Reconciling the Lexical Nature of Attachment Preferences With the Structural Nature of Reanalysis Using C-command

Abstract:

Though sentences (1) and (2) are both locally ambiguous at the asterisk, only the ambiguity in (2) causes processing difficulty:

Standard explanations for this contrast rests on two pillars: preferences for initial analysis, and a cost metric for reanalysis. Assuming that the initial hypothesis at * in (1) and (2) is that compatible with the first completion, the difference in processing the non-preferred completions must derive from reanalysis cost. The D-theory approach (Marcus, Hindle and Fleck, 1983) characterizes this in structural terms: any analysis derived from an initial hypothesis through monotonic addition of structural relations is easily processed.

In contrast to the use of domination that D-theory crucially exploits, we suggest that c-command is the structural primitive used by the parser. Using c-command, the contrast in (1) and (2) follows if the initial analysis in both cases involves the assertion that the verb c-commands the following NP (the structural correlate of a thematic dependency between verb and object). This relation continues to hold in the dispreferred completion in (1) but not in (2), reflecting the perceived difference. While c-command is less flexible than dominance in characterizing phrase structure (i.e., there are tree structures describable with dominance but not with c-command, cf. Frank and Vijay-Shanker, 1995), it nonetheless allows us to capture the cases which originally motivated this line of work. Moreover, c-command's restrictiveness explains previously problematic cases without resort to additional structural relations (Gorrell 1995) or restrictions on reanalysis (Sturt and Crocker, 1996). A significant body of examples is amenable to our analysis, unifying cases that previously received disparate treatments.

Other work has shown that initial parsing preferences are derived (at least) from thematic and lexical properties of words in the input. We believe however that such properties have their effects via the c-command relations they induce between lexical elements. Therefore, the parser's specification of c-command relations allows us to unify the determination of initial preferences on lexical/thematic grounds with a structural component of determining cost of reanalysis. This linkage between initial preference and reanalysis plays out in interesting ways in cases of head-complement and modification relationships. Additionally, there are intriguing consequences of a grammatically motivated assumption (Kayne 1994) that other things being equal, consecutive words stand in a c-command relation.

Frank, R., Vijay-Shanker, K. (1995). C-command and grammitical primitives. GLOW Newsletter 34:24--25. Paper presented at the 18th GLOW Colloquium.

Gorrell, P. (1995). Syntax and Parsing. Cambridge University Press, Cambridge, UK.

Kayne, R. (1994). The Antisymmetry of Syntax. MIT Press, Cambridge MA.

Marcus, M., Hindle, D., and Fleck, M. (1983). D-Theory: Talking about Talking about Trees. Proceedings of the 21st Annual Meeting of the Association for Computational Linguistics, Cambridge, MA. 129--136.

Sturt, P., and Crocker, M. (1996). Monotonic Syntactic Processing: A Cross-Linguistic Study of Attachment and Reanalysis. Language and Cognitive Processes, 11:5, 449--494.