Extractions from relative clauses, a type of long-distance filler-gap dependency (FGD), typically yield unacceptable sentences across the majority of languages. Noun phrases involving relative clauses are therefore assumed to universally comprise syntactic “islands” for extraction (Ross 1967). The fact that extractions from relative clauses (RCEs) (1) are judged as acceptable in Swedish (Erteschik-Shir 1973, Engdahl & Ejerhed 1982) is thus unexpected. None of the theoretical accounts have proven satisfactory. Furthermore, no online, processing-based studies have investigated these structures.

Our study uses an eyetracking while reading paradigm to determine (i) whether RCEs elicit similar processing costs as extractions from non-restrictive relative clauses (nRCE) (2), which show island-like behavior in Swedish (Engdahl, 1997), or if they pattern closer to nonproblematic FGD sentences in which an extraction has been made from a that-clause (TCE) (3), and (ii) whether non-structural factors (frequency, pragmatic fit, and working memory) contribute to any pattern of effects, as facilitatory effects would serve as a positive heuristic for the non-island status of Swedish RCEs (Sprouse, Wagers, & Phillips, 2012; Traxler & Pickering 1996). An intransitive control condition involving pseudo-coordination within a relative clause (4) was also included.

(1) Såna där gamla skottkärror såg jag en man som alltid tvättade på macken när...

such there old wheelbarrows saw I a man that always washed at gas-station-the when...

(2) Såna där gamla skottkärror såg jag en man som förresten tvättade på macken när …

such there old wheelbarrows saw I a man that by-the-way washed at gas-station-the when…

(3) Såna där gamla skottkärror såg jag att en man alltid tvättade på macken när...

such there old wheel barrows saw I that a man always washed at gas-station-the when...

(4) Såna där gamla skottkärror såg jag en man som alltid stod och tvättade på macken…

such there old wheelbarrows saw I a man that always stood and washed at gas-station-the...

For our experiment, we tested 80 critical items, each with four structural variants (1-4) rotated across four presentation lists. Sixty fillers were also included. Forty-five participants each read one list and then completed two working memory span (WM) tests. We used linear mixed models to analyze first fixation durations, gaze durations, regression path durations, and total dwell times for two regions (embedded verb: tvättade; PP: på macken). To assess the contribution of non-linguistic factors, two measures of WM, Ospan (OS) and Dspan (DS); the pragmatic fit (Prag) of the filler to the embedded verb; and the frequency by which the embedded verb is followed by the filler NP were included as predictors. The primary finding at the embedded verb was that RCE and TCE patterned faster than nRCE in both early measures and in regression path durations, with RCE’s facilitation often being enhanced by or conditional upon increases in OS and Prag. This suggests that RCE is processed more similarly to TCE than to nRCE when verb/object integration first occurs, as a function of non-structural factors. Total Durations at the verb region exhibited a three-way distinction, in which RCE patterned between nRCE and TCE (which showed the greatest facilitation), signaling that integrative processes may be somewhat more difficult for RCE than TCE over time, but are still easier for RCE than for nRCE. At the PP region, both RCE and TCE patterned together faster than nRCE, though for RCE this facilitation was again often dependent on non-structural factors, emerging (for Total Durations) and becoming stronger (Gaze Durations) only at higher values of Prag and Ospan.

Our findings provide evidence that RCEs are easier to process than nRCEs, and that this facilitation is dependent in part on high values of certain non-structural factors, such as working memory span and the pragmatic fit of the filler to the selecting verb. Our study thus provides novel processing evidence that Swedish RCEs are more appropriately categorized as long-distance FGDs and not as syntactic islands.