How much Prosody can open up Syntactic Islands? Evidence from Korean

Lee (1993) reported some island effects in long-distance scrambling in Korean. Based on adequacy of scrambling, she argued that a relative or an adjunct clause, which is not selected by a verb via selection (Cinque 1990), forms a strong island, whereas a whcomplement or a propositional complement clause forms a weak island or doesn’t show any island effect as they can be selected by a verb. Lee (1993) verified the islandhood of each clause type by using off-line questionnaire study. I claim that the type of clause is not the only or the best criterion to explain island effects found in long-distance scrambling in Korean. I claim that prosody plays a crucial role in deciding the strength of islandhood in on-line parsing. This study supports the claim that Korean parsing is incremental just as in English. To examine this, I adopted an auto-paced listening in conjunction with self-paced reading experiment. (n=25 in both cases). In the first experiment, I showed that even default prosody projected in silent reading can open up syntactic island and squeeze scrambled constituent back to its source structure very naturally. The evidence of squeezing can be found by the existence of Typing Mismatch Effect (=TME; significant delay means that parser’s certain expectation is discouraged: Miyamoto and Takahashi 2001) in parsing (1) and (2) via selfpaced reading experiments. TME at region 6 (=Yuna-ekey) in parsing (2) signals the possible competition between two dative NPs (=Jina-ekey and Yuna-ekey). And the reason of their competition is because the parser squeezed fronted dative NP (=Jina-ekey) already to the relative clause even before it knows the type of the current clause. In fact, in parsing (3), similar TME occurred when the second dative NP (=Yuna-ekey) is encountered. The fact that such delay occurs ahead of parsing the verbal suffixes (e.g., -nun for a relative clause, -ko for a complement clause) and TME between (2) and (3) at region 6 prove that the parser’s decision to perceive forthcoming structure as a strong or weak island is made pro-actively as early as possible rather than at the end or when the grammatical head is reached, as proposed in other head-driven approaches. (See Prichett 1992). The second experiment is via auto-paced listening experiment in conjunction with self-paced reading. Intonational break is represented as a blank word in the window with an auto-paced pause. As shown in (4), in a self-paced reading experiment, similar delay is observed in region 3. This is the result of honorification mismatch between the subject candidate kyojangsensengnim ‘head master’ and the non-honored verb mek‘eat’. However, this slow down effect can be cured if there is an intonational break right after the subject as shown at region 3 of Fig 2 (second line). Based on these results, I argue that when prosody is covertly given, the parser tends to build up a structure locally and incrementally. However, when prosodic cues such as intonational break are given overtly, structure-building pattern changes. In particular, intonational break triggers the parser not to associate preceding lexical item next to the following item in the same local structure. My finding strongly supports Fodor’s Implicit Prosody Hypothesis (i.e., even in a silent reading, default prosody needs to be accounted. Fodor 2002). Based on the these experiments, I argue that purely syntactic notion such as islandhood can be best understood when prosodic information.