Active filler effects in Japanese wh-scrambling constructions

Sachiko Aoshima, Colin Phillips & Amy Weinberg
University of Maryland, College Park

aoshima@wam.umd.edu

In this study, we present evidence for the operation of the Active Filler Effect in Japanese.  We show that Japanese speakers preferentially associate a fronted wh-phrase with an embedded clause.  This finding allows us to propose two refinements to the Active Filler Strategy's operation for all languages, specifically involving (i) the definition of 'first available position' and (ii) the parser's ability to rescind existing commitments in order to satisfy other constraints.  The experimental results indicate that the wh-phrase is filled in the linearly, not structurally, nearest position, and that the parser is able to make the necessary reanalyses.

We use Typing Mismatch Effects (TME; Miyamoto &Takahashi, 2001) as a diagnostic of active gap filling.  M&T showed for wh-in-situ sentences in Japanese that readers do not expect to encounter a declarative complementizer on the first verb (b), yielding a slowdown, but no such slowdown when the first verb contains a wh-scope particle (d).  We reason that if a TME slowdown is also observed in sentences with wh-fronting (a) in parallel to in-situ examples (b), the fronted wh-phrase must have been reconstructed to a position inside the embedded clause, closest to the first scope marker.  Thus, we predict a parallelism between (a)-(c) and (b)-(d) in reading time patterns at the embedded verb.  Since the fronted wh-phrase in our materials is a dative-marked NP, which is not subcategorized by the embedded verb (it receives a benefactive interpretation), the placement of the gap in the embedded clause cannot be attributed to lexical requirements of the embedded verb.  Therefore, the most likely reason for the interpretation of the wh-phrase in the embedded clause is that readers create a gap in the linearly first available position.  In order to associate the wh-filler with an embedded clause gap-site, readers must revise their initial matrix question interpretation of the sentence, yielding an indirect question interpretation (i.e., a declarative statement).  This reanalysis is not grammatically forced, and hence is unexpected under certain accounts of reanalysis.

a. Wh-dat NP-top [NP-nom NP-acc V-DeclC] Advb NP-dat V-QP [SCRAM+DECL]

b. NP-top [NP-nom Wh-dat NP-acc V-DeclC] Advb NP-dat V-QP [INSIT+DECL]

c. Wh-dat NP-top [NP-nom NP-acc V-QP] Advb NP-dat V [SCRAM+QP]

d. NP-top [NP-nom Wh-dat NP-acc V-QP] Advb NP-dat V [INSIT+QP]

In a self-paced reading task, subjects (n=21) read the embedded-verb region in (a) 81ms slower than in (c), as predicted.  Reading times in (a) and (c) did not differ elsewhere.  We also observed a parallelism between the differences in (a)-(c) and (b)-(d).  These results indicate that readers ultimately interpret fronted wh-phrases as indirect questions, and hence that the active filler effect operates linearly.  We discuss these results in the context of other Japanese studies, e.g., Fodor & Inoue (1995) and Kamide & Mitchell (1999).

Examples