Grammar Types in Language Explain Tone Sequence Processing in Music Christiane Neuhaus, Thomas R. Knösche, Jörg Bahlmann, Angela D. Friederici Max Planck Institute for Human Cognitive and Brain Sciences, Germany neuhaus@cbs.mpg.de, knoesche@cbs.mpg.de, bahlmann@cbs.mpg.de, angelafr@cbs.mpg.de ABSTRACT In this ERP study, linear and center-embedded musical sequences are built according to two artificial grammar types in language, named finite state grammar (FSG) and phrase structure grammar (PSG). The aim is to prove if neural sources and processing mechanisms for artificial grammar settings across domains are the same. Isochronous pitch sequences were constructed by two interval categories (3 rd and 6 th ) in upward and downward direction. FSG sequences, which have the general form ABAB in artificial grammar, are translated into “small up/small down/large up/large down”. PSG sequences of form A[AB]B are transposed to “small up/large up/large down/small down”. In two ERP recordings testing FSG and PSG separately, non-musicians had to distinguish between correct and false examples after getting familiar with each grammar type. Deviant sequences either include an item of reverse interval or contour. Our main results are: (1) N1 components indicate a 2-item- chunking in FSG and a 4-item-chunking in PSG based on immediate repetition between adjacent tones, thus low-level grouping is different for each grammar type. (2) A late processing negativity at sequence offset indicates syntax-based integration-and- memory processes primarily for PSG. The partially congruent ERP results for artificial grammar learning in language and music confirm that the linguistic perspective on music may be justified. I. INTRODUCTION Since the seminal work of Noam Chomsky, researchers draw parallels between music and language and seek common cognitive principles in both domains. Over the past two decades there has been an ongoing tendency to explain syntax processing in music by concepts in linguistics. Strong evidence has been found that processing mechanisms in both domains are partly the same and that, functio-anatomically, the same brain regions are activated. For the ease of handling this topic, let us concentrate on the aspect of syntax and leave aside the interactions with phonological and semantic effects. Three relevant insights into the processing of musical and language syntax should briefly be mentioned here. First, Maess and colleagues (2001) proved by MEG source localization that syntactic incongruities in music, in particular unexpected chords in musical cadences, are processed in Broca‟s area (BA 44/45) which was previously considered as specific to syntax processing in language (see also Koelsch et al., 2002, for comparable results using fMRI). Second, Levitin and Menon (2003) showed in an fMRI study that tone sequences of temporal coherence as compared to disrupted and random musical structures are mainly processed in the pars orbitalis region (BA 47) of the left inferior frontal cortex which was previously identified as the circumscribed brain region for verb generation and semantic word processing. Third, Patel (1998, 2003) suggests a shared cognitive mechanism for processing musical syntax and language syntax, called “shared structural integration resource” hypothesis. According to this hypothesis, two locally separate components are located in anterior and posterior parts of the brain, the former is assigned to allocating resources for integrating and memorizing lexical items, the latter is assigned to syntax representation and storage. Let us focus on center-embedded (nested) structures in language as compared to music and distinguish between natural and artificial types of grammar. In natural languages, such as English or Italian, center-embedded structures bring about hierarchical relations between subordinate items of the relative clause and the respective anchor words in the main clause which leads to distance-based dependencies in word structure (Gibson, 1998). In music, a substantial portion of structure reveals similar hierarchical relations, most evident in chord progressions with the tonic center as the main reference (cf. Patel, 2003). As several attributes in natural languages, for example animacy and case, play a substantial role in processing grammatical relations, it is, of course, a lot easier to explore the processing of center-embedded structures in artificial grammar learning tasks in which the only modified variables are category and word order. In artificial grammar learning, Friederici and Bahlmann (2006) studied long and short consonant-vowel sequences (e.g. “le ri se de ku bo fo mo”) based on two grammar types called finite state grammar (FSG) and phrase structure grammar (PSG). FSG is determined by local transition probabilities. Sequences follow the rule (AB) n , and the resulting structure is ABAB (e.g. “de bo gi fo”). PSG is characterized by recursive embeddings and accordingly, long-distant dependencies are processed. PSG sequences follow the rule (A n B n ), yielding the structure AABB (e.g. “de gi bo fo”). Thus, the study compares linear with this hierarchical sequence structure, and A and B categories are distinguishable from each other by bright and dark vowels. The present study uses the artificial grammar learning paradigm by Friederici et al. (2006) and translates it to music at a ratio of 1:1. Our objective is to prove if FSG and PSG grammar types in language can describe (pitch) structure in music appropriately, and if neural sources and processing mechanisms are once again similar across domains so evidence towards domain-general processing is accumulating. For modelling this kind of grammar setting in music, we use isochronous pitch patterns, each consisting of eight tones in two interval categories (3 rd and 6 th ) and two pitch directions (up and down). The finite state grammar (FSG), as illustrated by structure ABAB, is translated into the pitch sequence “3 rd up/3 rd down/6 th up/6 th down” whereas condition PSG, as represented by structure AABB, is transposed into “3 rd up/6 th up/6 th down/3 rd down” (see note examples in Figure 2). The Proceedings of the 7th Triennial Conference of European Society for the Cognitive Sciences of Music (ESCOM 2009) Jyväskylä, Finland Jukka Louhivuori, Tuomas Eerola, Suvi Saarikallio, Tommi Himberg, Päivi-Sisko Eerola (Editors) URN:NBN:fi:jyu-2009411302 372