A non-local attachment preference in the three-NP-site relative clause ambiguity in Japanese

Abstract:
Gibson et al. (1996) observed that English and Spanish speakers have the same biases for attachment of a relative clause (RC) to one of three preceding NP sites: The most local site is preferred, followed by the least local site, with the middle site least preferred. To explain these data and two-site ambiguity data (e.g., Cuetos \& Mitchell, 1988), Gibson et al. proposed a two-factor system in which one factor, Recency, favors local attachment and a second factor, Predicate Proximity, favors high attachment.

This poster presents self-paced reading data investigating a similar Japanese ambiguity. Because Japanese is head-final, the RC precedes the three sites, as in (1---2):
(1) RC / NP ₁ Postposition ₁ / NP ₂ Postposition ₂ / NP ₃ / main-predicate / remainder
(2) a. (low attachment, to NP ₁ )
Eda-ga oreteiru / shigemi-no yoko-no / hito-no ushiro-no / jitensha-wa / kireide. . .
branch-nom broken bush beside person behind bicycle-top beautiful

b. (middle attachment, to NP ₂ )
Paati-de & atta /. . .
party-at & met

Gakkou-made & notta / ...
school-to & rode
"The bicycle behind the person beside the bush that & /has a broken branch/ is beautiful. . . "
/I met at the party/
I rode to school/

Plausibility disambiguated each attachment. To control potential lexical/plausibility differences, nine conditions were created by rotating the location of the head noun through the three attachment sites for each of the three plausibility-biased RCs. For example, in conditions (a), (d) and (g), the attaching RC was 'that has a broken branch", and the head noun "bush" was located in positions NP _1, NP ₂ and NP ₃ respectively. 36 participants read 36 experimental items and 65 fillers, and answered a comprehension question about each item.

Gibson et al.'s theory predicts that NP ₁ attachment should be preferred, then NP ₃ attachment, with NP ₂ attachment hardest. If instead the high attachment factor has to do with anaphoric binding (Hemforth et al., 1997), then it should not apply to Japanese RCs, because they contain no lexical pronoun operator. Thus Hemforth et al.'s proposal predicts that NP ₁ attachment should be preferred, then NP ₂ attachment, with NP ₃ attachment hardest.

Comprehension performance was better in the NP ₁ attachment conditions than in the other two (ps $<$ .01). In reading times, the NP ₃ attachment condition was faster than the other two (ps > .01) in the NP ₂ and NP ₃ regions. There were no other RT differences. Thus the RT evidence indicated that high attachment was easiest, contrary to both the Gibson et al. and Hemforth et al. predictions. An account of this result, related to the head-final nature of Japanese, will be discussed.