Anticipating upcoming referents on the basis of discourse status

Comprehension studies of scrambling languages find that noncanonical sentences induce more difficulty than canonical sentences, with this difference being attributed to the structural complexity/infrequency of noncanonical forms (e.g.[1]). However, by presenting sentences in isolation, existing studies implicitly ignore that word order variation is largely guided by discourse factors. E.g., in Finnish, an articleless flexible-word-order language with canonical SVO order, word order encodes given/new distinctions: OVS specifically marks the object as old and the subject as new; SVO is more flexible, being used when the subject is old and the object new, and when both are old or, discourse-initially, when both are new.

If the general framework of referential theories (e.g.[2]) is correct, the observed dispreference for noncanonical orders may arise from the additional presuppositions needed to understand them out of context. Indeed, our prior work, involving self-paced reading in Finnish, showed that processing difficulty associated with OVS is greatly mitigated by discourses that establish appropriate given/new distinctions.

Reading studies, however, cannot easily reveal the on-line referential processes that we propose are involved in SVO/OVS comprehension. Given the description above, the Finnish OV... configuration predicts that a postverbal constituent will be new information, whereas SV... order makes no such prediction. In the current study, we examined whether Finnish listeners (N=16) can use this information on-line to predict referents of upcoming constituents.

Methods. Eye-movements were recorded as participants heard stories and viewed pictures. Target trials contained a picture with three characters (e.g., a man and a woman by a tree, and another woman elsewhere in the scene). Prior to hearing the critical sentence, only two of the characters had been introduced (man and woman by tree), leaving the other character discourse-new. Participants then heard either an SVO (man-subject.greeted.woman-object) or OVS (man-object.greeted.woman-subject) sentence. These 'ambiguous-referent' conditions were compared with 'unambiguous-referents' (man-subject.greeted.child-object), where the discourse-new woman was replaced in the picture by a child. In both cases, OV... should encourage anticipatory looks to the discourse-new referent. And, in ambiguous-referent conditions, SVO should prompt little consideration of the new-woman as the referent of the second noun, as compared to OVS.

Results. The predictions were borne out. As compared to SVO, OVS sentences showed anticipatory eye-movements to the discourse-new referent (other woman/child) at the second-noun onset, before participants had enough phonetic information to recognize this word (F=9.25,p<.05). Listeners in the SVO-unambiguous-referent condition didn't look to the new referent until well after the word-onset. SVO-ambiguous-referent showed few looks to the discourse-new referent: listeners assumed 'woman' referred to the previously-mentioned woman. In contrast, OVS-ambiguous-referent showed substantial looks to the discourse-new woman, but also later competition with the discourse-old woman, due to the ambiguous lexical cue. Off-line referential judgments matched these patterns.

Thus, comprehenders use discourse-status, encoded in object-verb order, to predict that the upcoming postverbal subject is a new, previously-unmentioned entity. Just as Altmann & Kamide's results[3] show anticipation based upon semantic restrictions, our findings illustrate that anticipation also arises on the basis of discourse-status information, as encoded by word order, even when this order is structurally complex.

Example (English translation and gloss)

In the park, near a tree, are standing a man and a woman, while some birds are flying overhead...

Then man-subject greets woman-object ... (SVO-ambiguous-referents)

Then man-object greets woman-subject ... (OVS-ambiguous-referents)

Then man-subject greets child-object ... (SVO-unambiguous-referents)

Then man-object greets child-subject .... (OVS-unambiguous-referents)


[1] Hyönä & Hujanen (1997). Quarterly Jrnl. of Exp. Psychology, 50A,pp.841-858.

[2] Altmann & Steedman (1988). Cognition, 30, pp.191-238.

[3] Altmann & Kamide (2000). Cognition, 73, pp.247-264.