Can a connectionist model explain the processing of regularly and irregularly inflected words in German as L1 and L2?(Report)

Abstract

The connectionist model is a prevailing model of the structure and functioning of the cognitive system of the processing of morphology. According to this model, the morphology of regularly and irregularly inflected words (e.g., verb participles and noun plurals) is processed in the same cognitive network. A validation of the connectionist model of the processing of morphology in German as L2 has yet to be achieved. To investigate L2-specific aspects, we compared a group of L1 speakers of German with speakers of German as L2. L2 and L1 speakers of German were assigned to their respective group by their reaction times in picture naming prior to the central task. The reaction times in the lexical decision task of verb participles and noun plurals were largely consistent with the assumption of the connectionist model. Interestingly, speakers of German as L2 showed a specific advantage for irregular compared with regular verb participles.

Keywords

connectionist models, German as L2, lexical decision task, morphological regularity

I Introduction

The cognitive processes involved in the comprehension and production of regular and irregular morphological word inflections have been debated intensively in terms of native speakers (e.g., Clahsen, 1999; Marcus, Brinkmann, Clahsen, Wiese, & Pinker, 1995; Rumelhart & McClelland, 1986) and second language (L2) learners (e.g., Grainger & Dijkstra, 1992; Ullman, 2004). Regular inflections of the English verb participles, for instance, include the past tense ending -ed (e.g., change--changed) and irregular inflections include unique past tense changes (e.g., spend--spent, go--went). One prevailing model explaining the nature of the cognitive mechanisms underlying the processing of regularly and irregularly inflected words is the connectionist model (e.g., Joanisse & Seidenberg, 1999; MeClelland & Patterson, 2002; Rumelhart & McClelland, 1986, for an alternative view see e.g., Pinker, 1999; Pinker & Ullman, 2002). This model postulates a network of phonological, orthographic and semantic representations of both word forms in patterns of interconnected simple processing units. In the connectionist model, it is important to stress that regular and irregular words are processed within a single network while alternative models assume different processing networks for these representations (e.g., dual-mechanism model; Pinker, 1999, see the Discussion section later in this article). Thus, the underlying cognitive architecture to process regular and irregular word forms is identical. In general, the connectionist model is part of a broader view in which cognitive processes are seen as graded, probabilistic, interactive, context-sensitive and domain-general (McClelland & Patterson, 2002).

Empirical evidence for the connectionist model of the processing of morphology was provided by priming studies including native speakers of German (e.g., Smolka, Rosler, & Wiese, 2003; Smolka, Zwitserlood, & Rosler, 2007), Italian (Orsolini & Marslen-Wilson, 1997) and French (Meunier & Marslen-Wilson, 2004), language acquisition during childhood (Brown, 1973; Maratsos, 2000; McClelland & Patterson, 2002), data from brain-damaged individuals (Faroqi-Shah, 2007), ERP-studies (e.g., Justus, Larsen, de Morney Davies, & Swick, 2008), and studies investigating the influence of semantic associations on the processing of morphology (Ramscar, 2002).

However, evidence that supports the connectionist model involved in the processing of regularly and irregularly inflected words in German as L2 is still lacking. German is of particular interest here because the regular forms of verb participles (e.g., Meier, 1964) and more strongly of noun plurals (e.g., Pfeffer, 1964) occur less frequently than irregular forms (see also Marcus et al., 1995). In general, we aim to achieve a better understanding of the processing of morphemes by investigating German as L2 and as L1 in our study. Therefore, the aim of the present study is to look for evidence for a single connectionist mechanism underlying the processing of two types of regular and irregular word inflections, verb participles and noun plurals, in German, in a group of adult L2 learners when compared to German native speakers.

Recently, Smolka et al. (2007) provided evidence for the connectionist model to explain the processing of regular and irregular inflections of verb participles in German native speakers. The authors measured the amount of facilitation that regular and irregular verb participles exerted on the recognition of the related infinitive verb form in a priming paradigm. Importantly, regularly and irregularly inflected participles were controlled for confounding variables, carefully matched for lemma frequency, number of letters and neighbours (i.e., the number of words that differ by only one letter with respect to the infinitive verb form). As regular and irregular verb processing underlie the identical single network according to the connectionist model, equal priming effects were predicted for both types of verb participles. Consistent with this prediction, Smolka et al. showed that regularly inflected participles primed their infinitive verb forms to the same degree as irregularly inflected participles.

In a neuropsychological study using event-related potentials (ERP), Justus et al. (2008) investigated the time course and scalp distribution of the processing of infinitive verb forms primed by their related participles in native speakers of English. Primes for infinitive verb forms were the related regularly inflected participles (which are generated with a regular -ed ending, e.g., pack-packed), weak irregularly inflected participles (which are similar to regularly inflected participles without stem changes, e.g., spend--spent) and strong irregularly inflected participles (which are truly irregular participles including stem changes, e.g., speak--spoken). Results showed a reduced N400 ERP component in trials with priming of strong irregular compared to priming of regular and weak irregular participles. Thus, the findings provide no evidence for a basic difference between regular and irregular participles. The differences between regularly and irregularly inflected words are rather gradual, a basic assumption of the single network system of regular and irregular inflections in the connectionist model.

I. I Formation of verb participles and noun plurals in German

In this section, we briefly summarize the formation of verb participles and noun plurals in German to enable a better evaluation of the effects to be reported in this study. Regular inflection of verb participles involves the affix ge- and the suffix -t without stem changes (e.g., trennen ['to separate', infinitive form]--trennte [past tense]--getrennt [past participle]). Word forms of irregular past participles are also generated with the affix ge- bur, instead of the suffix -t, they have the suffix -en (e.g., hauen ['to hit', infinitive form]--haute [past tense]--gehauen [past participle]). Additionally, irregular participle forms might undergo unpredictable phonological stem changes for a learner of German (e.g., gehen ['to walk', infinitive form]--ging [past tense]--gegangen [past participle]; schreiben ['to write', infinitive form]--schrieb [past tense]--geschrieben [past participle]). Depending on the size of text samples used to estimate verb frequency in German, the type frequency of strong irregularly inflected verbs ranged from 16% to 68%, the proportion of weak irregularly inflected verbs ranged from 1% to 31%, and the proportion of regularly inflected verbs ranged from 23% to 78% (Marcus et al., 1995).

Plural formation of nouns in German consists of five categories: four categories have the plural suffixes -e (e.g., Stein, 'stone'--Steine, 'stones'), -er (e.g., Gesicht, 'face'--Gesichter, 'faces'), -[e] n (e.g., Frau, 'woman'--Frauen, 'women') as well as -s (e.g., Hotel, 'hotel'--Hotels, 'hotels'), and one category is unmarked (e.g., Schuler, 'student'--Schuler, 'students'). Some plural forms in the categories with -e and -er plural suffixes co-occur with stem changes (e.g., Blatt, 'leaf'--Blatter, 'leaves'). Forms generated with -s endings act as the regular default, and the remaining categories were seen as the irregular forms of plural generation (e.g., Clahsen, 1999; Marcus et al., 1995; but see also McClelland & Patterson, 2002). Depending on the type of analysis, the type frequency of nouns taking regular -s endings ranged from 0% to 9% while the remaining nouns take irregular endings (Marcus et al., 1995).

To our knowledge, there are only a few studies focusing on the cognitive processing system of regulars and irregulars in L2 learners of German. In detail, Hahne, Mueller, & Clahsen (2006) looked at the generalization properties of regular and irregular participles (i.e., -t ending vs. -n ending) as well as of regular and irregular noun plurals (i.e., -s ending vs. -n endings) in learners with Russian as L1. That is, the authors analysed the amount of generalization of regular and irregular inflections during the production of nonce words (see also Clahsen, 1997). For participles, Hahne et al. found higher generalization rates in regular participles than in irregular participles. Investigating generalization rates of nouns, no differences in the regular and irregular plurals were found. While the data of plural generalization rather supports a single-mechanism hypothesis, the data of participle generalization showed no support for this view. Thus, evidence from the generalization data supporting a single processing mechanism within the framework of the connectionist model can be considered inconclusive at best. Furthermore, several critical issues may be raised with regard to the Hahne et al. study. First, there is no direct comparison with a control group of native speakers of German that allows an unequivocal interpretation of the results as being due to the German language or being specific to German as L2. Second, the instructed generalization task required very complex processing including multiple processing stages (e.g., for generalization of participles: (1) presentation of a simple past form, (2) repetition of a simple past form and (3) generating the participle form). This method only allows the assessment of frequency data but no processing of time data, which might be more sensitive to the effects of specific cognitive processes. Finally, the study of Hahne et al. leaves open the question of how L2 learners process well-formed irregular and regular word forms.

In regard of these issues, Neubauer and Clahsen (2009) applied the alternative lexical decision task (LDT) to assess processing of regularly and irregularly inflected participles in L2 learners of German (with Polish …