(22) Disguised prosodic cues for syntax

Poster session 1
Monday, September 5, 17:30
Yuki Hirose
University of Tokyo/ University of electro-Communications
hiroseatbozdotcdotu-tokyodotacdotjp
Prosody is regarded as a very important cue for syntax. However, non-syntactic factors can also affect it. The study demonstrates that certain prosodic patterns derived by non-structural factors can be exploited for ambiguity resolution.

In a left-vs. right-branching (LB/RB) ambiguity in Japanese as in (1), presence or absence of lexical accent (indicated by " ' ") on the initial adjective, (ao'i (blue) :accented vs. kiiroi (yellow): unaccented), influenced phrasal level prosody, that in turn affected listeners' branching analysis. The role of contrastive focus was tested by manipulating visual information to demonstrate that contrastive prosody can be (mis)perceived as a syntactic marker distinguishing LB from RB.

In the production task, naïve speakers (n=8) gave verbal instructions to their partners, who were separated but looking at the same scene, to identify the target object, using fixed phrase as in (1), where either RB or LB was intended. The visual context was manipulated as follows:

AMBIGUOUS context: Target sentences intended either LB or RB. Objects corresponding to both analyses were included in the scene, e.g.,

CONTRASTIVE context: only one possible target object was present with a color-contrasting counterpart, e.g.,:

For LB-intended sentences

For RB-intended sentences

(Note: case = CD case. No other object was colored blue or yellow)

Finally, all subjects were informed about the syntactic ambiguity prior to the last session (DISAMBIGUATED context). The recorded utterances were subject to the perceptual study and a phonetic analysis. In the perception study, new groups of subjects (n=12) made the forced choice judgment (RB orLB) on a picture.

There was a significant difference between items with an accented adjective (ao'i) (85%) and unaccented adjectives (kiiroi) (68%) in listener's performance (identifying the intended structure) in LB-intended utterances from the DISAMBIGUATED session. This is reasonably explained by the fact that successive downsteps (downstep occurs following an accented item) is a major cue indicating a left-branching structure; such cue is not available with an unaccented adjective.

Listener's performance on items with accented adjectives was significantly higher for LB-intended utterances collected in the CONTRASTIVE context (81%) compared to the AMBIGUOUS (54%). F0 peak declination between the adjective (blue) andN1 (CD) was significantly larger in the CONTRASTIVE context, accounting for the perception data. That is, the contrastive focus on the first item enlarged the F0 fall, exaggerating the typical LB-marking pitch pattern.

Example
(1) ao'i/kiiroi siidi'i-no ke'esu-o sa'sitekudasai
  blue/yellow CD-Gen case-Acc point
  Point the case of the{blue/yellow} CD. (LB-interpretation)
  / Point the {blue/yellow} case of the CD (RB- interpretation)