One of the most important discoveries regarding natural languages is that there are systematic constraints on the dependencies that exist between elements in a sentence. A sub-class of these constraints, referred to as island constraints in the literature, are generally assumed to be universal. Whether such constraints arise from factors specific to language or are influenced by general processing factors has been the subject of a lively international and interdisciplinary debate. Surprisingly, Swedish and the other Mainland Scandinavian do not consistently exhibit all of the expected effects of such constraints on dependencies, allowing e.g. relative clause extractions. Yet, despite the importance of this potential exceptionality, no current analysis has successfully managed to account for the nature of the deviation. Consequently, Mainland Scandinavian data has largely been omitted in the formulation and testing of island constraints as they relate to both language and general processing.

The main aim of our project has been to investigate to what extent Swedish comprises an exception to island constraints banning relative clause extraction. In addition, we investigated to what degree processing factors (e.g., working memory span) contribute to the processing and acceptability of the structures in question.

In a series of eyetracking while reading experiments, Tutunjian, Heinat, Klingvall, and Wiklund (2017) showed that Swedish sentences involving long-distance dependencies (LDDs) into relative clauses (RCs) in object position pattern between licit (non-island) and illicit (island) structures in terms of integration of the dependency at the RC verb. In addition, sentences involving LDDs into RCs in subject position were shown to display plausibility effects at the RC verb. Taken together, these findings demonstrate that, contrary to other claims in the literature, integrative processes are at work to some degree in such structures. We argue that these processes arise on account of weak island extraction and parasitic gapping and thus do not comprise an exception to the universality of island constraints across languages. In addition, eyetracking and working memory task (automated o-span and reverse digit span) results from Tutunjian, Heinat, Klingvall, and Wiklund (2015a) suggest that facilitation of these structures may be in part dependent on non-structural factors (working memory span and pragmatic fit).

In a series of acceptability judgment and online, word-by-word reading experiments, Müller (2019) and Müller, Tutunjian, and Wiklund (2019a) further investigated to what degree integrative processes are at work in strong islands by expanding our domain of investigation to Swedish and English adjunct islands. These studies provide the first experimental support for the claim that English sentences involving adjunct island extractions are sensitive to the finiteness of the adjunct clause and the degree of coherence between the main clause and the adjunct clause. The acceptability judgment studies also provide the first experimental support for cross-linguistic differences between Swedish and English in regard to finiteness effects in adjunct clauses. Whereas both English and Swedish displayed coherence effects, only English displayed finiteness effects.

As a further investigation of factors contributing to the integrative processes observed in Tutunjian et al. (2017), Tutunjian and Wiklund (2019) used a filled-gap, eyetracking while reading paradigm to investigate whether matrix verb transitivity affects the likelihood of the parser incrementally engaging in parasitic gapping (PG) in Swedish subject islands (involving a relative clause). It was found via patterns at the embedded verb, that prior cues such as transitivity affect the likelihood of the parser postulating and establishing PG in Swedish subject islands.

The findings from our study highlight the importance and need for a reconceptualization of the operationalization of syntactic islands both from a syntactic and processing point of view. We have identified a number of points for future consideration. First, from a syntactic point of view, islandhood has traditionally been measured via acceptability ratings. However, our study demonstrates that in formal, experimental settings, even licit extractions may receive low ratings despite having an apparent, intuitive acceptability. From a processing point of view, illicit extractions clearly show different gradations of processability which are not necessarily visible in acceptability judgments, suggesting that they are not impermeable to integration. In the absence of such impermeability, one must then question what it means to be a "strong island". Our study also shows that certain factors influence both the acceptability and processing of certain island structures (e.g., coherence, finiteness, and working memory) and that languages can vary in regard to their sensitivity to these factors. Of interest is the question of what other factors may contribute to so-called island effects and what the limits of variation are across languages and factors. Furthermore, it still remains an open question as to whether these factors contribute to the pattern of results independently of island-constraints, or if island constraints actually arise on account of the presence of such factors when present in some combination. Finally, there also remains a need to identify how real time processability and acceptability relate to each other and whether gradable differences in facilitatory patterns from processing data can consistently be mapped onto similar patterns of effects in acceptability judgments.

Our results have been presented at prestigious international conferences as well as at national workshops and linguistic seminars. We also organized a workshop on island constraints in the Scandinavian languages where we had the opportunity to address and discuss various issues relating to the theme of the workshop with colleagues from Norway and Denmark.