, all of which were presented with the same affix, i.e. as 1st/3rd pl. forms, e.g. sprachen 'spoke' (see Appendix A3). To determine effects of preterite stem frequency, the experimental items were arranged in pairs, resulting in two conditions. The items in the first condition had a relatively low mean preterite stem frequency of 79, while the preterite stems in the second condition had a higher mean frequency of 135. To control for other frequency effects, the items in both conditions were matched for their mean word form and their mean verb frequencies. Verb frequencies were calculated as the sum of the token frequencies of each base verb plus all its lexical variants that have separable prefixes, e.g. sprechen 'to 'speak', an-sprechen, vorsprechen 'to speak to'. Variants with inseparable prefixes, e.g. besprechen 'to discuss', versprechen 'to promise', were not taken into account. Preterite stem frequencies included all preterite forms of each base verb and its forms with separable prefixes, e.g. an-sprachst, ansprach. Word form frequencies were the token frequencies of the inflected word forms presented in the experiment, e.g. the frequency of the word form sprachen. 35 In addition to the 52 experimental items, we included 348 filler items: 40 weak verbs, 16 verbs of the mixed class (e.g. bracht-en 'brought'), 92 other strong verbs, and 200 pseudoverbs which were constructed by changing two or three letters of an existing weak or strong verb. In order to neutralize potential affix effects, the filler items including the pseudo-verbs had the same ending as the experimental items, i.e. -en. The entire stimulus list consisted of 400 items and was presented in a pseudo-randomized order. 5.2.2. METHOD. 34 adult native speakers of German (mean age: 26) were tested. The methods and procedures were parallel to those of Experiment 1. Errors as well as extreme reaction time values were determined using the same criteria as in the previous experiments. The total amount of excluded data was 11.7 %. The mean response times for the remaining test items were compared using matched t-tests for subjects and for items. 5.2.3. RESULTS. As Fig.3 illustrates, there is a significant effect of preterite stem frequency: lexical decision times for verbs with high preterite stem frequencies were significantly shorter (49 ms) than recognition times for verbs with low preterite stem frequencies (t(33) = 5.97, p < .001 for 'subjects' and t(25) = 4.33, p < .001 for 'items'). 36 Fig.3 Mean reaction times mean RT (in ms) 650 640 630 610 591 590 570 550 low preterite stem frequency high preterite stem frequency The error data showed the same difference. For low-frequency preterite stems the mean rate of incorrect responses was 8.8%, whereas for high-frequency preterite stems it was 4.5%, a statistically significant difference (t(33) = 4.00, p < .001). The experimental effect found here cannot be attributed to the frequencies of the verbs used in the experiment or the frequencies of the inflected word forms as a whole, as both were held constant in the two experimental conditions. Instead, the stem frequency effect indicates that the verb forms presented in this experiment are decomposed into stem and affix. 5.3. EXPERIMENT 4: A VISUAL LEXICAL DECISION TASK ON A-B-B VERBS. In associative models of inflected word forms, morpho-syntactic features and categories are not directly encoded; see for example (3) in section 2. In combinatorial accounts of inflection, however, morphological features and morpho-syntactic categories play an important role. In Wunderlich's (1996) analysis of strong verbs in German, for example, the various subnodes of a hierarchically structured lexical entry, such as the one in (14), are defined in terms of morphological features and morpho-syntactic categories (in addition to phonological strings). 37 An interesting test case to assess the role of morpho-syntactic features in the processing of inflected verb forms is provided by A-B-B verbs such as lügen 'to lie' which exhibit the same stem change in preterite and participle forms (log- / ge-log-en). If morpho-syntactic features are encoded in the representations of these forms, preterite and participle forms of A-B-B verbs should be stored separately (regardless of their phonological overlap), and there should be corresponding experimental effects. In a lexical decision experiment, for example, we would expect to find separate frequency effects for the preterite and participle forms of verbs such as lügen. Alternatively, the mental representations of verbs may contain phonological but no morpho-syntactic features. In this case, A-B-B verbs such as lügen would have two stored stem forms (lüg- and log-), and one would expect lexical decision times for preterite and/or participle forms to vary according to the frequency of the B-stem. 5.3.1. MATERIALS. 30 A-B-B verbs were selected as experimental items, all of which were presented as 1st/3rd pl. forms, e.g. logen 'lied' (see Appendix A4). To the 30 experimental items, we added 370 fillers (170 existing verbs and 200 pseudo-verbs). As in Experiment 3, all verbs and pseudo-verbs had the same inflectional ending as the experimental items (= -en). The experimental items were arranged pairwise in two conditions according to their preterite stem frequencies. The items in the first condition had a lower mean preterite stem frequency (= 101) than the items in the second condition (= 165). In addition, the items in both conditions were matched for their mean B-stem frequencies, i.e. the mean frequency of all preterite and participle forms of each experimental item. We also determined verb frequencies for the experimental items, in the same way as for Experiment 3. In contrast to Experiment 3, however, it was not possible to match the experimental items on verb frequency. As shown in Appendix A4, the mean verb frequency of the items in condition I, i.e. those with low preterite stem frequencies, is higher than that 38 for condition II items, i.e. those with high preterite stem frequencies. This means that if verb frequency affects lexical decision, response times should be shorter for condition I than for condition II. If, on the other hand, preterite stem frequencies count, condition II should yield shorter lexical decision times than condition I, since condition II has higher preterite stem frequencies. Finally, given the restricted number of A-B-B verbs, it was also impossible to keep the word form frequencies constant for each pair. Recall from Experiment 3, however, that word form frequencies did not have an effect on the reaction times during the processing of regularly inflected preterite forms. Moreover, the mean word form frequencies for the items in the two conditions of the present experiment were in a similar range (17 for condition I and 22 for condition II), and such a small difference is unlikely to have effects on lexical decision times. 5.3.2. METHOD. 34 adult native speakers of German (mean age: 26) were tested. The methods, procedures and time settings were the same as for Experiment 3. Incorrect responses as well as outliers were excluded from the reaction time analysis. The total amount of excluded data was 11 %. The remaining mean response times were compared using matched t-tests for subjects and for items. 5.3.3. RESULTS. Fig.4 shows a significant effect of preterite stem frequency. Recognition times for A-B-B verbs with high preterite stem frequencies were significantly shorter (54 ms for subjects and 60 ms for items) than recognition times for A-B-B verbs with relatively low preterite stem frequencies (t(33) = 7.11, p < .001 for 'subjects' and t(14) = 3.05, p < .001 for 'items'). 39 Fig.4 Mean reaction times mean RT (in ms) 650 640 630 610 586 590 570 550 low preterite stem frequency high preterite stem frequency The error data showed the same difference. The mean error rate was 8.6 % for experimental items with low preterite stem frequencies and 3.9 % for items with high preterite stem frequencies, a statistically significant difference (t(33) = 3.08, p < .001) that corresponds to the observed differences in reaction times. These results cannot be attributed to the B-stem frequencies, as these were held constant in both conditions. Neither can they be explained in terms of verb frequency differences. It is true that the items in the two experimental conditions were not matched for verb frequency. If, however, this affected the response times, the results should show the opposite pattern, shorter RTs for the first condition (because of higher verb frequencies) than for the second one. Instead, the observed response time difference appears to be linked to the different preterite stem frequencies, indicating that preterite stems of A-B-B verbs access their own mental representations. 5.4. PRELIMINARY SUMMARY. In Experiments 3 and 4, we examined marked stem forms of strong verbs in German that were combined with regular person and number affixes. The results from both experiments provide evidence against whole-word based representations for 40 these kinds of inflected verbs and support the view that regularly inflected word forms are decomposed into stems and affixes, even if they contain marked (irregular) stems. The stem frequency effects we found in both experiments show that marked stem forms are represented separately from the inflectional affixes with which they may occur. The frequency effect for A-B-B verbs in Experiment 4 indicates that preterite stems are stored separately from participle forms despite their phonological overlap. This finding indicates that morphosyntactic features form part of the lexical representations of marked stem forms. 5.5. EXPERIMENT 5: A CROSS-MODAL PRIMING TASK ON STRONG VERBS. While response times in lexical decision tasks seem to be sensitive to properties of lexical entries, results from priming tasks provide more direct psycholinguistic measures for examining relationships between lexical entries or subentries in the mental lexicon. Several previous studies have provided evidence for morphological priming effects in the mental lexicon (see e.g. Stanners et al. 1979, Marslen-Wilson et al. 1994, Münte et al. 1999 for English, Sonnenstuhl et al. 1999 for German, Orsolini & Marslen-Wilson 1997 for Italian, Lukács & Pléh 1999 for Hungarian). It is clear from these studies that regularly inflected word forms produce priming effects which manifest themselves in shorter lexical decision times on prime-target pairs such as walked→walk compared to a baseline condition in which the same target is preceded by an unrelated word (spilled→walk). We made use of the priming technique to examine relationships between the different stem forms of a structured lexical entry focusing on the representation of strong verbs in German. The experimental items were inflected forms of strong verbs that were composed of a stem and the 2nd pl. suffix -t, yielding prime-target pairs such as warft→werft 'threw-2nd pl.→throw-2nd pl.' vs. werft→warft. Like the past tense -ed in English, the 2nd pl. -t is fully predictable and regular in German. We would therefore expect the -t ending to be stripped off 41 from word forms such as werf-t and warf-t and the priming patterns to depend upon the morphological relationships between the stem forms involved. If strong verbs are represented in hierarchically structured lexical entries (e.g. (14) for werfen), one would expect the stems' feature specifications to affect the priming patterns, such that marked stems (which are further down on the inheritance tree) should be more difficult to prime than less specific stems. Specifically, we predict that preterite stems (= warf-t) should be less effectively primable by present tense forms (= werf-t) than vice versa, because in terms of the representation in (14) the target form warft (preceded by the prime werft) contains an unprimed feature (= [+PRET]), whereas the target werft (preceded by warft) does not have any unprimed morphological features. 5.5.1. MATERIALS. 32 strong verbs (see Appendix A5) from different subclasses were used. For each of these verbs, four prime-target pairs were constructed as shown in (15). (15) Conditions Auditory Primes I (pres.→ pres.) ihr helft 'you help-2nd pl. pres.' II (pret.→ pres.) ihr halft 'you help-2nd pl. pret.' III (pres.→ pret.) ihr helft 'you help-2nd pl. pres.' IV (pret.→ pret.) nd Visual Targets helft halft ihr halft 'you help-2 pl. pret.' All experimental items contained the 2nd pl. suffix -t. Note that 2nd pl. present tense forms of weak verbs and of many strong verbs are identical to 3rd sg. forms, e.g. ihr kommt / sie kommt 'you-pl. come / she comes'. To nullify any effects of this ambiguity, we presented all primes together with the 2nd pl. form of the personal pronoun. Moreover, for the experimental items, we used only verbs with distinct 2nd pl. and 3rd sg. forms, e.g. verbs such as helfen 'to help' for 42 which in the present tense the 2nd pl. is helft and the 3rd sg. hilft. Because no participant should see the same target more than once, four experimental versions were constructed. The prime-target pairs were distributed over four versions in a Latin Square Design, so that each version included 32 different prime-target pairs (8 from each of the conditions I-IV), and each verb appeared only in one prime-target pair. In addition to the experimental primetarget pairs, 176 unrelated word/word filler pairs and 208 word/pseudo-word filler pairs were constructed. The list of word/word fillers included German verbs in tense and agreement forms other than those of the experimental items (156 pairs). Moreover, all word/word pairs were counterbalanced so that present tense and preterite forms as well as the various person and number agreement affixes appeared equally often during the experiment. Parallel to the 208 word/word pairs in the experiment, the same number of word/pseudo-word pairs (which were inflected in the same way as the real-word items) was constructed. In 12 of the pseudoword items, the prime was either fully contained within the target (e.g. ich formte pluformtelst 'I formed - pluformtelst') or partially overlapping with the target (e.g. ich flüsterte 'I whispered' - flüspurtet). These pairs were added to the stimulus list to ensure that not all phonologically related pairs had real words as targets. The overall list of 416 primetarget pairs was pseudo-randomized. Each of the four versions exhibited the same order of experimental and filler items. 5.5.2. METHOD. The methods, procedures, and time settings were the same as in Experiment 2. Participants were 44 native speakers of German (mean age: 26). In addition to errors and extremely long reaction times, one item (trefft / traft 'meet / met') had to be entirely excluded from any further analysis because of an error in the preparation of the stimulus list. The total amount of excluded data was 7.2 %. To test the prediction that preterite stems are less effectively primable by present tense forms than vice versa, two separate ANOVAs were 43 performed (for 'subjects' and 'items') with the factors 'Prime Type' (Identity vs. Test) and 'Target Type' (Preterite vs. Present). 5.5.3. RESULTS. Fig.5 presents the overall mean RTs to the visual targets in the four experimental conditions. Fig.5 Mean reaction times to visual targets 657 660 mean RT (in ms) 630 600 580 566 570 541 540 510 480 Target Present Prime Present Target Preterite Prime Preterite Conditions I and IV (in which primes and targets were identical) produced shorter response times than the two morphological priming conditions II and III (in which primes and targets were different). Moreover, condition III where priming went from present tense forms to preterite targets produced much less priming compared to the Identity condition (= 77 ms) than condition II (= 25 ms) where priming went from preterite to present tense forms. These differences were confirmed statistically. The ANOVAs revealed a 'Prime Type' x 'Target Type' interaction which was significant for 'subjects' (F1(1,43) = 6.46, p = .015) and marginally significant for 'items' (F2(1,30) = 3.77, p = .062). In both subject and item analyses, there were also significant main effects of 'Prime Type', F1(1,43) = 28.49, p < .001, F2(1,30) = 44.95, p < .001, and 'Target Type', F1(1,43) = 21.88, p < .001, F2(1,30) = 31.72, 44 p < .001. Matched t-tests revealed that the differences between conditions I and II as well as those between III and IV are statistically significant (see footnote 6). Moreover, the differences between the priming effects in the two morphological priming conditions II and III (25 ms vs. 77 ms) were statistically significant in the 'subjects' analysis (p = 0.015) and marginally significant for 'items' (p = 0.062). This confirms the predicted priming patterns: preterite forms prime present tense forms better than present tense forms prime a preterite target. In addition to the reaction times, we also analyzed the distribution of errors. A comparison of the mean error rates for conditions I and II to those for conditions III and IV revealed that preterite forms produced a significantly higher error rate than present tense forms (6.3 % vs. 0.4 %, t(43) = 3.21, p = .003). This difference is consistent with the reaction-time data, in particular with the finding that preterite forms are less easily primable than present tense forms. To explain the observed priming patterns, we can rule out the possibility that the form of the inflectional affixes had an effect on the response times. Recall that all experimental items in all conditions had the same person and number affix (the 2nd pl. -t). Hence, the form of the affix cannot account for the observed differences between present tense and preterite forms. It is more likely that they are caused by the morphological feature specifications of the stem forms involved, in the following way. Suppose that the regular 2nd pl. -t affix is stripped off from all the inflected word forms we presented and that the remaining stem forms are accessed separately. When a marked stem such as half- is presented as the visual target preceded by an instance of the unmarked stem (helf-), the target contains a feature (= [+PRET]) that is unavailable from the prime, and this unprimed feature produces the increased target response times. On the other hand, when an unmarked stem (helf-) is the 45 target preceded by a marked stem (half-), the target does not contain any unprimed features, and hence there are significantly shorter target response times than for preterite targets. 6. GENERAL DISCUSSION. In this section, we summarize our main findings and discuss potential implications for the controversy between combinatorial and associative approaches to inflection. Our focus will be on three core aspects of morphological processing and representation: (i) morphological decomposition of inflected word forms, (ii) the representation of morphologically related word forms, and (iii) the specification of morphological form in entries of the mental lexicon. 6.1. EVIDENCE FOR MORPHOLOGICAL DECOMPOSITION. Previous research on morphological processing indicates that regularly inflected word forms are decomposed into smaller morphological units, such as stems, roots and affixes (see Clahsen 1999 for review). Evidence for this comes, for example, from cross-modal priming experiments on past participles and noun plurals in German (Sonnenstuhl et al. 1999). We found full stem priming for regularly inflected -t participles (gelacht Æ lach ‘ laughed Æ laugh’) and regular -s plurals (Autos Æ Auto ‘cars Æ car’), but reduced stem priming for irregular (-n) participles (geschlafen Æ schlaf ‘slept Æ sleep’) and irregular (-er) plurals (Kinder Æ Kind ‘children Æ child’). We explained these differences in morphological terms: -s plural and -t participle word forms are decomposed into stem+affix, and can thus prime their base stem directly. Irregular plurals and participles, however, access full-form entries stored in memory and cannot directly activate their corresponding base entries; therefore the priming route is less direct. At the same time, alternative non-decompositional accounts of the observed priming differences could be ruled out, since the prime-target pairs in the regular and the irregular 46 conditions were matched with respect to their frequencies and their phonological/orthographical overlap. The findings reported in the present paper provide further support for morphological decomposition. In three lexical-decision experiments, we found separate affix and stem frequency effects indicating that affixes access their own representations in the mental lexicon. Experiment 1 showed that adjectives inflected with (the more frequent affix) -s yielded shorter response times than the same adjectives inflected with (the less frequent affix) -m. At the same time, the word form frequencies of the items had no effect on the participants' response times. In Experiments 3 and 4, we found stem frequency effects for regularly inflected word forms of strong verbs. Preterite forms of A-B-A, A-B-C, and A-B-B verbs with high preterite stem frequencies elicited significantly shorter response times than corresponding forms with low preterite stem frequencies. Again, as in Experiment 1, word form frequencies, i.e. the frequency of an inflected preterite form as a whole, did not influence the response times. Interestingly, lexical decision tasks examining participle forms of A-B-A and A-B-B verbs produced different results (Clahsen et al. 1997), even though the design of these experiments was parallel to that employed in Experiments 3 and 4 above. Lexical decision times for high frequency participle forms of both subclasses of strong verbs were found to be shorter than those of low frequency participle forms. At the same time, stem and verb frequencies did not affect response times. The participle forms of A-B-B verbs tested in Clahsen et al. (1997) were matched with respect to B-stem frequencies and verb frequencies in the same way as in Experiment 4. Response times to participles with low word form frequencies, however, showed a significant 58 ms delay compared to participles with high word form frequencies7. 7 Note that this was not the case for -t participle forms of weak verbs such as gekauf-t 'bought'. In contrast to the strong participle forms, there were no word form frequency 47 These findings show that the lexical decision times of participle forms of strong verbs are sensitive to word form frequency, rather than to stem frequency. To account for the observed differences between the processing properties of preterite and participle forms, recall that the preterite forms from Experiments 3 and 4 were suffixed with a regular person and number suffix, while the strong participle forms used in Clahsen et al.'s (1997) lexical decision experiments (and those used in Sonnenstuhl et al.'s (1999) priming experiments) were -(e)n participles such as gelog-en 'lied' in which -(e)n is a class-specific ending that is unique to members of the strong class. Wunderlich (1996) posited different subnodes within the lexical entries of A-B-B and A-B-A verbs to capture this difference. Consider, for example, the lexical entry of the A-B-B verb lügen 'to lie' in (16); the preterite stem is log-, the participle form gelogen, and the subjunctive has a fronted vowel formed on the basis of the preterite stem (lög-). (16) [ly:g]+V [.o.]+PRET [.o..n]+PART [.∅.]+SUBJ Our experimental findings indicate that preterite stems and participle forms are stored in separate subnodes, as indicated in (16). The word form frequency effects for strong participle forms show that these forms do not have independent stem representations, in contrast to strong preterites for which stem frequency effects were found. Moreover, our results indicate that what triggers morphological decomposition is the presence of a regular affix. This was the case in the preterite forms tested in Experiments 3 and 4, but not in the strong participle effects for weak participle forms. This was taken to indicate that regular -t participles are decomposed into (weak) stems and a -t participle affix (see Clahsen 1999 for 48 forms examined in Clahsen et al. (1997). The form of the stem, on the other hand, does not determine whether an inflected word is decomposed or stored as a whole. Decomposition effects are found for inflected word forms such as walked or gezeigt 'shown' which have both a regular inflectional ending and an unmarked (regular) stem form as well as for items that contain a regular (agreement) ending with a marked (irregular) stem form (see Experiments 3 and 4). Hence, it seems (as one would expect from combinatorial models of inflection) that regular inflectional affixes are stripped off inflected words, irrespective of the form of the stem. 6.2. MORPHOLOGICAL RELATIONSHIPS IN THE MENTAL LEXICON. To address the question of how relationships between different inflectional affixes and between different stem forms of the same lexeme are mentally represented, we have investigated inflected adjectives and preterite stem forms of strong verbs in two cross-modal priming tasks. The empirical generalization that comes out of the results of these experiments is that reduced priming occurs in cases in which the target forms contain feature specifications that are not available from the primes. A paradigmatic analysis of German adjective inflection was shown to account for the priming patterns observed in Experiment 2. Recall, for example, that -m target forms were more difficult to prime than any other adjective form. We interpreted this as a 'specificity effect': -m is specified for the morpho-syntactic features [+OBL] and [+DAT], and these features are not available from any other adjective form. For this reason, -m forms are hard to prime by other adjective forms, and hence the longer response times for -m target forms in our experiment. We also found an asymmetry in the priming patterns of -s and -e forms: -s primes a target form -e as effectively as an identical prime, i.e. there was no statistically discussion). 49 reliable difference between -s→-e and -e→-e. On the other hand, -e adjective forms prime -s adjective targets much less effectively than identical targets. These priming differences are also determined by the feature contents of the two affixes. The prime-target pairs -e→-s contain unprimed target features (for gender and number), whereas the target forms in -s→-e pairs do not have any features that are not already available from the primes. Evidence as to how relationships between different stem forms of the same lexeme are mentally represented comes from the results of Experiment 5. We found that preterite stems (e.g. =warf- 'threw') are less effectively primable by present tense stems (= werf-) than vice versa. This priming difference can be explained in terms of structured lexical entries, such as those in (14) and (16). In these entries, preterite stems are marked forms that are represented further down on the inheritance hierarchy than present tense stems. Preterite stems (primed by present tense forms) are left with an unprimed [+PRET] feature in the target, whereas present tense forms (primed by preterites) do not have any unprimed target features, and hence the different priming patterns. Thus, despite different representations for stems and affixes, the observed priming patterns are parallel, in that for affixes and stems reduced priming was found when the target forms contained unprimed morphological features. 6.3. UNDERSPECIFICATION IN THE MENTAL LEXICON. The notion of underspecification is used in phonological and morphological representations of lexical entries to reduce lexical redundancies. In phonological representations, for example, redundant feature values that do not contrast two segments are left unspecified (e.g. Archangeli 1988, for psycholinguistic evidence see Lahiri & Marslen-Wilson 1991). Underspecification is also used in morphological representations, e.g. in inflectional paradigms to reduce the number of syncretic forms. Our experimental findings indicate that the mental lexicon contains entries that are underspecified for morpho-syntactic features. 50 Inflectional affixes, for example, were assumed to have impoverished entries in which only positive (=marked) feature values are directly specified, while negative (=unmarked) feature values are introduced into paradigms in opposition to a positive value on the same feature. With these assumptions, the priming patterns observed in Experiment 2 could be explained. The alternative possibility that all affixes have fully specified representations including positive and negative feature values, does not account for our findings. Consider, for example, the priming asymmetries between -e and -s adjective forms. If both these affixes had fully specified entries, we would have unprimed features in both -e and -s target forms, and the observed priming difference between the prime-target pairs -e→-s (= reduced priming) and -s→-e (= full priming) would be left unexplained. Priming asymmetries were also found for preterite and present tense forms of strong verbs; preterite forms prime present tense forms more effectively than present tense forms prime preterites. This is compatible with the view that (like inflectional affixes) stem forms also have impoverished underspecified entries in which only preterite stems are specified for a tense feature. If all stem forms were fully specified (e.g. present tense forms as [-Pret] and preterite forms as [+Pret]), there would be an unprimed tense feature in the targets of both prime-target pairs, since the [-Pret] feature of present tense targets cannot be primed by preterite forms, and the [+Pret] feature of preterite targets cannot be primed by present tense forms. The observed priming asymmetries are not compatible with this account. Instead, our findings support the hypothesis that the mental representation of lexical forms utilizes underspecified representations. 6.4. ATTEMPTS INFLECTION. TO EXPLAIN THE EXPERIMENTAL FINDINGS WITHIN ASSOCIATIVE MODELS OF It was argued that combinatorial models of inflection account for the results from both lexical decision and cross-modal priming. In this section, we will explore how 51 associative models of the mental lexicon, such as those presented in section 2, might explain the experimental results. It will be shown that some of our findings present difficulties for these approaches. In associative models of inflection, inflected word forms are stored as wholes. Hence we would have expected word form frequency effects in the three lexical decision experiments, i.e. shorter response times for high frequency adjective and preterite forms than for those with low word form frequencies. On the other hand, one would not predict affix and stem frequency effects. The results of Experiment 1, 3 and 4 showed the opposite pattern. Affix frequency effects paired with the lack of word form frequency effects are problematic for any account that assumes whole-word representations for regularly inflected word forms (see Clahsen 1999 for additional evidence from noun and verb inflection). As for the results from the two priming experiments, the satellite model (see (2) in section 2) provides a partial account of our findings. Recall that this model distinguishes between a base form (= nucleus) of a lexeme and its inflected variants (= satellites), which are connected to the nucleus. Thus, if lexical access is made via a satellite form, the corresponding nucleus is also activated. A base form, however, does not necessarily activate all its satellites. With these assumptions, the priming asymmetries reported in Experiment 5 can be accounted for. Strong preterite forms are satellites in this model and activate their corresponding base forms; hence they prime present tense forms more effectively than vice versa. Note, however, that the stem and affix frequency effects found in the three lexical decision experiments are incompatible with the full-form representations the satellite model posits for inflected word forms. The results of Experiment 2 on adjective priming are also hard to explain for the satellite model. Recall that we found priming differences between various inflected forms of the same adjective, e.g. priming asymmetries between -s and -e forms and differences 52 between -m and -s/-e target forms. These experimental differences between the various satellites of a nucleus are left unexplained by the satellite model. In other associative models of inflection, all inflectional variants (including the base form) are assumed to have word-level status and to be associatively connected to each other (see e.g. (3) in section 2). Experimental differences between inflected word forms are explained in terms of frequency and similarity differences of the forms involved. Schriefers et al. (1992) argued along these lines suggesting that inflected adjectives in German have associative fulllisting representations in the mental lexicon. The evidence they reported comes from a unimodal (visual) delayed repetition priming task examining adjectives in four morphological variants, for example gute, gutem, gutes, gut 'good'. Their main finding was an asymmetric priming pattern for -m adjectives: -m forms primed other adjective forms as effectively as an identical prime, but when -m adjectives were the targets and other adjective forms appeared as primes, priming was reduced. Schriefers et al. reasoned that the decomposition hypothesis 'would not only require full priming of a stem by its morphological variant, but also […] full priming between its morphological variants' (p.375). As this was not confirmed in their experiment, Schriefers et al. argued instead that each morphological variant of a stem is fully represented as a word node and that the observed priming asymmetries follow from frequency differences, claiming that -m target forms are difficult to prime because of the relatively low frequency of these forms. We do not think that these conclusions are borne out. According to decompositional accounts of morphological processing, stems and affixes of regularly inflected words access their own mental representations. Hence, in prime-target pairs in which the stems are identical but the affixes (or rather their feature contents) contain unprimed features in the targets, we would expect to find reduced priming (and not full priming). This is indeed what our Experiment 2 53 showed. Schriefers et al.'s finding, on the other hand, that -m forms used as primes produced the same target response times as an identical prime is not replicated in our experiment and is probably an episodic memory effect, caused by the long delay (of 10 to 14 unrelated items) they introduced between primes and targets. Episodic memory effects are based on a participant's remembering a prior event and have been shown to be stronger when there is a long delay between prime and target, and when the participant has to react in the same way to both elements of a related stimulus pair (Tenpenny 1995). Both points apply to Schriefers et al.'s experiment. It is therefore conceivable that when presented with a target such as akutes, participants remember having seen another instance of the same adjective 10 to 14 items before without necessarily remembering whether it was akutes or akutem, and therefore Schriefers et al. were unable to find a difference between these two conditions. The crossmodal immediate repetition task we used in Experiment 2 is less affected by episodic memory effects (since there is no delay between primes and targets and participants react only to the visually presented targets while listening to auditory primes), and in this experiment we found reduced priming in all cases in which the target forms contained unprimed features. Schriefers et al.'s attempt to explain priming effects in terms of frequency differences does also not lead very far. The finding (from both Schriefers et al. as well as from our Experiment 2) that -m forms are less easily primable than other adjective forms might be explained in this way. If, however, frequency was the determining factor for the target response times, we would also expect -e targets to be primed more effectively than -s targets, since -e is the most frequent adjective ending in German, as Schriefers et al. (1992:376) pointed out themselves. This prediction is not confirmed, either in our Experiment 2 or in Schriefers et al.'s data. Fig.5 shows that the response times to -e targets (primed by -m) are identical to the RTs to -s targets (primed by -m); 533 ms in both cases. The same holds for Schriefers et al.'s data, 688 54 ms for -m→-e and 689 ms for -m→-s. This shows that frequency is unlikely to be the decisive factor in accounting for the observed priming differences. In addition to frequency, proponents of associative models of inflection argue that priming is determined by the degree of surface form similarity, i.e. the orthographic and phonetic overlap between prime and target. Regular past tense forms in English are, for example, orthographically and phonologically more similar to their base forms than many irregular past tense forms (compare walked Æ walk versus taught Æ teach), and it has been argued that this is the reason why regular past tense forms are more effective primes for their corresponding stems than irregular ones (see e.g. Rueckl et al. 1997). This surface similarity account, however, can only handle a small subset of the observed priming patterns reported in the present study. Consider, for example, the prime-target pair -s→-e which in our experiment produced a non-significant 7 ms increase in response times compared to the Identity condition (see (12)). One might argue that effective priming in this case is due to the high degree of orthographic and formal overlap between prime and target, as the target form (e.g. karierte) is fully contained in the form of the prime (e.g. kariertes). However, other prime-target pairs yielded reduced priming effects, -m→-e (kariertem→ karierte), even though primes and targets show the same degree of surface similarity as -s→-e. This shows that a simple surface-based account does not explain the experimental findings. 7. CONCLUSION. We have investigated the question of how morphological relationships between inflected word forms are represented in the mental lexicon. Our focus was on paradigmatic relations between regularly inflected word forms and relationships between different stem forms of the same lexeme. We have also examined whether inflected word forms are morphologically decomposed or stored as wholes. Results from five psycholinguistic experiments were presented examining the processing of inflected adjective 55 and verb forms of German. We explained our findings in terms of combinatorial approaches to inflection with morphological paradigms to represent regular affixes and structured lexical entries to represent different stem forms of the same lexeme. Moreover, for both stems and affixes, we relied on linguistic analyses that posit underspecified lexical entries, i.e. minimally specified analyses in which only positive (= marked) feature values are directly specified. To the extent that our interpretation can be maintained, the experimental findings reported here provide psycholinguistic support for these theoretical notions. Associative models of inflection, on the other hand, in which morphological structure and morphosyntactic features and categories are not encoded, provide only partial explanations of our findings. Appendix: Experimental Items A1. Adjectives used in Experiment 1 Frequency counts are given in parantheses (stem / word-form with -m / word-form with -s) (i) -m dominant adjectives: mean frequencies (402/16/9) pur 'pure' (25/3/1), grell 'glaring' (35/3/1), spitz 'pointed' (46/4/1), sauer 'sour' (53/3/0), steif 'stiff' (82/7/0), schräg 'sloping' (104/5/0), flüssig 'liquid' (111/12/7), nass 'wet' (113/7/4), lebhaft 'lively' (281/27/19), dicht 'dense' (339/11/6), hell 'bright' (345/13/9), kalt 'cold' (503/24/17), rasch 'quick' (522/8/5), scharf 'sharp' (640/15/9), direkt 'direct' (706/14/10), ruhig 'quiet' (838/51/13), gering 'little' (929/31/9), frei 'free' (1560/42/38) (ii) -s dominant adjectives: mean frequencies (397/7/17) stumpf 'blunt' (25/1/4), fett 'fat' (34/0/3), grotesk 'grotesque' (48/1/4), dumpf 'dull' (57/1/7), toll 'great' (83/0/6), stabil 'stable' (101/1/4), süß 'sweet' (96/2/12), roh 'raw' (120/3/7), gesund 'healthy' (289/11/26), leer 'empty' (315/6/15), frisch 'fresh' (346/10/20), echt 'genuine' (523/12/29), schwach 'weak' (523/4/12), hart 'hard' (633/19/25), sozial 'social' (778/7/13), rein 'pure' (783/14/38), schlecht 'bad' (889/21/43), klar 'clear' (1510/16/31) A2. Adjectives used in Experiment 2 absurd 'absurd', adrett 'neat', akkurat 'precise', akut 'acute', albern 'silly', arrogant 'arrogant', barock 'baroque', blank 'shiny', brav 'well-behaved', brillant 'brilliant', brisant 'explosive', brutal 'brutal', delikat 'delicate', derb 'coarse', dezent 'discreet', diskret 'confidential', dominant 'dominant', dreist 'bold', dumpf 'dull', elend 'wretched', exklusiv 'exclusive', fair 'fair', fatal 'fateful', feucht 'damp', flink 'nimble', formal 'formal', frech 'cheeky', galant 'gallant', gemein 'mean', genial 'ingenious', global 'global', grell 'glaring', grotesk 'grotesque', harsch 'harsh', heiser 'hoarse', herb 'bitter', human 'humane', intakt 'intact', intern 'internal', kahl 'bare', kaputt 'broken', karg 'meagre', kariert 'chequered', kompakt 'solid', konfus 'confused', korrekt 'correct', korrupt 'corrupt', labil 'weak', legal 'legal', linear 'linear', mager 'lean', markant 'clear- 56 cut', mies 'rotten', mobil 'mobile', naiv 'naive', nuklear 'nuclear', oval 'oval', paradox 'paradoxical', pikant 'piquant', plump 'clumsy', rabiat 'violent', rapid 'rapid', rasant 'rapid', rauh 'rough', riskant 'risky', satt 'full', schief 'crooked', schlapp 'limp', schlau 'smart', schrill 'shrill', spontan 'spontaneous', steif 'stiff', steil 'steep', stumpf 'blunt', stur 'stubborn', tapfer 'brave', tolerant 'tolerant', trist 'dismal', verwegen 'daring', violett 'violet', zivil 'civil'. A3. Verb forms used in Experiment 3 Frequency counts are given in parantheses (stem / word-form / preterite stem) (i) verbs with low preterite stem frequency: mean frequencies (411/16/79) soffen 'drank' (8/0/3), wanden 'wound' (24/3/6), schliffen 'ground' (30/2/6), schwanden 'dwindled' (33/0/4), stanken 'stank' (40/2/9), fochten 'fenced' (42/3/6), priesen praised' (52/2/12), schworen 'swore' (69/6/19), verdarben 'spoilt' (76/0/8), gruben 'dug' (87/6/19), logen 'lied' (105/0/18), warben 'recruited' (124/4/14), pfiffen 'whistled' (135/13/58), stritten 'quarreled' (144/16/30), schieden 'separated' (276/10/39), banden 'tied' (385/7/39), schnitten 'cut' (386/20/87), sangen 'sang' (450/34/116), tranken 'drank' (474/39/146), luden 'loaded' (646/22/100), rissen 'tore' (661/53/282), brachen 'broke' (830/40/207), wuchsen 'grew' (935/43/165), glichen 'resembled' (1014/8/40), schufen 'created' (1479/48/149), schrieben 'wrote' (2160/41/473). (ii) verbs with high preterite stem frequency: mean frequencies (415/19/135). kniffen 'pinched' (22/0/8), schlangen 'wrapped' (27/3/19), flochten 'plaited' (24/2/10), sannen 'pondered' (31/1/15), mißlangen 'fail' (30/1/14), mieden 'avoided' (40/6/11), liehen 'lent' (66/2/14), schlichen 'sneaked' (74/11/40), erschraken 'scared' (72/3/51), stachen 'stung' (82/8/39), rieben 'rubbed' (96/8/46), bliesen 'blew' (105/4/41), rochen 'smelt' (135/6/73), schritten 'strode' (207/23/70), litten 'suffered' (254/23/78), empfahlen' recommended' (382/6/68), schwiegen 'remained silent' (360/18/133), empfingen 'received' (450/21/156), sprangen 'jumped' (516/42/229), zwangen 'forced' (668/25/112), trieben 'drifted' (631/40/160), stießen 'pushed' (815/64/322), fingen 'caught' (971/39/257), griffen 'grabbed' (1021/54/337), riefen 'called' (1514/45/524), schienen 'shone' (2050/36/634) A4. Verb forms used in Experiment 4 Frequency counts are given in parantheses (stem / word-form / B-stem / preterite stem) (i) verbs with low preterite stem frequency: mean frequencies (72/17/253/101) woben 'wove' (15/0/9/2), schliffen 'ground' (30/2/17/6), schmissen 'threw' (33/2/20/7), fochten 'fenced' (42/3/23/6), priesen 'praised' (52/2/27/12), schworen 'wore' (69/14/38/19), logen 'lied' (105/0/48/18), bissen 'bit' (144/4/75/52), strichen 'painted' (222/7/137/66), schieden 'separated' (276/10/134/39), schnitten 'cut' (386/20/218/87), glichen 'resembled' (1014/6/86/40), boten 'offered' (1643/50/651/291), schrieben 'wrote' (2160/41/1071/473), schlossen 'closed' (2393/90/1235/404) (ii) verbs with high preterite stem frequency: mean frequencies (444/22/251/165) kniffen 'pinched' (22/0/10/8), flochten 'plaited' (24/2/18/10), mieden 'avoided' (40/6/19/11), schmolzen 'melted' (62/4/26/13), wogen 'weighed' (80/2/32/18), ritten 'rode' (128/9/44/26), stritten 'quarreled' (144/16/42/30), rochen 'smelt' (135/6/79/73), schwiegen 'remained silent' (360/18/155/133), litten 'suffered' (254/23/122/78), trieben 'drifted' (631/39/254/160), flossen 'flowed' (245/30/81/60), hoben 'lifted' (1256/31/754/409), stiegen 'climbed' (1580/145/938/558), zogen 'pulled' (1693/5/1192/891) A5. Verbs used in Experiment 5 befehlen 'to order', bergen 'to recover', bersten 'to crack', brechen 'to break', empfangen 'to receive', empfehlen 'to recommend', essen 'to eat', fangen 'to catch', fressen 'to feed', geben 'to 57 give', graben 'to dig', halten 'to hold', helfen 'to help', laden 'to load', lassen 'to let', laufen 'to run', lesen 'to read', messen 'to measure', nehmen 'to take', raten 'to guess', schlafen 'to sleep', schlagen 'to hit', sehen 'to see', sprechen 'to speak', stechen 'to sting', stehlen 'to steal', sterben 'to die', stossen 'to push', treffen 'to meet', treten 'to step', werben 'to recruit', werfen 'to throw'. 58 REFERENCES Anderson, Stephen R. 1982. Where's morphology. Linguistic Inquiry 13. 571-612. Anderson, Stephen R. 1992. A-morphous morphology. Cambridge: Cambridge University Press. Archangeli, Diana. 1988. Aspects of underspecification theory. Phonology 5. 183-207. Aronoff, Mark. 1976. Word formation in Generative Grammar. Cambridge, MA: MIT Press. Aronoff, Mark. 1994. Morphology by itself. Cambridge, MA: MIT Press. Baayen, Harald; Richard Piepenbrock; and H. van Rijn. 1993. The CELEX lexical database (CD ROM). Philadelphia, PE: Linguistic Data Consortium, University of Pennsylvania. Balota, David A. 1994. Visual word recognition: The journey from features to meaning. Handbook of psycholinguistics, ed. by Morton Ann Gernsbacher, 303-349. San Diego, CA: Academic Press. Baumann, Herbert; John Nagengast; and G. Klaas. 1993. New experimental setup (NESU). Nijmegen: MPI für Psycholinguistik, ms. Bierwisch, Manfred. 1967. Syntactic features in morphology: General problems of so-called pronominal inflection in German. To honor Roman Jakobson, 239-270. Den Haag: Mouton. Blevins, James P. 1995. Syncretism and paradigmatic opposition. Linguistics and Philosophy 18. 113-152. Blevins, James P. 2000. Markedness and blocking in German declensional paradigms. Lexicon in focus, ed. by Barbara Stiebels and Dieter Wunderlich, 83-103. Berlin: Akademie-Verlag. Butterworth, Brian. 1983. Lexical representation. Language production, ed. by Brian Butterworth, 257-294. Vol. 2. London: Academic Press. Bybee, Joan L. 1985. Morphology: A Study of the relation between meaning and form. Amsterdam: Benjamins. Bybee, Joan L. 1991. Natural morphology: The organization of paradigms and language acquisition. Crosscurrents in second language acquisition and linguistic theories, ed. by Thom Huebner and Charles A. Ferguson, 67-92. Amsterdam: Benjamins. Bybee, Joan L. 1995a. Regular morphology and the lexicon. Language and Cognitive Processes 10. 425-455. Bybee, Joan L. 1995b. Diachronic and typological properties of morphology and their implications for representation. Morphological aspects of language processing, ed. by Laurie B. Feldman, 225-246. Hillsdale, NJ: Erlbaum. Cahill, Lynne, and Gerald Gazdar. 1997. The inflectional phonology of German adjectives, determiners and pronouns. Linguistics 35. 211-245. Carstairs, Andrew. 1987. Allomorphy in inflexion. London: Croom Helm. Chomsky, Noam. 1970. Remarks on nominalization. Readings in English transformational grammar, ed. by Roderick A. Jacobs and Peter S. Rosenbaum, 184-221. Waltham, MA: Ginn. Clahsen, Harald. 1999. Lexical entries and rules of language: a multi-disciplinary study of German inflection. Behavioral and Brain Sciences 22. 991-1060. Clahsen, Harald; Sonja Eisenbeiss; and Ingrid Sonnenstuhl. 1997. Morphological structure and the processing of inflected words. Theoretical Linguistics 23. 201-249. Corbett, Greville G., and Norman M. Fraser. 1993. Network Morphology: A DATR account of Russian nominal inflection. Journal of Linguistics 29. 113-142. 59 Elman, Jeffrey L.; Elizabeth A. Bates; Mark H. Johnson; Annette Karmiloff-Smith; Domenico Parisi; and Kim Plunkett. 1996. Rethinking innateness: A connectionist perspective on development. Cambridge, MA: MIT Press. Evans, Roger, and Gerald Gazdar. 1996. DATR: A language for lexical knowledge representation. Computational Linguistics 22. 167-216. Feldman, Laurie B., and Carol A. Fowler. 1987. The inflected noun system in SerboCroatian: Lexical representation of morphological structure. Memory and Cognition 15. 112. Fowler, Carol A.; Shirley E. Napps; and Laurie B. Feldman. 1985. Relations among regular and irregular morphologically related words in the lexicon as revealed by repetition priming. Memory and Cognition 13. 241-255. Frauenfelder, Uli H., and Robert Schreuder. 1992. Constraining psycholinguistic models of morphological processing and representation: The role of productivity. Yearbook of Morphology 1991, ed. by Geert Booij and Jaap van Marle, 165-183. Dordrecht: Kluwer. Günther, Hartmut. 1988. Oblique word forms in visual word recognition. Linguistics 26. 583600. Halle, Morris, and Alec P. Marantz. 1993. Distributed Morphology and the pieces of inflection. The view from building 20, ed. by Kenneth L. Hale and Samuel J. Keyser, 111176. Cambridge, MA: MIT Press. Jackendoff, Ray. 1975. Morphological and semantic regularities in the lexicon. Language 51. 639-671. Jackendoff, Ray. 1997. The architecture of the language faculty. Cambridge, MA: MIT Press. Kempley, S. T., and John Morton. 1982. The effects of priming with regularly and irregularly related words in auditory word recognition. British Journal of Psychology 73. 441-454. Kiparsky, Paul. 1982. From cyclic phonology to lexical phonology. The structure of phonological representations, ed. by Harry van der Hulst and Norval Smith, 131-175. Part 1. Dordrecht: Foris. Kiparsky, Paul. 1998. Paradigm effects and opacity. Stanford, CA: Stanford University, ms. Köpcke, Klaus-Michael. 1993. Schemata bei der Pluralbildung des Deutschen. Tübingen: Narr. Köpcke, Klaus-Michael. 1998. The acquisition of plural marking in English and German revisited: schemata versus rules. Journal of Child Language 25. 293-319. Lahiri, Aditi, and William Marslen-Wilson. 1991. The mental representation of lexical form: A phonological approach to the recognition lexicon. Cognition 38. 245-294. Laudanna, Alessandro, and Cristina Burani. 1985. Address mechanisms to decomposed lexical entries. Linguistics 23. 775-792. Laudanna, Alessandro and Cristina Burani. 1995. Distributional properties of derivational affixes: Implications for processing. Morphological aspects of language processing, ed. by Laurie B. Feldman, 345-364. Hillsdale, NJ: Erlbaum. Lieber, Rochelle. 1992. Deconstructing morphology: Word formation in syntactic theory. Chicago, IL: University of Chicago Press. Lukács, Ágnes, and Csaba Pléh. 1999. Hungarian cross-modal priming and treatment of nonsense words supports the dual-process hypothesis. [Commentary on H. Clahsen 'Lexical entries and rules of language: A multidisciplinary study of German inflection.'] Behavioral and Brain Sciences 2. 1030-1031. Lukatela, Georgije; Z. Mandic; B. Gligorijevic; Aleksandar Kostic; Milan Savic; and Michael T. Turvey. 1978. Lexical decision for inflected nouns. Language and Speech 21. 166-173. 60 Lukatela, Georgije; B. Gligorijevic; Aleksandar Kostic; and Michael T. Turvey. 1980. Representation of inflected nouns in the internal lexicon. Memory and Cognition 8. 415423. Lukatela, Georgije; Claudia Carello; and Michael T. Turvey. 1987. Lexical representation of regular and irregular inflected nouns. Language and Cognitive Processes 2. 1-17. MacWhinney, Brian, and Jared Leinbach. 1991. Implementations are not conceptualizations: Revising the verb learning model. Cognition 40. 121-157. MacWhinney, Brian; Jared Leinbach; Roman Taraban; and Janet McDonald. 1989. Language learning: cues or rules? Journal of Memory and Language 28. 255-277. Manelis, Leon, and David A. Tharp. 1977. The processing of affixed words. Memory and Cognition 5. 690-695. Marslen-Wilson, William; Mary Hare; and Lianne Older. 1993. Inflectional morphology and phonological regularity in the English mental lexicon. Cognitive Science Society 15. Marslen-Wilson, William; Lorraine K. Tyler; Rachelle Waksler; and Lianne Older. 1994. Morphology and meaning in the English mental lexicon. Psychological Review 101. 3-33. Münte, Thomas F; Tessa Say; Harald Clahsen; Kolja Schiltz; and Marta Kutas. 1999. Decomposition of morphologically complex words in English: Evidence from eventrelated brain potentials. Cognitive Brain Research 7. 241-253. Napps, Shirley E. 1989. Morphemic relationships in the lexicon: are they distinct from semantic and formal relationships? Memory and Cognition 17. 729-739. Orsolini, Margherita, and William Marslen-Wilson. 1997. Universals in morphological representation: Evidence from Italian. Language and Cognitive Processes 12. 1-47. Pinker, Steven. 1984. Language learnability and language development. Cambridge, MA: Harvard University Press. Pinker, Steven. 1999. Words and rules: The ingredients of language. New York, NY: Basic Books. Pinker, Steven, and Alan S. Prince. 1991. Regular and irregular morphology and the psychological status of rules of grammar. Berkeley Linguistics Society 17. 230-251. Plunkett, Kim, and Virginia Marchman. 1993. From rote learning to system building. Cognition 38. 43-102. Rueckl, Jay G.; Michelle Mikolinski; Michal Raveh; Caroline S. Miner; and Frank Mars. 1997. Morphological priming, fragment completion, and connectionist networks. Journal of Memory and Language 36. 382-405. Rumelhart, David E., and James L. McClelland. 1986. On learning the past tenses of English verbs. Parallel Distributed Processing, ed. by James L. McClelland, David E. Rumelhart and the PDP Research Group, 216-271. Vol. 2. Cambridge, MA: MIT Press. Schreuder, Robert, and Harald Baayen. 1995. Modeling morphological processing. Morphological aspects of language processing, ed. by Laurie B. Feldman, 131-154. Hillsdale, NJ: Erlbaum. Schriefers, Herbert; Angela D. Friederici; and Patty A. M. Graetz. 1992. Inflectional and derivational morphology in the mental lexicon: Symmetries and asymmetries in repetition priming. The Quarterly Journal of Experimental Psychology 44A. 373-390. Sonnenstuhl, Ingrid; Sonja Eisenbeiss; and Harald Clahsen. 1999. Morphological priming and the mental lexicon: evidence from German. Cognition 72. 203-236. Stanners, Robert F.; James J. Neiser; William P. Hernon; and Roger Hall. 1979. Memory representation for morphologically related words. Journal of Verbal Learning and Verbal Behavior 18. 399-412. Stump, Gregory T. 1993. On rules of referral. Language 69. 449-479. 61 Stump, Gregory T. To appear. Inflectional morphology: A theory of paradigm structure. Cambridge: Cambridge University Press. Taraban, Roman; Janet McDonald; and Brian MacWhinney. 1989. Category learning in a connectionist model: Learning to decline the German definite article. Linguistic Categorization, ed. by Roberta L. Corrigan, Fred Eckman and Michael Noonan, 163-193. Amsterdam: Benjamins. Tenpenny, Patricia L. 1995. Abstractionist versus episodic theories of repetition priming and word identification. Psychonomic Bulletin and Review 2. 339-363. Winer, Ben J. 1971. Statistical principles in experimental design. Tokyo: McGraw-Hill Kogakusha. Wunderlich, Dieter. 1987. Vermeide Pronomen - vermeide leere Kategorien. Studium Linguistik 21. 36-44. Wunderlich, Dieter. 1996. Minimalist Morphology: the role of paradigms. Yearbook of Morphology 1995, ed. by Geert Booij and Jaap van Marle, 93-114. Dordrecht: Kluwer. Wunderlich, Dieter. 1997. Der unterspezifizierte Artikel. Sprache im Fokus, ed. by Christa Dürscheid, Karl-Heinz Ramers and Monika Schwarz, 47-55. Tübingen: Niemeyer. Wunderlich, Dieter, and Ray Fabri. 1995. Minimalist Morphology: An approach to inflection. Zeitschrift für Sprachwissenschaft 14. 236-294. Zwicky, Arnold M. 1985. How to describe inflection. Berkeley Linguistics Society 11. 372386. Zwicky, Arnold M. 1986. The general case: Basic form versus default form. Berkeley Linguistics Society 12. 305-315. View publication stats

Log In

The Mental Representation of Infiected Words: An Experimental Study of Adjectives and Verbs in German

The Mental Representation of Infiected Words: An Experimental Study of Adjectives and Verbs in German

Related Papers

RELATED PAPERS