One or two derivations in (bimodal) bilinguals: that’s the question Josep Quer ICREA-Universitat Pompeu Fabra The study of bimodal bilingualism constitutes a rather new and fascinating field not only to better understand the nature of linguistic representations in bilinguals in general, but also their interface with production and comprehension systems. The amount of data and linguistic analyses of bimodal bilingual subjects is still quite limited, and Lillo-Martin, Müller de Quadros & Chen Pichler (2016) summarize one of the main threads of research in this domain, associated to the BiBiBi project developed on a database with ASL- English and Libras-Portuguese bilingual data. The other main thread is represented by Donati & Branchini (2013) and subsequent work, which relies on a corpus of LIS-Italian bilinguals. Some of the empirical generalizations drawn by the two groups of researchers are shared, but there are also some important differences, which have led to significantly differing models for the derivation and representation of (bimodal) bilingual utterances. One striking difference mentioned in Lillo-Martin et al. 2016 (section 5) is the amount of non-congruent syntax in blended structures: in their database of bimodal bilingual productions such structures are very rare. By contrast, in Donati & Branchini’s work on LIS-Italian bilinguals, even if they are not as frequent as congruent blended sentences, they are not at all residual, but rather quite common. In addition, they have been submitted to comprehension tasks by bimodal bilingual subjects and they have been deemed as totally acceptable and natural. As pointed out in Lillo-Martin et al. (2016), this discrepancy is most probably the consequence of divergent basic word orders in the bilingual pairing: SVO for Italian and SOV for LIS. This is arguably a major contrast to take into account when theorizing about the grammatical competence of bimodal bilinguals, since the conclusions drawn in Lillo-Martin et al. 2016 might carry the empirical bias of having examined only congruent language pairs. When referring to the set of facts of non-congruent blends, the authors state that “the late linearization analysis is completely compatible with the Language Synthesis Model” (p. 42, 1 st paragraph). For non-congruent syntax in blended sentences Donati & Branchini’s (2013: 120) proposal (from which they have departed in later work) amounts to having a single syntactic structure only encoding hierarchy, and two opposing linerarizations at PF “generating the two strings [that] are represented respectively as “>” (c-command = precedence, yielding Italian) and “<” (c-command = succession, yielding LIS)”: (1) WH>C>I>V vs. WH<C<I<V chi ha telefonato CALL DONE WHO ‘Who has called?’ Although no data are discussed in this respect, I think reducing non-congruent blends to linearization might be too simplistic. If linearization is a “harmonic” phenomenon that yields SVO vs. SOV orders in the two languages at hand, it fails to predict that Topic constituents, which hierarchically appear above WH, are robustly part of the left periphery both in Italian and LIS. The linearization account would give the right result for Italian, but arguably not for LIS, since it would linearize the topic to the right of the WH word, which is predicted to be ungrammatical: