Thematic Roles1
Running head: THEMATIC ROLES LEAVE TRACES
Can thematic roles leave traces of their places?
Franklin Chang, Kathryn Bock, and Adele Goldberg
Beckman Institute, University of Illinois at Urbana-Champaign
Postal Address:
Franklin Chang
Beckman Institute
University of Illinois at Urbana-Champaign
405 N. Mathews Ave.
Urbana, IL, 61801
USA
Email:
Abstract
An important question in the study of language is the nature of the information that speakers use to create syntactic structures. Research using structural priming has suggested that the construction of syntactic frames may be insensitive to variations in thematic roles within messages (Bock & Loebell, 1990; Bock, Loebell, & Morey, 1992). Because these studies involved structural alternations whose syntax covaries with the order of thematic roles, it is difficult to assess any independent contribution that role information makes to the positioning of phrases. In this study, we primed the order of the roles without changing the syntactic structure, and found that the order of the roles was influenced by the priming manipulation. This implies that thematic roles or the features that differentiate them are active within the mapping between messages and sentence structures.
Can thematic roles leave traces of their places?
When we speak, we typically try to convey some notion or set of notions in a sequence of words. Both of these tasks, conveying notions linguistically and developing a sequence of words for doing so, have been argued to depend on their own specialized representations (Garrett, 1988)(Garrett, 1988). The sequencing of words is constrained by syntactic structure, which represents hierarchical and linear relationships among phrases. Such structures can be seen as schemes for guiding the retrieval of words from memory (Bock, 1987).
On their own, structural constraints cannot insure that a sentence conveys the intended meaning, and the violations of sentence meaning seen in word exchanges and other speech errors indicate that in fact, structural integrity is not incompatible with semantic perfidy. Rather, another set of processes must mediate between intended messages and syntax. For example, the relationship between man and dog is different in the sentences The man bit the dog and The dog bit the man. In the first sentence, the dog did the biting and the man was bitten, while in the second sentence the fillers of the roles of the verb bitwere are reversed. Because Moreover, the same sets of roles can occur in different structural and serial positions (compare the active The man bit the dog with the passive The dog was bitten by the man)., languages (and Accordingly, the cognitive mechanisms for encoding them)languages must provide some means for securing intended relationships among entities across variations in the positions in which they occur. In theories of language production the maintenance of this equation is the responsibility of functional processing (Garrett, 1988).
The nature of the functional equation that is formulated during production is unknown, but an attractive hypothesis involves the construction(Goldberg, 1995). A construction is a scheme that relates an array of thematic roles (such as agent, goal, and patient) to a form, or aconceived as a set of structural positions. For example, iIn the dative sentence The man gave the girl the present, the agent is the man is the agent, the patient is the present, and the goalgirl is the girlgoal, and the present is the patient. For example in a double-object This dative construction relates the three roles to three structural positionsThere are 3 structural elements (corresponding to the subject, direct object, and second object, respectively). 2) and three corresponding roles (agent, recipient, and theme). For example, in thesentence The man gave the girl the present, the agent is the man, the patient is the present, and the goal is the girl. This sentence has three structural positions, subject, object, and second object. Thedouble-object construction thereby provides a mapping relating the agent to subject, the goal to direct object, and the patient to the second object position.
The aim of the present work was to evaluate the role of constructions in language production using the methodology of structural priming.
The aim of the present work was to evaluate the role of constructions in language production using the methodology of structural priming. Structural priming builds on the phenomenon of structural repetition. Structural repetition is seen in a tendency to reuse a previously produced sentence structure in a new, otherwise unrelated utterance (Bock, 1986). For example, after producing a double-object double objectative sentence sentence like The artist showed the police captain a sketch, speakers were more likely to describe a subsequent picture event using an analogous structureanother double-objectdouble-object sentence than if they had produced a structurally different priming sentence like such as (such as The artist showed a sketch to the police captain). That is, having formulated one double-objecdouble-objectt structure, speakers were more likelyprone to use another double-objectonestructure to describe an unrelated situation (e.g., The children are giving the teacher flowers). Conversely, the production of a prepositional object dative was associated with an increased likelihood of using a prepositional dative later on (e.g., The children are giving flowers to the teacher). Speakers thus tend to generalize structure from one sentence to another. The existence of this generalization makes it possible to probe the nature of the information that promotes it, exploring the kinds of relationships that support alternative arrangements of constituents in sentences.
A constructional account of functional processing predicts that constituents that play the same thematic roles in sentences should tend to occur in the same structural roles. Standing in the way of this kind of account is evidence that structural repetition does not hinge on the identity of thematic roles in prime and target sentences. Bock and Loebell (1990, Experiment 1) found that prepositional locatives (e.g., (The wealthy widow drove the Mercedes to the church) prime prepositional dative picture descriptions to the same degree as that prepositional datives primes e.g., (e.g., The wealthy widow gave the Mercedes to the church) prime other prepositional dativesrelative to double object controls (The wealthy widow gave the church the Mercedes). The prepositional locative and the prepositional dative share have a similar structural configuratione (NP VP [V NP PP[P NP]]), but they arguably differ in the event role of the prepositional argument. In the prepositional locative, the prepositional phrase encodes the location of the action, while in the dative it encodes the recipient of the action.
In aAnother experiment, Bock and Loebell found even stronger evidence against a purely thematic construal of structural repetition (Bock Loebell, 1990, Experiment 2). Locatives like The 747 was landing by the control tower primed passive utterances as much as passives like The 747 was alerted by the control tower, relative to active prime controls like The 747 radioed the control tower. The locatives and passives had similar syntactic structures (NP VP[AUX V PP[P NP]]), but the locatives had agents as subjects, while the passives had patients as subjects. These results suggest that structural repetition does not demand depend on overlap in the thematic roles of prime and target sentences.
More puzzling from the standpoint of the constructional hypothesis is that role overlap did not promote structural repetition: Locatives and passives were equally effective primes, suggesting that the roles of the constituents did not influence their positioning. Instead, Bock et al. (1992) proposed that basic conceptual features help to regulate functional mapping. Bock et al. varied the animacy of the subjects of active and passive sentences, and found that the presence of an animate subject in priming sentences increased the tendency to make animate entities the subjects of target sentences. Because this type of priming occurred over and above the general tendency to re-use the active or passive structure of priming sentences, the implication is that thematic roles per se may not participate in functional mapping. At the same time, elemental conceptual features (like animacy, which is a dimension of event-role taxonomies that distinguish agents from instruments, beneficiaries from goals, and so on) promoted similar positioning -- as subjects -- of like constituents between in the primes and targets.
The paradox in these findings is that aspects of both sentence meaning (the animacy of subjects) and sentence structure (active and passive form) appeared to be susceptible to priming, when the functional mapping itself was not. This clashes with a constructional account, where the emphasis is on the mapping between meaning and form involves an indissoluble association between them.
There is nonetheless a feature of existing research on structural repetition that may allowsmakes these results consistentto be reconciled with a constructional approach. Because English typically confounds the structural position of a phrase with its thematic properties, it is difficult to create changes to the thematic properties of a primeprimes and targets without corresponding changes to its their structural properties. For this reason, existing experiments orthogonally manipulated the correspondence between thematic roles and syntactic functions across primes and targetsin English has been manipulated in just one sentence position, leaving the normal correlation in other positions mostly intact. For example, in Bock et al. (1992), animacy and structure were manipulated independently in the priming sentences,, but of necessity, not in the targets: In the targets, a speaker's selection of an animate subject controlled the choice of the sentence structure; alternatively, andconversely, selection of a sentence structure controlled the animacy of the subject. Such manipulations, relying on a single role variation to drive a change in functional processing, may be insufficiently powerful to create the consistent relational differences that would could be necessary to showwould reveal the interaction of thematic roles and syntactic structure. In contrast, the structural components of repetition could have beenwaswere supported by the entire gross configuration of priming sentences.
Beginning with Pickering and Branigan (1998)(1998), structural priming more recent work has been shown that the magnitude of structural priming canto increase substantially with overlap in semantic and when there is lexical and semantic contentoverlap between prime and target sentences. Conversely,Lexical support may also increase the likelihood of priming between theorematically dissimilar but structurally similar sentences: Bencini, Bock, and Goldberg (2002)obtained significant locative/passive priming (replicating Bocks Loebell, 1990) when the preposition by was repeated (replicating Bock & Loebell, 1990), but in the absence of by repetition, the priming effect was not significant. This suggests Bencini, Bock, and Goldberg (in preparation) replicated Bock and Loebell’s locative/passive priming results, but showed that significant priming was eliminated when the preposition by was changed, suggesting that a shared preposition can helps to support passive priming. Overlap in thematic roles may also increase priming, according to findings from Hare and Goldberg (2000). Hare and Goldberg found evidence of thematic-role priming using the provide-withconstruction and the double- object constructions. The provide-with construction (e.g. The army provided the soldiers with blankets) has the same syntactic structure as the prepositional dative (NP VP[V NP PP[P NP]]), but the object argument is the recipient, and the second-object2 argument is a theme. This makes the mapping oforder of the roles similar to the order of roles in the double-object construction, which also puts the recipient directly after the verb, and beforefollowed by the theme. Hare and Goldberg found that provide-with sentences primed double-object expressions as much as double-object priminge sentences did. This suggests that the order of thematic roles can be primed. However, because Hare and Goldberg used role fillers that differ in animacy (animate recipients and inanimate themes), their results could be explained in the same terms as Bock et al. (1992) rather than in terms of thematic roles proper.
To achieve a straightforward test of eventthematic- role priming without the intrusion of animacy or the confounding influence of structural repetition requires a different tactic. What is needed are constructions in which the order of constituents varies without corresponding changes in syntactic structure or the animacy of the arguments. One type of sentence that fulfills these demands is the locative alternation (Anderson, 1971; Levin, 1993; Rappaport & Levin, 1985), also called the spray-load alternation. This alternation varies the order of the theme (the object that moves) and the location (the place that is moved to), both of which are typically inanimate. For example, in The man sprayed the wax on the car, wax is the theme and car is its final location. The alternative order puts the location before the theme, as in The man sprayed the car with the wax. The structure itself in both cases is traditionally NP VP[V NP PP[P NP]], so priming of the structural configuration should not differ. But if the order of the roles matters, then we should see theme-location orders priming other theme-location orders more than location-theme orders.
Since spray-load constructions are not easy to elicit reliably with picture description, we needed another task in which structural repetition occurs. One such task is Potter and Lombardi's (1998) rapid serial visual presentation (RSVP) sentence repetition paradigm. In this paradigm, participants simply silently read sentences that were shown onfrom a computer screen , sentences shown one word at a time, at a fairly fast rate. They The participants then performed a distractordistractoer task before being asked to repeating the sentence aloud. Because of the difficulty of explicitly remembering the surface forms of sentences, along with the high speed of presentation and the intervening task, speakers sometimes changed the syntactic structure of the sentences. Using this procedure, Potter and Lombardi (1998) found a tendency to reuse previously produced structures, which theywith participants tending to recall the sentence on a current trial in the grammatical structure used to recall a different sentence on the preceding trial. Potter and Lombardi attributed this to structural priming.
In addition, Potter and Lombardi found more priming of prepositional dative structures by prepositional datives than by prepositional locatives, in contrast to Bock and Loebell (1990). This suggests that their technique could be more sensitive to thematic role variations than the sentence-picture priming technique, but an alternative explanation is thatBecause this technique has not been used to contrast thematically different primes and targets when the lexical contents of their thematically contrasting primes differed. So, in the first experiment we set out to compare the RSVP paradigm to the sentence-picture priming paradigm, using primes and targets with one thematic role difference and a preposition difference, but no differences in open-class content (along the lines of Bock, 1989). Then, in the second experiment, we turned to the spray-load alternation, where there are two more substantial role differences of the primary from phrases is controlled, we first set out to see whether we could replicate the ending of Bock (1989) using Potter and Lombardi’s paradigm. Then is relatively new, we wanted to replicate Potter and Lombardi's (1998) results using materials that were previously employed in picture description paradigms, to gain information about the relative strength of priming effects. We therefore performed two experiments, one with the dative alternation and a second within the second experiment, we examined the spray-load alternation.
Two competing hypotheses were tested. The first, which we call the no-roles hypothesis, claims that the message representations that support sentence production do not individuate thematic roles as primitives, making it impossible to prime the roles themselves. The second hypothesis, which we call the construction hypothesis, says that event thematic roles are a proper part of the messages that drive production mechanisms. The hypotheses make the same predictions for the dative experiment, where the purpose was primarily methodological, because either animacy or thematic roles (the order of the patient and goal) could support priming. But in the spray-load experiment, the no-roles hypothesis predicts no difference in priming between the different orders of theme and location, while the construction hypothesis predicts just such a difference.
Experiment 1: Dative Priming
The first study was designed to examined the sensitivity of replicate work finding structural priming with dative sentences using the RSVP recall, as in production paradigm(Potter and Lombardi (, 1990), to differences in thematic roles with the content words of priming sentences controlled. We also varied the preposition used in the target sentence to see if there were variations in the likelihood of priming for datives with changed prepositions and slightly different roles.If priming in the RSVP paradigm is more vulnerable to thematic role effects, the dative priming observed by Bock (1989) between transfer and benefactive datives in the sentence-picture task could be absent from the RSVP task.