Sentence Construction by Stored-Form Retrieval and by Combination: ERP Evidence for Early and Interactive Processing

Introduction: The human brain stores an immense repertoire of meaningful signs and combines them into a virtually unlimited set of well-formed strings. Communicative function is impaired if strings include meaningless, not stored items (“pseudowords”), or if the combinatorial rules are violated. Success or failure of access to stored signs and of application of combinatorial mechanisms are reflected in brain responses, so that event-related potentials, ERPs, can be used for determining the time course and putative interaction of combinatorial and access processes. Methods: We probed spoken mini-constructions (16 short pronounverb German sentences) composed of stored meaningful morphemes and novel meaningless pseudo-morphemes that obey or violate combinatorial syntactic rules. Information about morpho-syntactic and lexico-semantic violations was first available at the same point in time, the “divergence point” (DP) that corresponded to the onset of the last syllables of stimulus phrases and served as time locking point for the ERPs. The 4 groups of 4 constructions each were either 1) morphosyntactically well-formed [ich leide (=suffer), ich zeige (=show), wir schweigen (=keep silent), wir scheiden (=separate)], 2) violated pronoun-verb agreement (*ich schweigen, *ich scheiden, *wir leide, *wir zeige), 3) containing a morphologically correct but unfamiliar and meaningless pseudo-verb (ich schweide*, ich scheige*, wir leigen*, wir zeiden*), or contained a pseudo-verb that also violated morphosyntactic agreement (*ich leigen*, *ich zeiden*, *wir schweide*, *wir scheige*). This design therefore orthogonally varied the Storage [stored word vs. pseudoword] and Combinatorial [well- vs. ill-formed] factors with the critical, phrase-final syllables being constant over conditions. Spoken phrases were presented in 2 blocks of a MMN multifeature paradigm that used the pronoun (ich or wir) as standard stimulus and the 8 (pseudo)verbs as deviants. Data from 23 subjects were analyzed with repeated-measures ANOVAs. Results: The ERPs showed significant [F(1,22)=4.6, p=0.04, ηp2=0.17] neurophysiological differences between stored morphemes as compared with novel pseudomorphemes 100-200ms after the DP. 270ms after the DP and later, concordant violation responses emerged for both infelicitous combinations and pseudo-verb-containing strings, but double violations failed to elicit a violation response, possibly indexing abortive processing of entirely implausible strings. This pattern was manifest in a significant interaction of the Storage and Combination factors [F(1,22)=6.9 , p=0.01, ηp2=0.25]. Conclusions: These results show that combinatorial morpho-syntactic and storage-related lexico-semantic processes occur in parallel and do interact in the brain, already at 270ms. They also confirm that lexico-semantic processing takes place very early, around 100-200ms after information is available. Construction processes operating on stored forms and combinatorial schemas appear as nearsimultaneous and independent initially and interactive slightly later. Supported by the DAAD, the DFG and the Freie Universität Berlin