[PS-1.7] ERP-evidence from strong adjectival inflection in German supports morphological underspecification

Opitz, A. 1 , Regel, S. 2 , Müller, G. 1 & Friederici, A. D. 2

1 Institute of Linguistics. Universität Leipzig. Leipzig, Germany.
2 Max-Planck-Institute for Human Cognitive and Brain Sciences. Leipzig, Germany.

In our study we investigated the relevance of the concept of underspecified inflectional markers for the processing of language in the human brain. This concept is used in a wide range of frameworks in order to account for syncretism (Bierwisch (1968), Halle & Marantz (1993), Stump (2001)) although relatively little is known about its cognitive status (Clahsen (2006)). In underspecification-based theories a morphological marker can be incorrect for a given context for two reasons: i) its feature set either contains conflicting features, or ii) it is not specific enough (i.e., there is a more specific marker). Our hypothesis was that, if ungrammatical phrases are parsed that contain one of these different types of violations, this difference should be mirrored in divergent brain responses. We used the ERP violation paradigm to approach this hypothesis in the domain of German adjectival declension. For each correct phrase we provided two incorrect versions by manipulating agreement markers on adjectives preceding a noun. According to our hypothesis, both incorrect versions represent different kinds of violations and should therefore influence ERP-components relevant for morpho-syntactic integration differently (i.e., LAN and P600, see Friederici (2003)). Our results strongly support the underspecification-hypothesis. Although alternative explanations for single effects are available (e.g. surface-near phonological repair), they can be ruled out by the overall pattern and a second observation: LAN was strongest for more specific markers even in correct contexts. This is remarkable as it indicates increased processing demands for highly specific markers. LAN-effects may thus be sensitive not only to morpho-syntactic errors but to the degree of processing-effort as well. Our study suggests that underspecification is real, and that it even may qualify as optimal from an optimal design perspective on language (Chomsky (2005)) as it reduces complexity for both: i) the lexicon, and ii) the procedural part.