| |
In a Principles and Parameters syntax framework, sentences are derived by two operations, merger and movement. Merger takes two categories as input (e.g., V and NP) and merges them into a single, higher-order category (e.g., VP). There are, however, constraints on the categories that can be merged successfully. Consider the following pairs:
1a. [NP ;The girl] sneezed
1b. * [NP The girl] sneezed [NP the boy]
2a. [NP The girl] defeated [NP the boy]
2b. * [NP The girl] defeated
3a. [NP The girl] gave [NP the prize] [PP to [NP the boy]]
3b. * [NP The girl] gave [NP the prize]
The (a) examples above are well-formed sentences; the (b) versions, containing the same verbs but different structures following the verbs, are ill-formed. Thus, not all verbs can fit into all sentence structures. How, then, does a theory of syntax account for these facts? Borrowing from logic, we can say that sentences are composed of a verb (i.e., predicate) and a set of arguments. A verb denotes an activity or event and an argument denotes a participant in the activity or event. So, in the grammatical (a) versions above, the sentences contain the appropriate number of arguments the verb entails; in the ungrammatical (b) versions there is either an extra argument (as in (1b)) or a required argument is missing (as in (2b) and (3b)), hence violating the argument structure of the verb.
Not all the phrases in a sentence function as arguments of a verb. Consider:
4. [The girl] defeated [the boy] on the beach/this afternoon/with great finesse.
The bracketed NPs are clearly participants in the event denoted by defeated, and thus are arguments of the verb. However, the italicized phrases do not represent participants in the event. Instead, they carry additional information (i.e., where the event took place, when, and the manner in which it took place). These expressions are considered to be adjuncts; that is, they are adjunctive or in addition to the information specified by the verb. Simplifying a bit, adjuncts are typically optional while arguments are often required.
Thematic Roles
The NPs that are participants/arguments of the verb in (4) play different semantic roles in relation to the verb defeated. A more comprehensive account of argument structure, then, needs to consider a description of these roles. For example, in (4) the NP the boy is the Agent of defeat, while the girl is the affected object and hence the Patient (or Theme). Some common thematic roles, then, are Agent, Experiencer, Theme/Patient, Goal, and Location.
A verb (that is, a lexical category) assigns its thematic roles to its arguments through theta marking. For example, the verb defeat is said to theta-mark the subject argument with the Agent role and the object argument with the Theme (Patient) role:
Thus, each lexical category (e.g., verb) has a set of argument structure features that must be satisfied in the sentence in which the word appears. If those features are not satisfied, the sentence will be ungrammatical.
Verbs can also have clauses as arguments; these too need to be theta-marked:
The verb know assigns Experiencer to the subject NP argument and Proposition to the CP argument.
Processing
Canonical linking rules have been hypothesized to play a key role in the acquisition (van der Lely, 1994) and the processing of such verb-argument structures (see, e.g., McRae, Spivey-Knowlton, and Tanenhaus, 1998). This linking or mapping refers to the regular, most frequent relation found between thematic roles and syntactic functions (Pesetsky, 1995). For example, if an individual knows that a verb involves an Agent, Patient/Theme, and Goal, she can infer that those arguments can serve the role of subject, object, and oblique object, respectively. The verb donate, for example, requires three arguments—a subject NP, a direct object NP, and an indirect (oblique) object NP—as in:
7. [The girl/AGENT] donated [the present/THEME] to [the boy/GOAL].
Unlike (7), where the properties of the verb donate entail canonical linking, there are verbs with properties that entail noncanonical linking. Consider, for example, receive, which entails a reversal of the canonical assignment of Agent and Goal arguments:
8. [The boy/GOAL] received [the present/THEME] from [the girl/AGENT].
Sentences (7) and (8) reflect a well-known bias that suggests the “sender” seems more volitional and is a more plausible candidate for the Agent role than a “receiver” (Dowty, 1991). Importantly, there are no positional or configurational distinctions between the arguments to signal the difference in thematic order in the above examples; that is, the underlying syntax between the constructions appears to be the same. Thus, the distinction is based on the inherent properties of the verb.
Unlike the sentences above, where thematic roles can be directly assigned from inherent lexical information and linking relationships, sentences with so-called displaced arguments require indirect thematic role assignment. For example, consider the following noncanonical cleft-object sentence: “It was the boy who the girl kissed ___ yesterday.” The direct object NP the boy has been displaced from its canonical, post-verb argument position in the sentence, leaving a gap. Such constructions are often referred to as filler-gaps. Verb-argument structure properties influence such constructions rather directly; for example, in the cleft-object case, a verb must license a direct object argument position in order to form a filler-gap dependency.
Given its syntactic and semantic importance, then, argument structure (in various forms) has played a privileged role in accounts of language processing. One of the earliest attempts to show that such lexically based information has repercussions for normal adult sentence processing was that of Fodor, Garrett, and Bever (1968). They inserted verbs differing in grammatical complexity (defined, in current terms, by the types of arguments each allowed) into matched sentences and found that off-line performance on those sentences decreased when verbs were more complex. Similar effects were later found on-line (e.g., Shapiro, Zurif, and Grimshaw, 1987).
What is important here is not just the fact that there are observed “effects” of argument structure, but what those effects suggest about the architecture of the sentence processing system. Briefly, most current accounts claim that when a verb (or any theta-assigning head of a phrase, including prepositions) is encountered in a sentence, its various argument structure configurations are momentarily activated (see, e.g., Pritchett, 1992; MacDonald, Pearlmutter, and Seidenberg, 1994). On some accounts this information is ordered in terms of “preference,” which then helps determine which of a set of parses the system initially attempts (Shapiro, Nagel, and Levine, 1993; Trueswell, Tanenhaus, and Kello, 1993). Such preference effects suggest that argument structure information may be used immediately to help analyze sentence input. Indeed, an influential set of theories (e.g., MacDonald, Pearlmutter, and Seidenberg, 1994) suggests just that.
However, there remains an equally influential alternative which suggests that there are (at least) two passes through a sentence (Frazier and Clifton, 1996). The first pass considers only categorical information (e.g., DET, N, NP, V, VP, etc.) and perhaps the number of arguments a verb entails, and essentially builds a skeletal phrase structure representation of the input; the second pass considers lexical-semantic and contextual information. In some of these accounts, detailed thematic information is explicitly claimed to be part of the second-pass analysis (Friederici, Hahne, and Mecklinger, 1996).
Finally, the representation and processing of argument structure has important implications for language disorders underlying aphasia. Briefly, the “mapping deficit” account (e.g., Schwartz et al., 1987) has suggested that the sentence comprehension patterns evinced by some agrammatic Broca's aphasic individuals may be explained by their inability to “map” thematic roles onto grammatical (i.e., subject, object) positions, particularly in sentences that have noncanonical mapping. A more detailed and circumscribed account of the deficit is offered by the trace deletion hypothesis (e.g., Grodzinsky, 2000). Here, the claim is that knowledge of argument structure is intact for these individuals (for on-line evidence of this fact, see Shapiro et al., 1993). However, traces of moved referential NPs or arguments are deleted, and hence indirect thematic role assignment is blocked. Instead, these individuals appear to use an “agent-first” strategy for arguments that cannot receive a grammatically computed thematic role, explaining performance on a wide range of sentence types.
Unlike Broca's aphasic individuals, those individuals most likely characterized as Wernicke's syndrome type appear to be insensitive to the argument structure properties of verbs, even where on-line comprehension is at issue (Shapiro et al., 1993; Russo, Peach, and Shapiro, 1998). Yet, their deficit does not seem to affect on-line comprehension of sentences with moved arguments (Zurif et al., 1993). These patterns therefore suggest a double dissociation between the activation of argument structures and the syntactic parsing routines underlying the comprehension of sentences with moved arguments.
| |
