Journal of Psycholinguistic Research, Vol. 27, No. 2, 1998

Getting There (Slowly) Lyn Frazier1,2 Accepted July 1, 1997 The general foundations for a theory of human sentence processing and the basic goals and questions of the 1960s are argued to be solid and still very relevant today, even if the particular processing model of the time (the clausal model) does not hold up. Today there is considerable agreement that sentence processing is both fast and grammatically controlled from the outset, and that minimal structure and recent attachments are favored across a diverse range of languages. Parsing is not purely bottom-up but instead allows phrases to be postulated before all daughters have been parsed; it works similarly for head-initial and head-final languages. The parser takes as input a rich prosodic/intonational representation which influences processing in ways that extend far beyond use of intonational boundaries as local (juncture) cues. It is argued that psycholinguistic evidence disconfirms the use of prepackaged fully articulated X' templates for identifying phrase structure, instead supporting the extended projections of Grimshaw (1991). Further, considerable evidence shows that the processor respects grammatical distinctions among types of dependencies, as indicated by differential processing effects for antecedent-government chains versus binding relations, by the processing difficulty of composite dependencies involving more than one grammatical type of dependency, and by distinctions between deep and surface anaphora. The central problem for future theories of sentence processing is claimed to be the development of theories of sentence interpretation. Various possible approaches are discussed, including parallel computation of alternatives with subsequent selection, a task-driven interpretive system, and underspeciftcation. The question is raised whether sentence-level psycholinguistics is in the middle of a paradigm shift from models with symbolic representations and an intrinsically serial (Von Neumann) architecture to constraint-satisfaction/connectionist models with inherently parallel architectures. Reasons are offered for rejecting current constraint-satisfaction models, including lack of supporting evidence discriminating between constraint-satisfaction and "conventional" models, serious problems with the use of prepackaged X' templates for identifying phrase structure (the only proposal to date for handling syntactic structures in constraint-based models) and an inability to deal with cross-language gen-

1 2

University of Massachusetts, Amherst, Massachusetts 01003-7130. Address all correspondence to Lyn Frazier, Department of Linguistics, University of Massachusetts, Amherst, 226 South College, Amherst, Massachusetts 01003-7130.

123 0090-6905/98/0300-0123$15.00/0 © 1998 Plenum Publishing Corporation

124

Frazier

eralizations. Pure connectionist models are unlikely to fare better in the future for two reasons. They are apparently unable to handle restricted universal quantification (Marcus, 1998) which clearly lies within the capacity of the human sentence processor. Further, competition between alternatives lies at the very heart of such models. Maximal competition should occur in processing fully ambiguous sentences and thus they are predicted to take longer to process than biased or temporarily ambiguous ones (Frazier & Clifton, 1997). But the empirical evidence suggests that fully ambiguous sentences are not systematically harder to process than their disambiguated counterparts.

WHERE WE'VE BEEN: EARLY CONTEMPORARY PSYCHOLINGUISTICS One cannot help but be impressed by the accomplishments of language comprehension investigators in the 1960s—the era extending from 1960 to 1974 when Fodor, Bever, and Garrett's (1974) seminal textbook was published. Investigators in this era inherited an earlier view of sentences as simple chains of words with relations between adjacent words, and the results of numerous paired-associate learning experiments designed to discover how those relations were established. A few psychologists—unlike today, language comprehension investigators then were nearly all psychologists— embraced Chomsky's (1957) theory of generative grammar. They were swayed, for example, by clear linguistic arguments that discontinuous dependencies exist between nonadjacent words of a sentence. Equipped with a generative (explicit) theory of grammar and a new focus on language competence (the ability underlying the productive use of novel sentences), they devised an impressive set of imaginative experiments demonstrating the psychological reality of grammatical structure (Fodor et al., 1974). Verbal behavior, and the results of their experiments, could not be explained without appeal to the syntactic structures linguists postulate. Transitional probabilities would not suffice; see Bever, 1973, Johnson, 1965, and many others. Semantic plausibility would not suffice (e.g., Forster & Ryder, 1971). In the process of demonstrating the reality of grammatical structure, they developed a battery of new methods for investigating sentence processing experimentally, including (click) monitoring (Fodor & Bever, 1965; Ladefoged & Broadbent, 1960), probe studies (Caplan, 1972; Stewart & Gough, 1967), rapid serial visual presentation (Forster, 1970), phoneme monitoring (Foss, 1970), and dichotic presentation of ambiguous sentences and biasing context (Lackner & Garrett, 1973). In this period, the questions changed from how do humans learn associations between words to questions of what structures are assigned, un-

Getting There (Slowly)

125

derlying or surface structures (Bever, Lackner, & Kirk, 1969; Wanner, 1968), and how structure is assigned, including in cases of discontinuous and long distance dependencies (Wanner & Maratsos 1971, 1978). The importance of ambiguity as a window on internal sentence processing routines was clearly recognized (Bever, 1970; Bever, Garrett, & Hurtig, 1973; Cairns, 1970; Lackner & Garrett, 1973). Convincing arguments were presented for drawing a distinction between grammaticality and acceptability (Miller & Chomsky, 1963). Restrictions on immediate memory capacity (Miller, 1956) were viewed as central then, as they are today (Lewis, 1996), even if today they are no longer cast primarily in terms of the number of chunks to be remembered but instead in terms of resource limitations on reapplying the same grammatical kind of operation (Stabler, 1994; but also see Miller & Isard, 1963). Important cross-language questions were addressed (Forster, 1966, 1968; Yngve, 1960), though studies specifically targeting cross-language comparisons were the exception, not the rule. General questions about why particular structures are preferred were highlighted (Bever, 1970), and important insights were achieved concerning the relation between distinct performance systems (Bever & Langendoen 1971).

WHERE ARE WE NOW? CURRENT THEORIES OF SENTENCE PROCESSING Results Concerning the Sentence Processor So what has changed? The assumptions and questions of the 1960s sound deeply familiar today. What contributions have been made in the decades following this auspicious beginning? In my opinion, two things have changed. The particular model of how perceivers assign a grammatical structure to a sentence has changed dramatically. The theory of grammar has also changed with the development of nonderivational grammars, and a much better understanding of the phonology/syntax interface and of the syntax/semantics interface. In the 1960s investigators of language processing were hampered by an unaccommodating grammar. The grammar of the time was crucially cast as a derivational mapping defined for at least a clause-sized domain. This sort of grammar created a dilemma: Either grammatical processing could be fast and not grammatically controlled (Marslen-Wilson, 1973; Marslen-Wilson & Tyler, 1980) or it could be "slow" (clause-based) and the grammar could be applied heuristically in analysis by synthesis mode or, problematically, backward in analysis-by-analysis mode. Fast (word-by-word) and grammatically controlled sentence processing was not really an option until

126

Frazier

the theoretical contract "trace" was postulated (Fiengo, 1977; but see Wanner & Maratsos, 1971, for a clever attempt which approximates a processing equivalent of trace theory). Hence, grammatically guided guesses about within-clause structure were coupled with a later grammatical check at clause boundaries in the "clausal model" of sentence processing (Fodor et al., 1974). Most theories of sentences processing today incorporate the claim that parsing is both fast and grammatically controlled. A syntactic representation is developed essentially as each input word is encountered, as evidenced by eye movement data (e.g., Frazier & Rayner, 1982, and many others) and evoked-response potentials (e.g., Neville, Nicol, Barss, Forster, & Garrett, 1991, and many others). Each new word is incorporated into the current representation in accord with the grammar (Frazier, 1987a; Gibson, 1991; Gorrell, 1995; Pritchett, 1992). Hence preference strategies, favoring one rather than another grammatical analysis, not guessing heuristics, guide early syntactic analysis in most current models. Further, today the emphasis is on temporary, not full, ambiguity. From the perspective of an immediate commitment parser which assigns structure to incoming items immediately, ambiguity of analysis poses a serious problem even if the ambiguity does not persist until the end of the sentence. Current models range from strictly serial (e.g., Frazier, 1987a) to models with limited parallelism (e.g., Gibson, 1991). Fully parallel models of syntactic processing have received no empirical support. What has never been found, for example, is that the processing complexity of a sentence patterns together with other sentences with the same number of (in principle) alternative structures, as would have been expected on fully parallel models, e.g., with two-way ambiguous sentences processed faster on either analysis than three-way ambiguous sentences on any of their analyses. Discriminating empirically between serial versus limited parallel models is difficult. The predictions of these two types of models are very similar once limited parallel models rank syntactic analyses so that the single analysis computed initially in a serial model corresponds to a favored analysis computed in a parallel model. Two questions emerge for limited parallel models: If all relevant syntactic analyses of an input have been computed, why should the processor then favor one particular analysis, e.g., the one with new material attached to the most recent node? And why should long reanalysis times, or regressive eye movements, occur at the point where a favored analysis must be abandoned? Why doesn't the processor simply give up its current favored analysis for another already computed one? In the "beam search" model of Gibson (1991), by far the most explicit and well motivated of the limited parallel models that have been proposed,

Getting There (Slowly)

127

the processor is viewed as having the capacity to develop multiple syntactic analyses, but not to maintain all of them. This makes the counterintuitive prediction that unfavored analyses that will be pruned over time will be easiest to comprehend at fast presentation rates, because multiple analyses need not be maintained in memory for as long a duration as slow presentation rates. (See Waters & Caplan, 1996, for evidence that fast presentation rates do not facilitate the processing of garden-path sentences.) As for explaining why potentially long and variable reanalysis times are observed in processing studies of unfavored syntactic structures that lie within the processor's capacity (i.e., acceptable but difficult to process structures), the challenge for the beam search model is accounting for evidence that the complexity of diagnosing the error in the first or favored analysis predicts reanalysis times (Fodor & Inoue, 1994). This is mysterious in a system where the correct analysis is claimed to have been computed already by the time disambiguating evidence arrives. Across proponents of distinct models, considerable agreement exists concerning the basic principles governing the preferred (or initially only) syntactic analysis pursued by the processor. Basically the processor postulates the minimal structure consistent with current evidence whether this is cast as minimal attachment (Frazier, 1978; Frazier & Fodor, 1978), thematic reception (Gibson, 1991), simplicity (Gorrell, 1995), or minimal everything (Inoue & Fodor, 1995). Depending on details, the particular minimal-structure principle either entails an argument-preference principle (Abney, 1989; Pritchett, 1992) or needs to be supplemented with one. The minimal-structure principle also applies to empty categories, e.g., the active filler strategy (Frazier, 1987b), the active trace strategy (Crocker, 1992), the minimal chain principle (De Vincenzi, 1991), the thematic reception or thematic transmission (Gibson, 1991), and including the logical form (Frazier, in press; Pritchett & Whitman, 1995; Tunstall, 1998). Most theories include a preference for attachment to recently postulated nodes in the case of ambiguity (right association—Kimball, 1973; late closure—Frazier, 1978; recency—Gibson, 1991; Gibson, Pearlmutter, Canseco-Gonzalez, & Hickock, 1996) or more generally (visibility—Frazier & Clifton in press; Stevenson, 1994). Many detailed theories of syntactic processing exist. Occasionally they differ in their gross empirical predictions (see discussion of head-final structures below). But for the most part (several important debates aside), the theories differ in the explanation for the parser's preferences (frequency— Mitchell, Cuetos, Corley, & Brysbaert, 1996, but cf. also Gibson, Schutze, & Salomon, 1996; the need to quickly structure material—Frazier, 1978; or the desire to interpret rapidly—Crocker, 1992). Theories also diverge on whether nonsyntactic information influences competition between alterna-

128

Frazier

tives from the outset (MacDonald, Pearlmutter, & Seidenberg, 1994) or whether such effects arise from interpretation and possible revision of the first analysis (see Fodor & Ferreira, in press). One of the most striking advances in psycholinguistics in the last decade has been the rapid emergence of sentence processing studies on languages other than English. Studies of Germanic languages and various Asian languages with head-final word order have systematically indicated that phrases can be postulated before the head of the phrase is encountered (for Dutch, see Frazier, 1987b; Frazier & Flores d'Arcais, 1989; for German, see Bader & Lasser, 1994; Hemforth, Konieczny, Scheepers, & Strube, 1993; Schriefers, Friederici, & Kuhn, 1995; for Korean, see Koh, 1997; for Japanese, see Inoue & Fodor, 1995, and many others). The results of these studies exclude bottom-up processing models where phrases are not postulated until all of their constituents have been recognized (Pritchett, 1991). Studies of these languages have also raised important issues concerning the role of overt or morphological case in processing (Bader, in press) and possible differences across languages in the ease or method of reanalysis (Inoue & Fodor, 1995; Sturt & Crocker, 1995) as well as effects of already received lexical heads (Konieczny, Hemforth, Scheepers, & Strube, 1997). Broad agreement exists concerning the processing of distinct language types. Though important debates continue, the same underlying processing principles have been argued to apply in head-initial and head-final languages (Inoue & Fodor, 1995), in strict word order and "free" word order languages (Sekerina, 1997), in overt versus null subject languages (De Vincenzi, 1991), and in languages with prosodically versus syntactically marked focus (Rado, 1998). Some processing differences among languages have been identified (Carreiras, 1992; Carreiras & Clifton, 1993; Cuetos & Mitchell, 1988; Gilboy, Sopena, Clifton, & Frazier, 1995; Konieczny, Hemforth, Scheepers, & Strube, 1997) which may or may not be attributed to differences in the grammars of the languages. However, these on-going debates should not be allowed to obscure the incredible similarities that have been established. For example, each of the above properties of the sentence processor, that it is fast and grammatically controlled, not fully parallel or strictly bottom-up, and that it exhibits preferences for minimal structure and recency in initial analysis, have been established for widely diverse languages. Notable progress has also been made in our understanding of prosody/intonation and its effects on syntactic analysis (as evidenced by two recent special issues of journals devoted to prosody, Journal of Psycholinguistic Research [Nicol, 1996] and Language and Cognitive Processes [Warren, 1996]). In the 1960s formal theories of intonation (the phonological components of prosodic constituents and their phonetic interpretation) were

Getting There (Slowly)

129

not well developed and explicit theories of the phonology-syntax mapping essentially did not exist. This situation changed dramatically in the 1980s (Pierrehumbert, 1980; Pierrehumbert & Beckman, 1988; Selkirk, 1984, 1995). Although many early studies recognized the importance of intonation in parsing (Lehiste, 1972), the bulk of such studies demonstrated that some acoustic cue to a boundary is (or is not) sufficient to disambiguate a particular structural ambiguity. Working with explicit theories of intonation and sophisticated speech software, investigators are now in a position to offer a reasonable description of the particular pronunciations of the experimental materials perceivers receive and to manipulate the speech input in order to test explicit hypotheses. Results of such studies indicate that perceives are sensitive to a rich prosodic/intonational structure, not just a local intonational boundary. For example, Schafer (1998) has presented experimental evidence showing that the interpretation of an intonational boundary preceding a syntactically ambiguous phrase depends on the presence and location of other intonational boundaries, and the presence and kind of pitch accent preceding the intonational boundary. She has also shown that reanalysis of an ambiguous word is more difficult following an intonational boundary, and that intonational boundaries can govern the interpretation of prenominal modifiers, e.g., whether an expensive Porsche receives an intersective interpretation (expensive and a Porsche) or a subsective one (expensive for a Porsche). In sentences with a moved constituent, the trace bound by the moved constituent (who) is preferentially located in the same intonational phrase as its binder (who) (Broderick, 1996; Nagel, Shapiro, & Nawy, 1994). In short, intonational structure can no longer be treated merely as a local cue signaling the position of an ambiguous syntactic boundary; it provides a richly structured input to syntactic and semantic analysis, not just a local boundary marker, and it influences a wide array of processing decisions, not just "grouping" decisions. Results Concerning Linguistic Representation Important, if sometimes overlooked, results have emerged from sentence processing studies concerning the nature of linguistic structure. Consider ambiguities involving the attachment of a prepositional phrase (PP) following a direct object in a verb-noun phrase-prepositional phrase sequence. There is an ongoing debate concerning whether the PP is initially preferentially attached to the verb phrase (VP) under all circumstances (Clifton, Speer, & Abney, 1991) or whether this preference holds only for certain verbs (Spivey-Knowlton & Sedivy, 1995) or only with head nouns that do not take arguments (Schutze & Gibson, 1996). However, no one on any side of this debate argues that the PP is preferentially and generally interpreted

130

Frazier

as modifying the direct object when the PP cannot be an argument of either the verb or the noun. Yet this NP attachment is precisely what would be predicted by all theories incorporating a right association/late closure/recency principle if the phrase structure in (la) is projected even when X does not branch.

This can be seen in (1b) where clearly the NP attachment involves adjunction to a more recently postulated nominal node rather than to a more distant verbal node. The NP modification would thus allow a more recent attachment into the (noun) phrase currently being processed. There is near-universal agreement that some form of a recency principle is required to characterize syntactic processing preferences. Further, nothing in contemporary linguistic theory requires an X node to be postulated in cases of exhaustive (nonbranching) immediate self-domination, as in (la) where X exhaustively dominates X. The implication of PP attachment preferences is thus obvious—a nonbranching X need not be and is not postulated during sentence processing. It must therefore be assumed that it is the structure in (1c) that the parser builds. It follows that a prepackaged, fully articulated X template like that shown in (la) does not offer the correct description of phrase structure either in linguistic theory (which doesn't need it) or psycholinguistic theory, which can employ it only at the cost of incorrectly predicting a general NP modification preference for (nonargument) PPs in a V-NP-PP sequence. Note that this argument does not in any way depend on the sensitivity of current psycholinguistic techniques for measuring processing complexity. The argument does not rest on the absence of any observed complexity effect at the point in processing where the nonbranching N1 would be postulated. It rests entirely on the consequence of the presence or absence of this node given an independently motivated syntactic processing preference principle (recency). Grimshaw (1991) proposed that functional projections extend lexical projections. For example, an inflectional element may extend the (lexical) VP projection, forming "IP." A complementizer may be also added to this extended VP projection (the IP), to form a "CP." Similarly, a (lexical) NP projection may be extended by adding a determiner to form an extended nominal projection "DP." This theory of phrase structure receives psycholinguistic support from two independent lines of evidence. Frazier and Clif-

Getting There (Slowly)

131

ton (1997) argued that nonprimary phrases such as adjuncts are "associated" into the current thematic processing domain—defined as the extended projection of the last theta assigner. Empirical evidence supporting the theory derives from examples with extended verbal projections (e.g., analysis of ambiguous adverbial clauses and adjunct predicates) as well as examples with extended nominal projections (e.g., the daughter of the colonel who . . .). Hence precisely the cut that Grimshaw's theory makes is essential to the characterization of processing preferences. Gorrell (1996) presented additional support for extended projections, crucially employing them to account for effects of word order variations on the processing of German sentences. Thus dramatically distinct types of evidence, drawn from different languages, support Grimshaw's view of phrase structure. Let me emphasize that these results could not be recast in terms of clauses, and thus they represent another sharp difference between the early clausal model of processing and recent models. Turning to the analysis of moved constituents, abundant evidence supports the existence of "traces" (the empty category that a moved constituent binds). The evidence derives from cross-modal priming studies (Love & Swinney, 1996, and references therein), reading studies involving eye movements (e.g., Traxler & Pickering, 1996), self-paced reading (e.g., Frazier & Clifton, 1989; Stowe, 1986), and event related potential (ERP) studies (e.g., Friederici & Mecklinger, 1996, Mecklinger, Schriefers, Steinhaus, & Friederici, 1995), for example. Some of the evidence involving postverbal direct object traces might be attributed to a relation between the moved constituent and the theta-assigning verb itself (Pickering & Barry, 1991; but cf. also Gibson & Hickok, 1993). However, many of the results could not. For example, early gap preferences are observed in subject-object-verb (SOV) languages, before the verb is encountered (Frazier, 1987b, for Dutch; Schriefers et al., 1995, for German; Nagai, 1995, for Japanese, and many others). In Italian, De Vincenzi (1996) has studied ambiguous extraction from embedded clauses containing or lacking an overt complementizer (che/"that"). The generally observed subject extraction preference observed in main clauses and embedded clauses without an overt complementizer disappears in clauses with an overt complementizer. Rizzi (1982) has argued that extraction following an overt complementizer (which is ungrammatical in English) is grammatical in Italian because extraction may occur from postverbal subject position, though this entails a longer chain than direct extraction from preverbal position assuming that the subject must first move to postverbal position (leaving a trace) and then move again (leaving another trace). Given De Vincenzi's (1991) minimal chain principle, the loss of the subject extraction (over object extraction) preference in clauses with overt complementizers follows immediately. However, given a theory without

132

Frazier

empty categories but only direct associations between moved constituents and theta assigners, the result is completely unexplained and mysterious. De Vincenzi (1991, 1996) also showed that the grammatical type of dependency matters. The Italian counterpart to who, which enters into antecedent-government chains, shows more robust early gap preferences than the counterpart to which-N. Following Cinque (1992), she assumed that which-N enters into binding relations, not antecedent-government chains (see also Hickok & Avrutin, 1996). At logical form (LF), minimal chains seem to be preferred. Quantifiers are interpreted with their surface scope when possible, taking scope over phrases which c-command them at surface structure only when evidence warrants movement (which entails a more complex chain) (see Frazier, in press; Pritchett & Whitman, 1995; Tunstall, 1998). With respect to NP-trace, the processing evidence is less clear (see Fodor, 1989, for discussion). The binder of an NP-trace acts like it's reactivated after the NP-trace (Bever & McElree, 1988); however, one can imagine accounts of such effects that do not entail the existence of an empty category in the syntactic representation of the sentence, as Fodor (1989) pointed out. Rado (1994) showed that "mixed chains" or composite dependencies of distinct grammatical types are difficult to process, namely, dependencies involving both reflexive binding and movement to a nonargument position (A' movement). In Hungarian and English questionnaire studies, a particular NP is less likely to be chosen as antecedent for a reflexive if that NP has been A' moved (as in wh movement) than if it has not. Further on-line studies have shown longer processing times when the only grammatically permissible antecedent for the reflexive has moved than when some other NP has been A' moved. Processing studies of deep versus surface anaphora lend support to Hankamer and Sag's (1976) distinction between types of anaphors (see Tanenhaus & Carlson, 1990). Topicality profoundly affects the processing of deep anaphors but not surface anaphors (Ward, Sproat, & McKoon, 1991), whereas the syntactic form of the antecedent seems to have a larger or more general impact on surface anaphors than deep anaphors. Summary Obviously the level of refinement and detail of sentence processing theories has sky-rocketed since 1960 as well as the sophistication of the grammatical questions addressed. But just as obviously, the theories of today rest on the foundations built in the 1960s. The underlying assumptions of today's theories and the basic kinds of questions addressed do not differ: Only the answers to the questions have changed.

Getting There (Slowly)

133

To summarize, there is broad if not complete agreement among proponents of distinct processing theories for the following points: (i) The grammar guides sentence processing from the outset; its role is not restricted to a post hoc checking phase following the application of nonlinguistic heuristics. (ii) Input items are rapidly incorporated into a syntactic representation without substantial delays (e.g., until the next potential clause boundary) Corollary: Clause boundary identification is a consequence of syntactic analysis of constituents, not a precondition as in the clausal model. (iii) Syntactic analysis is not fully parallel—it is either strictly serial with rapid revision (Fodor & Inoue, 1994; Frazier, 1978) or there is constrained parallelism (as in the beam search model of Gibson, 1991). (iv) The syntactic processor favors recent attachments and minimal structure in the domain of phrase structures and chains (filler-gap structures) at both surface structure and logical form. (v) Syntactic analysis is not purely bottom-up; a phrase may be postulated before its head is encountered and before all its constituents have been encountered. (vi) Cross-language similarities in processing (with respect to all of the above points) by far outweigh the relatively few differences between languages that have been documented. The cross-language differences may or may not themselves be attributed to variations in the grammars of the particular languages, being only distinct superficial outcomes of applying identical underlying principles. (vii) A rich prosodic/intonational representation is the input to syntactic analysis. Intonation cannot be treated as a local cue anymore. Concerning linguistic implications of processing studies, reasonable evidence supports the following additional conclusions: (viii) Phrases are not stored as prepackaged fully-articulated X templates. Nonbranching self-domination is not represented, at least at the X level. (ix) Functional projections extend lexical projections. (x) Trace theory is supported at least for traces of movement to nonargument positions. (xi) Grammaticality distinct dependency types (nonreferential, antecedent government vs. "discourse-linked" or binding relations) differ in their impact on processing.

Frazier

134

(xii) Composite dependencies where a phrase enters into more than one type of dependency (A' movement, binding) increase processing complexity. (xiii) The distinction between deep and surface anaphora receives psycholinguistic support: deep anaphors exhibit more robust topic/focus effects whereas surface anaphors are apparently more open to effects of syntactic form and possibly to the amount of material intervening between antecedent and anaphor when topicality effects are controlled. Clearly psycholinguists are not done with their work yet. Indeed we're still getting started. We may rest assured that tomorrow will hold surprises. Nevertheless, the broad points of agreement above allow more sophisticated questions to be addressed as well as permit new results to be interpreted in light of independently motivated generalizations and assumptions. Thus, to my mind, the above results represent significant progress in the area of sentence processing theories. Today we need not investigate sentence processing in a vacuum. Broadly established generalizations can be reasonably if tentatively assumed. Competing theories exist which help both to elucidate unstated assumptions and to guide hypothesis formation. Experimental techniques have been developed to test hypotheses, including nonintrusive online techniques such as eye movement recording in reading environments and in situations involving the use of spoken language (Eberhard, SpiveyKnowlton, Sedivy, & Tanenhaus, 1995) as well as rapidly developing techniques involving brain recording and imaging.

WHERE WE'RE GOING Turning to today's controversies and the questions of tomorrow, anyone familiar with psycholinguistics will have detected the omission in earlier sections of this paper of certain fundamental contributions (as well as numerous detailed ones) and a loud silence concerning certain raging controversies. The controversies primarily concern the architecture of the processor (Fodor, 1983), the contributions of referential theory (Crain & Steedman, 1985), and the interplay of syntactic and nonsyntactic information/phenomena (e.g., MacDonald et al., 1994). Fodor's (1983) important modularity thesis has shaped the way many psycholinguists think about the human language capacity. With respect to initial form-based phonological, morphological, and syntactic processing, Fodor's thesis is at least arguably true—though not everyone would agree. Where it obviously falls down is with respect to semantic/pragmatic/discourse

Getting There (Slowly)

135

processing—a domain not clearly included in the grammatical processing system on Fodor's view. Only the purely linguistic form-based system can be shallow, fast, automatic, domain specific, and information encapsulated. Pragmatic and discourse processing clearly involve world knowledge and nonlinguistic inferences found also in nonlinguistic communication. Hence, they could not be domain specific or information encapsulated. Semantics is difficult to isolate from these systems, especially if the computation of sentence-meaning does not precede the computation of speaker meaning. What to do? One possibility is to compute multiple structural possibilities within the grammatical processing system and then later choose the best alternative at later levels (as in Forster's, 1979, autonomy model). Crain and Steedman's (1985) referential theory is just such a system where multiple grammatical analyses are computed but then pragmatic principles (the avoidance of presupposition violations) are used on a word-by-word basis to identify the best analysis to pursue. Conceptually this is a very attractive solution. However, there are empirical problems with it. Both minimal structure preferences (e.g., the VP-attachment preference in Spivey-Knowlton & Sedivey, 1995) and recency preferences (e.g., Gibson et al., 1996) have been established. In some cases, these preferences are not predicted by referential theory and in others they explicitly disconfirm predictions of the theory. To account for these and other findings establishing asymmetries in the structure initially preferred by perceivers,one might devise a hybrid theory with generation of syntactic possibilities ordered to reflect minimal structure and recency principles, with referential evaluation of the alternatives as they become available. The problem with this is that the referential/presuppositional properties of alternative analyses will determine the selection of an analysis to pursue only in the rare situation where referential evaluation can be accomplished at the first word of an ambiguity. Otherwise the syntactically nonpreferred alternatives will be dismissed before they can be evaluated and the processor will behave exactly like a serial syntactic processor. Consider The horse quickly (raced past the barn . . .). At quickly, the processor will already have to choose between attaching the adverb to the main clause or to a reduced relative clause. Presumably (by the minimal structure principle) it will opt for the main clause analysis. At this point, only one alternative, the main clause continuation, will remain. Except in limited circumstances, the hybrid model will behave like a serial model and need to incorporate precisely the reanalysis mechanisms needed in the serial model. The restriction on word-by-word evaluation and pruning of alternatives might be relaxed to circumvent this problem. But then the model will again fail to account for the minimal structure and recency preferences, because it will pursue multiple structures for several words and thus it should not

136

Frazier

matter which structure is consistent with subsequent material (though see also Ni, Braze, Conway, Crain, & Shankweiler, 1996, for one demonstration supporting the parallel view). Another kind of problem for the referential theory does not concern the parallelism implicit in the model. Instead it stems from recent studies suggesting that "common ground" (presupposed information assumed to be shared by speaker and listener) does not influence early stages of comprehension, which instead are based only on information salient to the perceiver (Horton & Keysar, 1996; Keysar, in press). Only after an initial interpretation is assigned do perceivers check whether the information used to identify a referent as familiar is shared information (part of the common ground with the speaker) or privileged information available only to the perceiver. The late checking for the common ground status of the perceiver's assumptions is time consuming and hence may be omitted under time pressure, according to Keysar's model and evidence. Clearly there is a tension between these findings and referential theory, which, claims that the avoidance of presupposition violations is the core mechanism for early syntactic ambiguity resolution and the determination of which syntactic analysis should be pursued by the parser. Another problem, not for referential theory but for the general approach of pursuing multiple grammatical analyses with later selection of the most plausible one, is the observation that, other things being equal, scope preferences follow surface c-command relations (Frazier, in press; Ioup, 1975; Pritchett & Whitman, 1995; Tunstall, 1998; Van Lehn, 1978). If the sentence processor computed all permissible scope relations and later chose the pragmatically most plausible one, this basic generalization would remain unexplained. Task-driven processing is another option for how a domain-specific information-encapsulated grammatical processing system could compute interpretations for sentences even though those interpretations are based in part on nongrammatical information. This is the approach pursued in Frazier (1990, 1995b). The idea is that the linguistic representations themselves implicitly define tasks and the interpretive system may import or consult information needed to execute those tasks. On this view, the interpretive subsystem is "pseudoencapsulated." Its task is linguistically determined in advance of any particular sentence (helping to explain the speed and automaticity of the system; Marslen-Wilson & Tyler, 1987) but the information it consults may include information imported from outside the interpretive module. For example, by its very form, an unbound pronoun defines a task of identifying an antecedent. A referential phrase in an argument position must find or establish a discourse referent. Lexical requirements, such as the fa-

Getting There (Slowly)

137

miliarity restriction of the, must be met. Presumably the content of context variables, such as the implicit variable in "local" (meaning "near X"), must be determined, at least under some circumstances. On this view, each linguistic requirement may be viewed as implicitly defining a task. A task-driven interpretive system fits well with an approach to semantics which identifies the context-dependent aspect of semantics and introduces a corresponding "hole" into the semantic representation such as a variable that needs to be bound or an anaphor which requires an antecedent. Rooth (1992) took this approach to focus; von Fintel (1994) took this approach to topic. See Rado (1998) for processing studies of Hungarian of this general type. Underspecification in the semantic representation is another way to proceed with interpretation without requiring the processor to compute multiple fully determinate syntactic and semantic representations. Frazier and Rayner (1990) presented evidence supporting this approach to word sense alternatives. The processor is not faced with grammatically incompatible alternatives when it encounters a word with, say, an abstract versus a concrete interpretation (e.g., book as a concrete object vs. an abstract intellectual product) and thus it may remain noncommittal, choosing a particular sense if and when biasing evidence is encountered. Whether underspecification applies only in cases of vagueness is not known. In principle, it might also be used as a way of "collapsing" grammatically incompatible alternatives, e.g., if the surface syntax were used as an underspecified representation for sentences with multiple (contradictory) scope assignments or using the surface form of the word bank as an underspecified representation of its various meanings. Whatever the answer turns out to be, theories of sentence processing will not be complete until the answers to such question are known. On all views of processing, syntactic processing is only a part of the language comprehension problem. Further, even deciding between alternative accounts of syntactic parsing may depend increasingly on an explicit account of how and when the constituents of a sentence are interpreted. At present psycholinguistics is a long way from understanding how interpretations are assigned. Setting aside the study of pronouns, the vast majority of studies of interpretation have focused either on the processing of prestored representations (word meanings) or text-level story comprehension, where it is typically assumed that the perceiver has somehow arrived at a propositional representation of the meaning of a clause and the questions addressed concern the bridging inferences that are or are not drawn to integrate that proposition with preceding text. Of course, there are important exceptions. Carlson and Tanenhaus (1988) examined the role of thematic grids in discourse comprehension. Murphy (1990) and Potter and Faulconer

138

Frazier

(1979) investigated how nouns are modified. Several studies of quantifiers also exist (Kurtzman & MacDonald, 1993; Sanford, Moxey, & Patterson, 1994, and references above) as well as studies of how metaphorical interpretations are assigned (e.g., Onishi & Murphy, 1993). But the fact remains that psycholinguistic investigations of those interpretive processes intervening between word interpretation and higher-level text processing are isolated exceptions. To my knowledge today there simply do not exist any empirically motivated theories of this phase of comprehension. Assuming that developing such theories is the central task for psycholinguists in the future, the question is what it takes to get there.

WHAT IT TAKES TO GET THERE: A PARADIGM SHIFT? Is a paradigm shift required, or perhaps already in progress? If a paradigm shift involves a new set of questions or goals, an overturning of old assumptions, often coupled with novel techniques or methods for answering those questions, some might argue that a paradigm shift is already underway. On this (possibly straw-man) view, psycholinguistics is or was characterized by theories couched in a symbolic representational system, implemented in an intrinsically serial (Von Neumann) architecture where sentence processing is in some form an instantiation of the grammar or of the representations it defines. But psycholinguistic theories are or will be cast in terms of inherently parallel connectionist networks that are not (necessarily) symbolic. Processing is viewed as implementing contextual probabilities through modulation of activation spreading through the network (MacDonald, 1994). In the area of sentence processing, arguments taken to support this general sort of "constraint-satisfaction" model derive from effects of the frequency of alternatives on processing time and effects of "contextual constraints" identified by means of off-line normative rating of the experimental materials for an on-line experiment. Until it is known precisely what frequencies matter in sentence processing (what's being counted and why), the existence of frequency effects does not necessarily favor a constraint-satisfaction approach. The existence of lexical frequency effects, for example, is equally compatible with constraint-satisfaction theories of sentence processing and other theories. Contextual constraints stemming from response (off-line) response (on-line) correlations are also neutral between types of theories. Subjects in an off-line rating study do not necessarily employ the competing activation levels of stored representations to generate those ratings. For proponents of constraint-satisfaction models to assume that they do is to assume their conclusions. Further, an explanatory theory of human sentence processing should not need to use human sentence processing data from a par-

Getting There (Slowly)

139

ticular sentence to generate its predictions about the processing of that sentence. Another line of evidence taken to support the activation component of constraint-satisfaction models comes from studies of individual memory span, assessed by a comprehension task (Daneman & Carpenter, 1980). Reading memory span correlates with reading time and perhaps also influences what structures can be built during sentence comprehension (Just & Carpenter, 1992; King & Just, 1991; MacDonald, Just, & Carpenter, 1992; Pearlmutter & MacDonald, 1995). Recently, however, the idea that memory capacity influences the nature of the processor's use of constraints and its ability to cope with difficult (garden path) structures has been challenged strongly (e.g., Caplan & Waters, 1991; Waters & Caplan, 1996). More generally, though constraint-satisfaction models admit a role for grammatical constraints on a par with contextual probabilities, concrete proposals about how to deal with productive syntax have not been forthcoming. To their credit, MacDonald et al. (1994) did make some specific suggestions for how to compute local phrase structure. The proposal is seriously flawed. It involves storing in the lexicon an X template [as in (la)] with every potential head of phrase. Earlier in this paper arguments have been presented against precisely this approach. The approach also presupposes that the processor has some means of identifying the head of a phrase though no such mechanism is offered (see Frazier, 1995a, for discussion). Another problem with the proposed template approach is that it utterly fails in head-final languages, given that syntactic analysis is not delayed until the head is encountered. There are reasons to doubt that pure connectionist models will fare better in the future. The formal properties of connectionist systems may be inherently incapable of handling natural language. Like other feature conjunction systems, connectionist models cannot cope with restricted universal quantification (Marcus, 1998), though clearly human perceivers can. Also, connectionist models are intrinsically parallel. At the very core of connectionist networks is the notion of competition between alternatives. But, if alternatives are not prestored components of the model, it is quite unclear how the competition mechanism works. How does the processor arrive at the alternatives when they are not prestored? To make matters worse, the limited empirical evidence available disconfirms the notion that competition is at work in choosing an interpretation for a sentence. Fully ambiguous sentences promote maximal competition and therefore should be processed more slowly than biased or unambiguous sentences. But they aren't (see Frazier & Clifton, 1997, Clifton, in press). To live up to the goals established in the 1960s, theories of sentence processing must go beyond "transitional probabilities" or "contextual-con-

140

Frazier

straints" (you choose the name). The pioneers in the early days of modern psycholinguistics clearly recognized and demonstrated the need for a theory of sentence processing capable of dealing with productive syntax, including discontinuous and long-distance dependencies, a theory capable of dealing with the full syntactic and expressive range of natural languages. And linguists, especially Chomsky (1957), emphasized the need for explicit, predictive, and explanatory theories of language. Slowly psycholinguistic theories are approaching these goals with theoretically and empirically more adequate accounts of the human sentence processing ability. But simple feature-conjunction systems (simultaneous activation of features or nodes) won't get us there. REFERENCES Abney, S. (1989). A computational model of human parsing. Journal of Psycholinguistic Research, 18, 129-144. Bader, M. (in press). On reanalysis: Evidence from German. In B. Hemforth and L. Konieczny (eds) Cognitive parsing in german: An introduction. Dordrecht: Kluwer Academic Publishers. Bader, M., & Lasser, I. (1994). German verb-final clauses and sentence processing: Evidence for immediate attachment. In C. Clifton, Jr., L. Frazier, & K. Rayner (Eds.), Perspectives on sentence processing. Hillsdale, NJ: Erlbaum. Bever, T. G. (1970). The cognitive basis for linguistic structures. In J. R. Hayes (Ed.), Cognition and the development of language. New York: Wiley. Bever, T. G. (1973). Serial position and response biases do not account for the effect of syntactic structure on the location of brief noises during sentences. Journal of Psycholinguistic Research, 2, 287-288. Bever, T. G., Garrett, M. F., & Hurtig, R. (1973). Ambiguity increases complexity of perceptually incomplete clauses. Memory & Cognition, 1, 279-286. Bever, T. G., Lackner, J. R., & Kirk, R. (1969). The underlying structures of sentences are the primary units of immediate speech processing. Perception and Psychophysics, 5, 225-231. Bever, T. G., & Langendoen, D. T. (1971). A dynamic model of the evolution of language. Linguistics Inquiry, 2, 433-463. Bever, T. G., & McElree, B. (1988). Empty categories access their antecedents during comprehension. Linguistics Inquiry, 19, 35-43. Broderick, K. (1996). Focus and adjunct extraction. Poster presented at the CUNY Sentence Processing Conference, New York. March, 1996. Cairns, H. S. (1970). Ambiguous sentence processing. Unpublished doctoral dissertation, The University of Texas at Austin. Caplan, D. (1972). Clause boundaries and recognition latencies for words in sentences. Perception and Psychophysics, 12, 73—76. Caplan, S., & Waters, G. S. (1991). Short-term memory and language comprehension: A critical review of the neuropsychological literature. In G. Vallas & T. Shallice (Eds.), Neuropsychological impairments of short-term memory. Cambridge, England: Cambridge University Press.

Getting There (Slowly)

141

Carlson, G., & Tanenhaus, M. (1988). Thematic roles and language comprehension. In W. Wilkins (Ed.), Syntax and semantics. Vol. 21: Thematic relations. London, England: Academic Press. Carreiras, M. (1992). Estrategias de analisis en el procesamiento de frases: Cierre temprano versus cierre tardio. Cognitiva, 4, 3-27. Carreiras, M., & Clifton, C, Jr. (1993). Relative clause interpretation preferences in Spanish and English. Language and Speech, 36, 353-372. Chomsky, N. (1957). Syntactic structures. The Hague, The Netherlands: Mouton. Cinque, G. (1992). Types of A' dependencies. Cambridge, MA: MIT Press. Clifton, C. (in press). Evaluating models of human sentence processing. In M. Crocker, M. Pickering and C. Clifton (Eds.), Architectures and mechanisms for language processing. Cambridge: Cambridge University Press. Clifton, C., Speer, S. & Abney, S. (1991). Parsing arguments: Phrase structure and argument structure as determinants of initial parsing decisions. Journal of Memory and Language, 30, 251—271. Crain, S., & Steedman, M. (1985). On not being led up the garden path: The use of context by the psychological parser. In D. Dowty, L. Karttunen, & A. Zwicky (Eds.), Natural language parsing. Cambridge, England: Cambridge University Press. Crocker, M. (1992). A logical model of competence and performance in the human sentence processor. Doctoral dissertation, University of Edinburgh. (Available as Human Communications Research Centre. Publication RP34.) Cuetos, F., & Mitchell, D. C. (1988). Cross-linguistic differences in parsing: Restrictions on the use of the late closure strategy in Spanish. Cognition, 30, 73-105. Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behavior, 19, 450—466. De Vincenzi, M. (1991). Syntactic parsing strategies in Italian. Dordrecht, The Netherlands: Kluwer. De Vincenzi, M. (1996). Syntactic analysis in sentence Comprehension: Effects of dependency types and grammatical constraints. In J. L. Nicol (special guest editor) Journal of Psycholinguistic Research (Special Issue on Sentence Processing), 25, 117-134. Eberhard, K.., Spivey-Knowlton, M., Sedivy, J., & Tanenhaus, M. (1995). Eye movements as a window into real time spoken language comprehension in natural contexts. Journal of Psycholinguistic Research, 24, 409-436. Fiengo, R. (1977). On trace theory. Linguistic Inquiry, 8, 35-62. Fodor, J. A. (1983). Modularity of mind. Cambridge, MA: MIT Press. Fodor, J. A., & Bever, T. G. (1965). The psychological reality of linguistic segments. Journal of Verbal Learning and Verbal Behavior, 4, 414—420. Fodor, J. A., Bever, T. G., & Garrett, M. F. (1974). The psychology of language: An introduction to psycholinguistics and generative grammar. New York: McGrawHill. Fodor, J. D. (1989). Empty categories in language processing. Language and Cognitive Processes, 4, SI155-210. Fodor, J. D., & Ferrieira, F. (Eds.), (in press). Reanalysis in sentence processing. Dordrecht, The Netherlands: Kluwer. Fodor, J. D., & Inoue, A. (1994). The diagnosis and cure of garden paths. Journal of Psycholinguistic Research, 23, 407—434. Forster, K. (1966). Left to right processes in the construction of sentences. Journal of Verbal Learning and Verbal Behavior, 5, 285-291.

142

Frazier

Forster, K. (1968). Sentence completion in left- and right-branching languages. Journal of Verbal Learning and Verbal Behavior, 7, 296-299. Forster, K. (1970). Visual perception of rapidly presented word sequences of varying complexity. Perception and Psychophysica, 8, 215-221. Forster, K. (1979). Levels of processing and the structure of the language processor. In W. E. Cooper & E. C. T. Walker (Eds.), Sentence processing. Hillsdale, NJ. Erlbaum. Forster, K. I., & Ryder, L. A. (1971). Perceiving the structure and meaning of sentences. Journal of Verbal Learning and Verbal Behavior, JO, 285-296. Foss, D. J. (1970). Some effects of ambiguity upon sentence comprehension. Journal of Verbal Learning and Verbal Behavior, 9, 699-706. Frazier, L. (1978). On comprehending sentences: Syntactic parsing strategies. Unpublished doctoral dissertation, University of Connecticut. Frazier, L. (1987a). Sentence processing: A tutorial review. In M. Coltheart (Ed.), Attention and performance XII. Hillsdale, NJ: Erlbaum. Frazier, L. (1987b). Syntactic processing: Evidence from Dutch. Natural Language and Linguistic Theory, 5, 519—560. Frazier, L. (1990). Exploring the architecture of the language system. In G. T. M. Altmann (Ed.), Cognitive models of speech processing: Psycholinguistic and computational perspectives. Cambridge, MA: MIT Press. Frazier, L. (1995a). Constraint satisfaction as a theory of sentence processing. In J. L. Nicol (special guest editor), Journal of Psycholinguistic Research (Special Issue on Sentence Processing), 24, 437-468. Frazier, L. (1995b). Representational issues in psycholinguistics. In J. L. Miller & P. D. Eimas (Eds.), Handbook of perception and cognition (vol. 11). San Diego: Academic Press. Frazier, L. (in press). On sentence interpretation. Dordrecht: Kluwer Academic Publishers. Frazier, L., & Clifton, C. (1989). Successive cyclicity in the grammars and the parser. Language and Cognitive Processes, 4, 93—126. Frazier, L., & Clifton, C. (1996). Construal. Cambridge, MA: MIT Press. Frazier, L., & Clifton, C. (in press). Sentence reanalysis, and visibility. In J. D. Fodor & F. Ferreira (Eds.), Reanalysis in sentence processing. Dordrecht, The Netherlands: (in press) Kluwer. Frazier, L., & Clifton, C. (1997). Construal: Overview, motivation and some new evidence. Journal of Psycholinguistic Research. Frazier, L., & Flores d'Arcais, I. (1989). Filler-driven parsing: A study of gap-filling in Dutch. Journal of Memory and Language, 28, 331-344. Frazier, L., & Fodor, J. D. (1978). The sausage machine: A new two-stage parsing model. Cognition, 6, 291-326. Frazier, L., & Rayner, K. (1982). Making and correcting errors during sentence comprehension: Eye movements in the analysis of structurally ambiguous sentences. Cognitive Psychology, 14, 178-210. Frazier, L., & Rayner, K. (1990). Taking on semantic commitments: Processing multiple meanings vs. multiple senses. Journal of Memory and Languages, 29, 181-200. Friederici, A. D., & Mecklinger, A. (1996). Syntactic parsing as revealed by brain responses: First-pass and second-pass parsing processes. In J. L. Nicol (special guest editor), Journal of Psycholinguistic Research (Special Issue on Sentence Processing), 25, 157-176.

Getting There (Slowly)

143

Gibson, E. A. F. (1991). A computational theory of human linguistic processing: Memory limitations and processing breakdown. Unpublished doctoral dissertation, Carnegie Mellon University. Gibson, E. A. F., & Hickok, G. (1993). Sentence processing with empty categories. Language and Cognitive Processes, 8, 147—162. Gibson, E. A. F., Pearlmutter, N. J., Canseco-Gonzalez, E., & Hickok, G. (1996). Recency preference in the human sentence processing mechanism. Cognition, 59, 2359. Gibson, E. A. F., Schutze, C. T. & Salomon, A. (1996). The relationship between the frequency and the processing complexity of linguistic structure. In J. L. Nicol (special guest editor), Journal of Psycholinguistic Research (Special Issue on Sentence Processing), 25, 25-58. Gilboy, E., Sopena, J. M., Clifton, C., Jr., & Frazier, L. (1995). Argument structure and association preferences in Spanish and English compound NPs. Cognition, 54, 131 — 167. Gorrell, P. (1995). Syntax and parsing Cambridge, England: Cambridge University Press. Gorrell, P. (1996). Parsing theory and phrase-order variation in German V2 clauses. In J. L. Nicol (special guest editor), Journal of Psycholinguistic Research (Special Issue on Sentence Processing), 25, 135-154. Grimshaw, J. (1991). Extended projections. Unpublished manuscript, Brandeis University. Hankamer, J., & Sag, I. A. (1976). Deep and surface anaphora. Linguistics Inquiry, 7, 391-428. Hemforth, B., Konieczny, L., Scheepers, C. & Strube, G. (1993). First analysis, reanalysis and repair. Freiburg, Germany: Institut fur Informatik und Gesellschaft. Hickok, G., & Avrutin, S. (1996). Comprehension of wh-questions in two Broca's aphasics. Brain and Language, 52, 314-327. Horton, W. S., & Keysar, B. (1996). When do speakers take into account common ground? Cognition, 59, 91-117. Inoue, A., & Fodor, J. D. (1995). Information-paced parsing of Japanese. In R. Mazuka & N. Nagai (Eds.), Japanese sentence processing. Hillsdale, NJ: Erlbaum. loup, G. (1975). Some universals for quantifier scope. In J. Kimball (Ed.), Syntax and Semantics (Vol. 4). New York: Academic Press. Johnson, N. F. (1965). The psychological reality of phrase structure rules. Journal of Verbal Learning and Verbal Behavior, 4, 469—475. Just, M. A., & Carpenter, P. A. (1992). A capacity theory of comprehension: Individual differences in working memory. Psychological Review, 99, 122-149. Keysar, B. (in press). Language users as problem solvers: Just what ambiguity do they solve? In S. R. Fussell & R. J. Kreuz (Eds.), Social and cognitive psychological approaches to interpersonal communication. Hillsdale, NJ: Erlbaum. Kimball, J. (1973). Seven principles of surface structure parsing in natural language. Cognition, 2, 15–47. King, J., & Just, M. A. (1991). Individual differences in syntactic processing: The role of working memory. Journal of Memory and Language, 30, 580—602. Koh, S. (1997). The resolution of the dative ambiguity in Korean. Journal of Psycholinguistic Research, 26, 265—273. Konieczny, L., Hemforth, B., Scheepers, C., & Strube, G. (1997). The role of lexical heads in parsing. Evidence from German. Language & Cognitive Processes. 12, 307-348.

144

Frazier

Kurtzman, H., & MacDonald, M. E. (1993). Resolution of quantifier scope and ambiguities. Cognition, 48, 243-279. Lackner, J. R., & Garrett, M. F. (1973). Resolving ambiguity: Effects of biasing context in the unattended ear. Cognition, 2, 359-372. Ladefoged, P., & Broadbent, D. E. (1960). Perception of sequence in auditory events. Quarterly Journal of Experimental Psychology, 13, 162-170. Lehiste, I. (1972). Timing of utterances and linguistic boundaries. Journal of the Acoustical Society of America, 51, 2018-2024. Lewis, R. L. (1996). Interference in short-term memory: The magical number two (or three) in sentence processing. In J. L. Nicol (special guest editor), Journal of Psycholinguistic Research (Special Issue on Sentence Processing), 25, 93—116. Love, T., & Swinney, D. (1996). Coreference processing and levels of analysis in objectrelative constructions: Demonstration of antecedent reactivation with the cross-modal priming paradigm. In J. L. Nicol (special guest editor), Journal of Psycholinguistic Research (Special Issue on Sentence Processing), 25, 5-24. MacDonald, M. C. (1994). Probabilistic constraints and syntactic ambiguity resolution. Language and Cognitive Processes, 9, 157-202. MacDonald, M. C., Just, M., & Carpenter, P. A. (1992). Working memory constraints on the processing of syntactic ambiguity. Cognitive Psychology, 24, 56-98. MacDonald, M. C., Pearlmutter, N. J., & Seidenberg, M. S. (1994). The lexical nature of syntactic ambiguity resolution. Psychological Review, 101, 676-703. Marcus, G. (1998). Cognition, variables and connectionism. Unpublished manuscript, University of Massachusetts. Marslen-Wilson, W. D. (1973). Linguistic structure and speech shadowing at very short latencies. Nature, 244, 522-523. Marslen-Wilson, W. D., & Tyler, L. K. (1980). The temporal structure of spoken language comprehension. Cognition, 8, 1-72. Marslen-Wilson, W. D., & Tyler, L. K. (1987). Against modularity. In J. L. Garfield (Ed.), Modularity in knowledge representation and natural-language understanding. Cambridge, MA: MIT Press. Mecklinger, A., Schriefers, H., Steinhaus, K., & Friederici, A. (1995). The processing of relative clauses varying on syntactic and semantic dimensions: An analysis with event-related potentials. Memory & Cognition, 23, 477-494. Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63, 81-96. Miller, G. A., & Chomsky, N. (1963). Finitary models of language users. In R. D. Luce, R. R. Bush, & E. Galanter (Eds.), Handbook of mathematical psychology (Vol. II). New York: Wiley. Miller, G. A., & Isard, S. (1963). Some perceptual consequences of linguistic rules. Journal of Verbal Learning and Verbal Behavior, 2, 217—228. Mitchell, D. C., Cuetos, F., Corley, M. M. B., & Brysbaert, M. (1996). Exposure-based models of human parsing: Evidence for the use of coarse-grained (nonlexical) statistical records. In J. L. Nicol (special guest editor), Journal of Psycholinguistic Research (Special Issue on Sentence Processing), 24, 469-488. Murphy, G. (1990). Noun phase interpretation and conceptual combination. Journal of Memory and Language, 29, 259-288. Nagai, N. (1995). Constraints on topics and their gaps: From a parsing perspective. In R. Mazuka & N. Nagai (Eds.), Japanese sentence processing. Hillsdale, NJ. Erlbaum.

Getting There (Slowly)

145

Nagel, H. N., Shapiro, L. P., & Nawy, R. (1994). Prosody and the processing of fillergap sentences. Journal of Psycholinguistic Research, 23, 473—486. Neville, H., Nicol, J., Barss, A., Forster, K., & Garrett, M. (1991). Syntactically based sentence processing classes: Evidence from event-related brain potentials. Journal of Cognitive Neuroscience, 3, 155—170. Ni, W., Braze, D., Conway, L., Crain, S., & Shankweiler, D. (1996). A fork in the garden path. Poster presented at the Ninth Annual CUNY Sentence Processing Conference, New York. Nicol, J. (guest editor). (1996). Journal of Psycholinguistic Research (Special Issue on Prosody), 25. Onishi, K. H., & Murphy, G. L. (1993). Metaphoric reference: When metaphors are not understood as easily as literal expressions. Memory & Cognition, 21, 763—772. Pearlmutter, N. J., & MacDonald, M. C. (1995). Individual differences and probabilistic constraints in syntactic ambiguity resolution. Journal of Memory and Language, 34, 521-542. Pickering, M., & Barry, G. (1991). Sentence processing without empty categories. Language and Cognitive Processes, 6, 229-259. Pierrehumbert, J. B. (1980). The phonology and phonetics of English intonation. Unpublished doctoral dissertation, MIT. Pierrehumbert, J. B., & Beckman, M. E. (1988). Japanese tone structure. Cambridge, MA: MIT Press. Potter, M. C., & Faulconer, B. A. (1979). Understanding noun phases. Journal of Verbal and Learning and Verbal Behavior, 18, 509—522. Pritchett, B., & Whitman, J. (1995). Syntactic representation and interpretive preference. In R. Mazuka & N. Nagai (Eds.), Japanese sentence processing. Hillsdale, NJ: Erlbaum. Pritchett, B. L. (1991). Head position and parsing ambiguity. Journal of Psycholinguistic Research, 20, 251-270. Pritchett, B. L. (1992). Grammatical competence and parsing performance Chicago: University of Chicago Press. Rado, J. (1994, March). Processing moved antecedents. Poster presented at the Seventh Annual CUNY Conference on Human Sentence Processing, New York. Rado, J. (1998). Processing Hungarian: The role of topic and focus in language comprehension. Unpublished doctoral dissertation, University of Massachusetts. Rizzi, L. (1982). Issues in Italian syntax. Dordrecht, The Netherlands: Foris. Rooth, M. (1992). Ellipsis redundancy and reduction reduncancy. Unpublished manuscript, University of Stuttgart, Stuttgart, Germany. Sanford, A. J., Moxey, L. M., & Patterson, K. (1994). Psychological studies of quantifiers. Journal of Semantics, 10, 153—170. Schafer, A. (1998). Prosodic parsing: The role of prosody in sentence comprehension. Unpublished doctoral dissertation, University of Massachusetts. Schriefers, H., Friederici, A., & Kuhn, K. (1995). The processing of locally ambiguous relative clauses in German. Journal of Memory and Language, 34, 499-520. Schutze, C., & Gibson, E. (1996). Argument preference in English PP attachment. Paper presented at a meeting of the Linguistic Society of America, San Diego. Sekerina, I. (1997) The syntax and processing of scrambling constructions in Russian. The City University of New York doctoral dissertation. Selkirk, E. O. (1984). Phonology and syntax: The relation between sound and structure. Cambridge, MA: MIT Press.

146

Frazier

Selkirk, E. (1995). Sentence prosody: Intonation, stress and phrasing. In J. Goldsmith (Ed.), Handbook of phonological theory. Oxford, England: Blackwell. Spivey-Knowlton, M., & Sedivy, J. (1995). Resolving attachment ambiguities with multiple constraints. Cognition, 55, 227-267. Stabler, E. P. (1994). The finite connectivity of linguistic structure. Unpublished manuscript, University of California, Los Angeles. Stevenson, S. (1994). Competition and recency in a hybrid network model of syntactic disambiguation. Journal of Psycholinguistic Research, 23, 295-322. Stewart, C., & Gough, P. (1967). Constituent search in immediate memory for sentences. Proceedings of the Midwestern Psychology Association. Stowe, L. (1986). Parsing wh-constructions. Language and Cognitive Processes, 2, 227246. Sturt, P., & Crocker, M. (1995, March). Monotonic parsing and reanalysis. Paper presented at the Eighth Annual CUNY Conference on Human Sentence Processing, Tucson, AZ. Tanenhaus, M., & Carlson, G. (1990). Comprehension of deep and surface verb phrase anaphors. Language and Cognitive Processes, 5, 257-280. Traxler, M. J., & Pickering, M. J. (1996). Plausibility and the processing of unbounded dependencies: An eye-tracking study. Journal of Memory and Language, 35, 454475. Tunstall, S. (1998). The interpretation of quantifiers and indefinites. Unpublished doctoral dissertation, University of Massachusetts. Van Lehn, K. A. (1978). Determining the scope of English quantifiers. Unpublished master's thesis, MIT. von Fintel, Kai. (1994). Restrictions on quantifier domains. Unpublished doctoral dissertation, University of Massachusetts. Wanner, E. (1968). On remembering, forgetting and understanding sentences: A study of the deep structure hypothesis. Unpublished doctoral dissertation, Harvard University. Wanner, E., & Maratsos, M. (1971). On understanding relative clauses. Unpublished paper, Harvard University. Wanner, E., & Maratsos, M. (1978). An ATN approach to comprehension. In M. Halle, J. Bresnan, & G. A. Miller (Eds.), Linguistic theory and psychological reality. Cambridge, MA: MIT Press. Ward, G., Sproat, R., & McKoon, G. (1991). A pragmatic analysis of so-called anaphoric islands. Language, 67, 439-473. Warren, D. (Guest editor). (1996). Language and Cognitive Processes (Special Issue on Prosody and Parsing. Volume II (v2)). Waters, G. S., & Caplan, D. (1996). Processing resource capacity and the comprehension of garden path sentences. Memory & Cognition, 24, 342-356. Yngve, V. (1960). A model and an hypothesis for language structure. Proceedings of the American Philosophical Society, 104, 444-466.