RUNNING HEAD: AVOIDING ATTACHMENT AMBIGUITIES Avoiding Attachment Ambiguities: the Role of Constituent Ordering

Three experiments investigated whether speakers use constituent ordering as a mechanism for avoiding ambiguities. In utterances like "Jane showed the letter to Mary to her mother", alternate orders would avoid the temporary PP-attachment ambiguity ("Jane showed her mother the letter to Mary", or "Jane showed to her mother the letter to Mary"). A preference judgment experiment confirmed that comprehenders prefer the latter orders for dative utterances when the former order would have contained an ambiguity. Nevertheless, speakers in two online production experiments showed no evidence of an ambiguity avoidance strategy. In fact, they were slightly more likely to use the former order when it was ambiguous than when it was not.. Speakers’ failure to disambiguate with ordering cannot be explained by the use of other ambiguity mechanisms, like prosody. A prosodic analysis of the responses in Experiment 3 showed that while speakers generally produced prosodic patterns that were consistent with the syntactic structure, these patterns would not strongly disambiguate the PP-attachment ambiguity. We suggest that speakers do not consistently disambiguate local PP-attachment ambiguities of this type, and in particular do not use constituent ordering for this purpose. Instead, constituent ordering is driven by factors like syntactic weight and lexical bias, which may be internal to the production system. Avoiding Attachment Ambiguities 3 To what extent do speakers design the form of their utterances for their addressees? Some choices are clearly dependent on the need for clear communication, for example the choice of language (in a bilingual community, for instance), or the decision to speak loudly if the addressee is far away. But many production decisions may proceed independently of a consideration of either specific or general properties of the listener. A full understanding of the production system will need to explain which processes, if any, are addressee-oriented. A specific instance of this question is whether speakers avoid ambiguous sentence structures. For example, a speaker who plans to say (1) may realize that it contains a local PPattachment ambiguity, or at least that it may be difficult to understand. (1) The judge sent the letter to the president to the members of the congressional subcommittee. Of course all sentences contain ambiguity at numerous levels, for example the beginning of the word “president” is consistent with cohort competitors like “present,” and “The judge sent the letter” could be a reduced relative clause. However, here we only focus on local PP-attachment ambiguities, which have been shown to sometimes cause parsing difficulty (e.g., Altmann & Steedman, 1988; Boland & Bohem-Jernigan, 1998; Frazier, 1978; Rayner, Carlson, & Frazier, 1983; Spivey-Knowlton & Sedivy, 1995), with potentially long-term consequences for comprehension (Christianson, Hollingworth, Halliwell, & F. Ferreira, 2001). If speakers believe local ambiguities like this will cause difficulty for the listener, they may take pains to avoid them. Several potential mechanisms for this process have been proposed. The speaker may rephrase the utterance, for example “There’s a letter to the president, and the judge got it and sent it...”. In spoken language, the prosodic realization of the utterance has also been shown to facilitate the interpretation of PP-attachment ambiguities, whether the Avoiding Attachment Ambiguities 4 speaker intended the prosody to disambiguate (Snedeker & Trueswell, 2003; Snedeker et al., 2000), or not (Warren, Shafer, Speer, & White, 2000). A third, very natural mechanism for avoiding ambiguity is word order, the issue of interest here. For example, if the speaker wanted to convey the message evoked by the sentence in (1), the PP-attachment ambiguity could be avoided with the double-object construction: “The judge sent the members of the congressional subcommittee the letter to the president.” Indeed, writing manuals often cite word order as a desirable technique to avoid ambiguity (e.g. the National Archives and Records Administration’s guide to legal writing, 2003), and some theories have posited that ambiguity avoidance constrains the shape of grammars (Frazier, 1985; Hankamer, 1973). If speakers make ordering choices to avoid ambiguity, it would additionally be specifically relevant to a debate in the literature about whether choices in ordering are made to facilitate comprehension (Hawkins, 1994), or because of constraints on planning and production (Arnold, Wasow, Losongco, & Ginstrom, 2000; Stallings, MacDonald, & O'Seaghdha, 1998; Wasow, 1997). We investigated this issue by asking whether speakers use phrasal ordering to avoid temporary PP-attachment ambiguities. Certain constructions in English allow variation in the order of the postverbal constituents, for example the Dative Alternation (see 2), where one order contains a local ambiguity, but the other orders avoid that ambiguity. We characterize these orders in terms of whether the goal argument comes relatively early in the utterance, since all constructions with the goal before the theme avoid the ambiguity. Avoiding Attachment Ambiguities 5 (2) Examples of the Dative Alternation: a. theme-early (prepositional): Give the letter to Kim to me. b. goal-early (double-object): Give me the letter to Kim. c. goal-early (prepositional shifted): Give to me the letter to Kim. The choice between orders in dative and other constructions has been shown to be influenced by a variety of factors. The most extensively documented determinant of postverbal ordering is the syntactic weight or complexity of the two constituents, such that shorter, “lighter” constituents tend to precede longer, “heavier” ones (see among others, Arnold, et al., 2000; Behagel, 1909/10; Hawkins, 1994; Stallings, et al., 1998; Wasow, 1997; but see Hawkins, 1994, and Yamashita and Chang, 2001, for evidence of the opposite pattern in Japanese). Ordering has also been shown to be influenced by the accessibility of the concepts referred to by each constituent, such that given or accessible arguments tend to precede new or inaccessible ones (Arnold et al., 2000; Bock & Irwin, 1980). Biases associated with particular lexical items have also been shown to play a role (Stallings et al., 1998; Wasow, 1997). However, none of these factors, together or independently, completely determines the order a speaker will choose. Therefore constituent ordering offers a possible mechanism for avoiding PP-attachment ambiguities, if speakers choose to do so. The current study complements a number of recent studies that have debated whether ambiguity avoidance drives the use of other features of language production. For instance, two unrelated studies have shown that untrained speakers spontaneously produce instructions with prosody that disambiguates global PP-attachment ambiguities (Snedeker & Trueswell, 2003; Warren et al., 2000). Snedeker and Trueswell argue that such prosody is produced for the Avoiding Attachment Ambiguities 6 purpose of disambiguation, and show that disambiguating prosody is only produced when the referential context does not disambiguate the meaning. By contrast, Warren et al. argue that instead it is a natural consequence of producing particular syntactic structures, and does not provide evidence of a disambiguation strategy. Similar findings have also been presented for local PP-attachment ambiguities (Kraljic & Brennan, 2003; Warren, 1985; Straub, 1997). Given these findings, in Experiment 3 we investigated whether speakers used prosody to disambiguate, using utterances like (2b) with two prepositional phrases. However, our study differs from all of the above studies, with the exception of Kraljic and Brennan (2003), in that they studied utterances that contained only a single prepositional phrase, for example “Tickle the frog with the flower” (Snedeker & Trueswell, 2003). By contrast, the presence of two prepositional phrases, as we describe below, reduces the availability of prosodic cues to NP-attachment of the ambiguous PP. Ambiguity avoidance has also been investigated in a study of the production of optional words in sentences like The chiropracter observed (that) {you / I} couldn’t stand up straight (V. Ferreira & Dell, 2000). In the absence of an explicit that, the embedded subject you initially could have been interpreted as either the direct object of “observe”, or the subject of a sentential complement. By comparison, “I” is morphologically constrained to be the subject of the following clause. In both cases the local ambiguity was disambiguated by the following context. In a series of experiments, V. Ferreira and Dell found that that-insertion was equal across these two conditions, leading them to conclude that ambiguity avoidance does not affect this choice. Instead, speakers were more likely to produce a that when the following information was less available, suggesting that optional word choice was driven by the needs of the production system. Avoiding Attachment Ambiguities 7 A host of other studies have investigated the extent to which speakers take the addressee’s perspective into account when designing referential expressions. On one hand, it has been argued that reference is established through a collaborative process between speaker and addressee, where the speaker takes the addressee’s attention and state of knowledge into account dynamically (Clark & Wilkes-Gibbs, 1986; Clark & Krych, 2002). Even children as young as 5 have been shown to design referring expressions with respect to “common ground”, or the information shared between them and their addressee (Nadig & Sedivy, 2000). On the other hand, some researchers have argued that common ground is only used at a second, monitoring stage of production, and not during initial utterance formulation (Horton & Keysar, 1996). As the literature on this topic has shown, at issue is not simply whether speakers pay attention to the listener’s needs (see Schober & Brennan, in press, for a review). Rather, a complete model of the production system will need to specify the circumstances under which speakers do or do not adapt to the needs of the addressee, and the precise processes and mechanisms at work. While the current study will not be able to answer the higher-order question, it contributes to the answer by addressing a plausible mechanism for avoiding local ambiguities – constituent ordering. The role (or lack of a role) that constituent ordering plays is suggestive of both how speakers order constituents and choose syntactic constructions, and the extent to which speakers consider ambiguity when planning an utterance. We will additionally provide evidence about the use of prosody in sentences like “The judge sent the letter to the president to the members of the congressional subcommittee.” The logic used in these experiments is to investigate the production of dative sentences that contain an ambiguity in the theme-early (prepositional) order, as in “Give the letter to Kim to me.” These are compared with control sentences that contain no such ambiguity or where the Avoiding Attachment Ambiguities 8 structural ambiguity is unlikely to cause a garden-path, like "Give the letter regarding Kim to me" and "Give the letter about Kim to me." If speakers tend to choose a goal-early order more often when the theme-early order contains an ambiguity than when it does not, it will be interpreted as evidence for an ambiguity avoidance strategy. In all experiments below, the relative size of the two constituents was held constant, so that weight was not a major determinant of constituent ordering (cf. Hawkins, 1994; Wasow, 1997). In one preference judgment experiment and two on-line production experiments, we investigated ordering preferences in dative utterances like those in (2). All three experiments also investigated Heavy-NP-shift constructions, for example “The chef put the jello in the fancy mold in the refrigerator “ or “The chef put in the refrigerator the jello in the fancy mold,” which also contain an ambiguity in the unshifted order that can be avoided with the shifted order. However, the shifted version of this construction is highly dispreferred, and none of our experimental manipulations did much to modulate this dispreference in either the preference judgment task or the spoken production task. Therefore these data will be excluded from further discussion. Our first experiment investigated whether comprehenders prefer goal-early orders when they avoid an ambiguity. In a forced-choice questionnaire, participants were asked to read pairs of sentences such as those in Table 1 (below), and choose the one they preferred. We manipulated “potential ambiguity”, that is, whether an utterance would be ambiguous if it occurred in the theme-early order. Experiments 2 and 3 investigated ordering preferences in the on-line production of the same kind of utterance. Avoiding Attachment Ambiguities 9