Argument structure saturation in a constraint based model

John Drury,1 Philip Resnik,2 Amy Weinberg,2 Silvia Gennari2 & Sachiko Aoshima2
Georgetown University, 2 University of Maryland


Juliano and Tanenhaus (1994) have argued that the apparently rigid garden path attachment of postverbal NPs in cases like (1) is compatible with frequency based accounts, if we assume that the absence of the comma is itself a frequency based cue to treat the following object as a dependent of the preceding verb.  This allows the intransitive bias of the verb to be overridden by the strong bias of the lack of comma for postverbal attachment.

(1) After the boy sneezed the girl looked curious.

Ferreira and McClure (1996) ( henceforth FM) have presented evidence from reciprocal clauses casting doubt on this interpretation.  The interpretation of a reciprocal verb allows either the second conjunct of a conjoined subject, or the direct object, to be taken as the internal argument of a transitive reciprocal verb.

(2) a.  Mary and John kissed.
b.  Mary kissed John.

In the reciprocal case, (3a), subjects had no trouble interpreting the underlined noun as the subject of a following clause.  This is true despite the fact that there is no comma.  Non-reciprocal verbs showed clear garden path effects under the same circumstances (3c).  This yielded a first pass reading time difference dependent on the presence of an intervening comma only for non reciprocal verbs.

(3) a.  When the golfer and the opponent tied THE RULES were changed.
b.  When the golfer tied the opponent the rules were changed.
c. When the golfer and the opponent disputed THE RULES were changed.

(FM) argued that frequency constraints were thus subordinate to argument structure constraints. These results could not be attributed to frequency since the verb classes were matched for transitive bias.  One problem with their data however is that the difference between (3a and c), and many other of FMs stimulus sets could be also be attributed to the unbalanced implausibility of the postverbal noun phrase as an object in these cases.

Our own eye tracking experiments decouples plausibility from argument structure biases.  Our results show the opposite pattern from that observed by FM.  Reciprocal verbs yield significantly longer residual first pass reading times at the capitalized noun phrase than non reciprocals in the no comma condition. The pattern is reversed for 2nd pass and residual reading time measures.  We interpret these results within a constraint based model.  We show however, that this pattern argues for an inherent argument structure saturation bias that cannot be reduced to frequency.



Ferreira, F., & McClure, K. (1997).  Parsing of garden-path sentences with reciprocal verbs.  Language and Cognitive Processes, 12, 273-306.

Juliano, C., & Tanenhaus, M. K. (1994).  A constraint-based lexicalist account of the subject/object attachment preference.  Journal of Psycholinguistic Research, 23, 459-471.