This poster presents data collected as part of the first known investigation of sentence processing in an indigenous North American language. It argues that any complete model of human sentence parsing should include heuristic-based mechanisms that take over when faced with costly syntactic reanalysis. A self-paced listening experiment was performed in Odawa (Ottawa). Odawa has extremely flexible word order without overt case marking or prosodic marking of non-canonical orders, and a rich verbal inventory. A subject-verb-object sentence can occur in any of the six logical word orders. Odawa's "direct" (dir.) verb form shows agreement with subject/agent, the "inverse" form (inv.) with object/patient.
Distance between theta role assigners and assignees were varied by crossing word order and verb form with 0-2 adverbs, which were situated immediately after the first constituent, creating distance between thematic assigners and assignees. Certain theories predict distance to increase processing load as unassigned thematic roles accrue, others predict distance will strengthen ultimately correct parsing predictions.
The experiment yielded both on-line and off-line (picture verification task) results partially consistent with models predicting increased processing costs as unassigned thematic roles accrue. Adding one adverb to all word orders reduced accuracy levels in the off-line task. In verb-final orders, adding another adverb further reduced accuracy. However, when two adverbs were added to verb-initial and verb-medial orders, accuracy rose in some orders, approaching performance on sentences without adverbs. On-line data for sentences with no adverbs and with two adverbs also patterned together in non-verb-final orders. The results are explained if several Odawa facts are considered. Despite flexible word order, canonical orders exist. When speakers encounter alternative orders, empty canonical argument positions are assumed to be filled with phonologically null 'pro' elements. When it is discovered that scrambled word orders rather than pro-drop obtains, syntactic revisions must be made. If the revisions are delayed due to intervening adjuncts a linear heuristic is employed, which is successful in certain orders. This heuristic leaves the syntactic structure built with pros in place, and assigns coreferential indexes to the pros in argument positions and the full NPs, which, it is proposed, are alternatively parsed in right-dislocated positions. Verb-final orders were not subject to this heuristic, as no pros are posited until an empty canonical argument position is encountered.
These data suggest that a heuristic-based processor takes over when processing load becomes too great. The syntax of Odawa allows for efficient use of a linear heuristic in certain orders.
Picture verification data (% correct) by word-order, verb form, and # adjuncts
| VSO | VOS | SVO | OVS | SOV | OSV | ||||||||
| Dir. | Inv. | Dir. | Inv. | Dir. | Inv. | Dir. | Inv. | Dir. | Inv. | Dir. | Inv. | ||
| 0 adj. | 84.4 | 58.7 | 80.4 | 82.6 | 95.7 | 69.6 | 76.1 | 69.6 | 78.3 | 71.7 | 54.4 | 87 | |
| 1 adj. | 54.4 | 54.4 | 56.5 | 58.7 | 58.7 | 52.2 | 47.8 | 56.5 | 78.3 | 67.4 | 67.4 | 65.2 | |
| 2 adj. | 69.6 | 73.9 | 58.7 | 47.8 | 76.1 | 80.4 | 63 | 65.2 | 65.2 | 52.2 | 54.4 | 60.9 |