THE UNACCUSATIVE HYPOTHESIS

According to the unaccusative hypothesis (Perlmutter, 1978), intransitive verbs are broadly classified into unergatives and unaccusatives, depending on the semantic characteristics of the argument. For some linguists, the difference between the two classes of verbs is purely semantic (Dowty, 1991; Van Valin, 1990) and reduces to the notions of agentivity and telicity. Telicity (from Greek telos “limit, end, goal”) is the property that makes situations telic (with an inherent endpoint) or atelic (with no inherent endpoint). Thus, unergative verbs are usually agentive and denote a process (atelic; run, walk, work), whereas unaccustive verbs (die, disappear, exist) are nonagentive and telic, typically denoting a change of some sort. For other linguists (Burzio, 1986; Rosen, 1984), however, the distinction between the two classes of verbs is largely syntactic.¹

These days, traces no longer have theoretical status in generative grammar. In current minimalist terms (Chomsky, 1995), movement of constituents is performed through an “attract” operation between two elements (a target and a goal), forming a chain between them (the target for movement and the goal). A “trace” is simply an unpronounced copy of the moved element (the antecedent of the trace).

In English, the resultative construction (a small clause describing the effect on the object of the event denoted by the verb) has been used as a syntactic diagnosis for unaccusativity (Carrier & Randall, 1992; Levin & Rappaport Hovav, 1995; but cf. Levin & Rappaport Hovav, 2002). Because resultatives can only be predicated of the object, they are possible with transitive (Example 3) and unaccusative (Example 4) verbs (which have an object), but impossible with unergatives, whose sole argument is a subject (Example 5). Note that the sentence in Example 6 with an unergative verb is grammatical when a fake reflexive, which acts syntactically as an object, is added.

For Levin and Rappaport Hovav (1995), unaccusativity is semantically determined but syntactically represented. The linking rules that regulate the projection of arguments to syntactic positions are sensitive to semantic notions like change of state, immediate cause, and so forth. In English, an argument denoting a directed change is an internal argument, whereas one denoting an immediate cause is an external argument.

If the distinction between unaccusative and unergative verbs is purely structural, then we expect all unaccusative and unergative verbs to behave alike syntactically. However, Sorace's (2000) research has uncovered that, although many verbs display consistent syntactic unaccusative or unergative behavior across and within languages, others display variable syntactic behavior, depending on the contribution of other aspectual or thematic elements in the sentences in which they appear. For example, change of state verbs like decay, agentive verbs of directed motion like run and march, verbs of existence like exist, and so forth systematically display variable syntactic behavior with auxiliary selection and are sensitive to aspectual and thematic elements in the sentences in which they appear. The verb run in Italian (and many other languages) selects the perfect auxiliary “have” and displays unergative behavior when it is atelic (“Maria ha corso velocemente.” “Mary has run fast.”), but it selects “be” and displays unaccusative behavior when a goal prepositional phrase is added, making the event telic: “Maria è corso in farmacia.” “Mary ran to the pharmacy.” By contrast, talk and sing always select the perfective auxiliary “have” in languages that have the auxiliary distinction, and the telicization of the predicate with another element in the sentence does not select “be” (“I colleghi hanno chiaccherato tutto il pomeriggio.” “My colleagues chatted the whole afternoon.”). Similarly, the unaccusatives come and arrive always select “be” as perfective auxiliary and agentivity or detelicization of the event does not affect the choice of auxiliary (“Sono arrivati ospiti per ore e ore.” “Guests arrived for hours.”). Sorace (2000) proposed that there is a continuum (what she calls a “hierarchy”) of gradients of unaccusative/unergative verbs of potential universal significance, as in Example 7. This hierarchy is related to the semantic concepts of telicity and agentivity. The extremes of the continuum (which she calls “core”) are the prototypical unaccusative and unergative verbs. The verbs in the middle are more or less unaccusatives or unergatives, depending on their position in the continuum with respect to the extremes.

Sorace (2000) notes that the potential reality of this hierarchy is confirmed in a series of psycholinguistic studies (Bard, Robertson, & Sorace 1996; Sorace, 1995). Findings from L1 acquisition have shown that, although auxiliary selection with intransitive verbs is learned quite early, it is learned earlier with verbs displaying stable behavior than with those that shift from unaccusative to unergative. Furthermore, experiments with native and near-native speakers of Italian have shown that native speakers have very strong intuitions concerning auxiliary selection with the “stable” verbs and less determinate intuitions about syntactic behavior with the others, attesting to these gradients of unaccusativity.

To summarize thus far, the unaccusative/unergative distinction is a universal phenomenon but languages implement different syntactic and morphological reflexes of unaccusativity, which have to be learned implicitly in L1 or L2 acquisition from exposure to primary linguistic data (input). Debate exists as to whether the distinction could be better characterized as a purely semantic phenomenon (Dowty, 1991; Van Valin, 1990), a purely syntactic phenomenon (Burzio, 1986; Rosen, 1984), or both (Levin & Rappaport Hovav, 1995; Sorace, 2000). Whereas the variable syntactic behavior of these verbs indicates that there is a strong correlation between semantic notions and unaccusativity, it is still far from clear why particular semantic components are sensitive to certain syntactic reflexes when others are not, or why languages are sensitive to different semantic notions for the classification of intransitive verbs.²

SYNTACTIC TESTS FOR UNACCUSATIVITY IN SPANISH

Unlike the situation in Italian and French, present-day Spanish no longer has two perfect auxiliaries to distinguish between the two classes of intransitive verbs. However, other syntactic and morphological tests of unaccusativity in Spanish have been recently identified in the literature. One such test is the distribution of bare plurals (NPs without determiners; Demonte, 1985; Torrego, 1989). Transitive (Example 8) and unaccusative (Example 9) verbs allow bare plurals in postverbal position; unergatives do not, as Example 10 shows (examples from Torrego, 1989, p. 254).³

A second syntactic test discriminating between unaccusative and unergative verbs in Spanish is the absolutive construction (de Miguel, 1992). Absolutive constructions are clauses with a past participle that modify a postposed noun phrase and agree with it in number and gender. Absolutive constructions are grammatical with transitive (Example 11) or unaccusative (Example 12) participles, that is, with structures that have an object but are clearly ungrammatical if the participle is an unergative verb (Example 13), which presumably does not have an object.

Furthermore, de Miguel (1992) notes that among unaccusative verbs, this construction is sensitive to the aspectual notion of telicity. In other words, the participial absolutive construction is grammatical with telic unaccusative verbs (those that have an end point), as in Example 12, but not with atelic unaccusative verbs, such as verbs of existence, durative verbs, and iterative verbs (e.g., some stative and achievement verbs in Vendler's, 1967, classification), as in Examples 14 and 15.

Another difference between unaccusative and unergative verbs in Spanish concerns the position of the subject with presentational verbs (Contreras, 1978; Suñer, 1982), although this is an optional rule (Demonte, 1985). Stylistically, subjects of unaccusative verbs show a marked tendency to appear in object position (postposed), as in Example 16.

By contrast, if postposed, subjects of unergative verbs sound odd, unless there is special emphasis in the discourse context.⁴

PSYCHOLINGUISTIC EVIDENCE

Assuming that there is a syntactic basis for the unaccusative/unergative distinction in Spanish, Bever and Sanz (1997) conducted an experiment with 32 Spanish native speakers to assess the role of empty categories in sentence comprehension. Recall that in terms of linguistic representations, the arguments of unaccusative verbs leave an empty category—an NP-trace—in object position. Unergative verbs do not have a trace. To examine the differential effect of the NP-trace left by the movement of the argument during on-line processing, Bever and Sanz used a visual probe-recognition task. According to Bever and McElree (1988) and Bever, Straub, Shenkman, Kim, and Carrithers (1989), a trace, or “copy,” provides an extra representation of its antecedent. This is reflected in shorter latencies to recognize a word contained in the antecedent, when compared to other, similar sentences that do not have traces. Thus, Bever and Sanz (1997) hypothesized that sentences with unaccusative verbs and preverbal arguments should display shorter reaction times (RTs) than sentences with unergative verbs because the trace in object position in unaccusative verbs is “more recent” in the string as the string is being processed.

The task consisted of sentences containing 16 frequency-matched unaccusative and unergative verbs presented on a computer screen. The critical verbs were 8 unaccusatives, crecer (grow), aparecer (appear), desaparecer (disappear), aterrizar (land), morir (die), caer (fall), llegar (arrive), and arribar (arrive by boat) and 8 unergatives, caminar (walk), pasear (stroll), viajar (travel), correr (run), dormir (sleep), hablar (talk), cantar (sing), and llorar (cry).

Eight of the experimental sentences had preverbal subjects and the other eight had postverbal subjects, of the structures shown in Example 18.

Bever and Sanz (1997) included both concrete and abstract adjectives as probes. Sentences were presented on a computer screen in four fragments and without punctuation marks, as Examples 19 and 20 show:

After the segment fragments disappeared from the screen, participants were asked to perform two tasks: decide if the probe word (in this case, the adjective appearing between asterisks) occurred in the sentence and answer a question about the content of the relative clause, by pressing one of two keys (yes or no) for each task. Both questions required the experimental subjects to refer back to the subject NP containing the adjective. Question 1 tested the subjects' sensitivity to the sequence of words in the sentence (i.e., syntactic structure). Question 2 tested whether subjects were paying attention to the meaning of the sentence as well. The computer recorded the time that participants took to respond in milliseconds.

In addition to the 16 experimental trials, the task also included a set of 80 filler sentences of varying length (15–58 syllables), some with probes and some without. Bever and Sanz (1997) then used these distracters with negative answers to classify the participants into two groups, based on how the sentence comprehension process works. According to these authors, comprehending a sentence involves understanding the conceptual representation for its meaning and the syntactic representation for its form. In understanding a sentence, subjects can initially scan the conceptual semantic representation, the syntactic representation, or both. Bever and Sanz (1997) claim that there are individual differences in this respect, because people can be conceptual or syntactic scanners. Because there is no linguistic theory in which semantics can be accessed independently of syntax, this assertion does not make much sense. However, Townsend and Bever (2001, p. 206) discuss how this is a possible strategy in on-line sentence comprehension. To examine this distinction, a correlation was taken between the time it took each subject to recognize that the probe was not in the sentence and the length of the filler sentence in syllables. Half of the subjects (n=16) showed a positive correlation between sentence length and RT, whereas the other half (n=16) showed a negative correlation. The first group of subjects was classified as sequence sensitive, meaning that they pay attention to syntactic structure when they access the probe. The subjects for whom the correlation was negative were classified as sequence insensitive, supposedly because these subjects scan the conceptual representation of a sentence to search the probe. Bever and Sanz (1997) further confirmed the distinction between these two groups by examining the RTs in sentences with concrete and abstract adjectives. It is only the syntactic scanners who should show shorter latencies for unaccusative verbs than for unergative verbs. The results from Bever and Sanz (1997) are displayed in Table 1.

As can be seen from Table 1, sequence-sensitive individuals reacted faster to unaccusative verbs with preverbal subjects (950 ms) than to unergative verbs (1070 ms). The 120-ms difference was significant according to analysis of variance (ANOVA). The sequence-insensitive individuals showed an effect in the opposite direction, responding faster to unergative verbs (960 ms) than to unaccusative verbs (1070 ms). There were no differences between the two types of verbs with postverbal subjects, regardless of type of comprehender. According to Bever and Sanz (1997), the data from the sequence-sensitive participants confirm that there is a trace in the syntactic representation of sentences with unaccusative verbs and preverbal subjects in Spanish. Just like Felser, Nakano, and Clahsen (in press), who used a cross-modal lexical decision experiment with native speakers of Japanese and confirmed the psychological reality of syntactic wh- traces left by movement, Bever and Sanz (1997) results also suggest that the unaccusative/unergative distinction in Spanish is syntactic and psycholinguistically real. This raises the question of whether L2 learners of Spanish attain this type of linguistic knowledge as well and whether it is reflected in an on-line task.

Before turning to L2 acquisition, I would like to mention some shortcomings of the Bever and Sanz (1997) experiment. The first question that arises is the division of subjects by type of comprehender. Unlike individual differences related to working-memory capacity for which there is substantial independent evidence, the distinction between sequence-sensitive and sequence-insensitive comprehenders appears to be post hoc and a way to explain the results. Second, if indeed the sequence-insensitive individuals do not scan the syntactic representation of sentences, then no differences between unaccusatives and unergatives would be expected. Why then did unergative verbs induce shorter reaction times than unaccusatives for these individuals?

Sanz (2000) defends the sequence-sensitive or sequence-insensitive distinction by alluding to the correlations between length and RTs, as well as to the differential responses to the concrete/abstract adjective in the Bever and Sanz (1997) experiment. The correlation between sentence length and reaction time for the sequence-sensitive individuals was r²=.22 (p<.01) and r²=−.10 (ns) for the sequence-insensitive group. Even if significant, the correlation for the sequence-sensitive individuals is very weak and could perhaps be explained by the great variability in individual means, which Sanz (2000, p. 172) herself acknowledges. Moreover, and as an anonymous referee also correctly noted, the use of concrete and abstract adjectives is not well motivated and may even seem spurious. For example, the adjective apuesto (handsome) is presented as a concrete adjective but severo (strict) is classified as an abstract adjective, when in fact these are very impressionistic, relative, or subjective qualities. In short, both the correlations and the responses on adjectives are behavioral evidence consistent with the type of comprehender classification proposed in this particular study and they constitute evidence internal and dependent on this study.

As for the faster responses to unergatives by sequence-insensitive individuals, Sanz (2000, pp. 158–159) citing Bever (1970) explains that in English the first noun of a sentence is usually an agent. However, if speakers assign the thematic role of agent to an unaccusative verb, a reassignment of theta role (from agent to theme) must occur, as it happens in the garden path sentence the horse raced the barn fell, where horse is initially assigned the role of agent and then must be reassigned the role of theme, causing a processing delay. Thus, sequence-sensitive individuals read off the conceptual representation of the sentence and perform a thematic role reassignment. It is this reassignment that causes the delayed response with unaccusative verbs as opposed to unergative verbs. However, this explanation might work for English, where word order is more fixed and subjects are usually agents. In contrast, in Spanish word order is far more flexible, and it is very common for objects or themes to be in preverbal position (see MacWhinney, 1987 and VanPatten, 1996). Because the experiment reported here is on Spanish and unless it can be shown that this “agent first” theta assignment strategy is also frequent in this language, it is dubious whether this explanation could be extended in this case.

Despite these shortcomings, the Bever and Sanz (1997) study is original and important in several ways and merits replication. First, it is the first study on unaccusativity in Spanish with native speakers. Second, the study attempts to provide psycholinguistic evidence for a well-motivated linguistic universal, for which substantial independent evidence exists from a variety of languages. Third, and to the extent that the methodology is valid, the study provides evidence for approaches to the nature of linguistic representations that make use of empty categories like traces or copies, thus making a contribution to linguistic theory.

PROCESSING UNACCUSATIVES IN L2 ACQUISITION

The most widely used methodology to investigate subtle linguistic knowledge in L2 acquisition has been grammaticality judgment tasks. These tasks allow researchers to test constructions that are grammatical and ungrammatical, particularly structures that are likely to not occur in spontaneous production very often. However, like any other technique that is used with native and nonnative speakers, these tasks tap linguistic competence only indirectly, given that they are also measures of some sort of performance. Many researchers have argued that grammaticality judgment tasks do not tap linguistic intuitions (Schütze, 1996); rather, they engage the participants' conscious or learned grammatical knowledge and metalinguistic skills (Birdsong, 1989).

It is also believed that language processing (parsing) and grammatical knowledge are related (Fodor, 1999; Klein, 1999). Successful parsing involves segmenting speech into phrases, assigning category labels to words, and determining syntactic dependencies and hierarchical organization. In recent years L2 researchers have begun to expand their methodological repertoire to include processing tasks, which are also assumed to provide an indirect way of tapping grammatical competence, but without the added metalinguistic factor (see Bley–Vroman & Masterson, 1989; Duffield & White, 1999; Duffield, White, Bruhn de Garavito, Montrul, & Prévost, 2002; Eubank, 1993; Juffs 1998a, 1998b; Juffs & Harrington, 1995, 1996; and see particularly Juffs, 2001, and Marinis, 2003, for recent overviews of the existing research and possibilities). However, many times, the relationship between language processing and grammatical knowledge is not always clear. If L2 learners do not parse L2 speech or written input appropriately, it is likely that they cannot develop the same kind of linguistic competence as native speakers (VanPatten, 1996). According to MacWhinney (1987), L2 learners use processing routines from their native language in their L2. Results of relative clause attachment preferences have shown that this is largely the case: L2 learners transfer low or high attachment parsing strategies from their L1 (Dussias, 2003; Felser, Roberts, Gross, & Marinis, 2002; Fernández, 1999). Or, it may be the case that learners have acquired a certain type of linguistic knowledge, but it is labored or delayed access to that knowledge during processing that gives the impression that grammatical knowledge has not been acquired, as Juffs and Harrington (1995) and White and Juffs (1998) have shown with knowledge of subjacency (or constraints on wh- movement) in English by Chinese-speaking learners.

Despite the above findings, other research has shown that, as it is the case with native speakers, on-line processing indirectly reflects mental and linguistic processes in L2 acquisition as well. For example, Duffield and White (1999) and Duffield et al. (2002) found that the results of an on-line sentence-matching task corroborated and complemented the results of an off-line grammaticality judgment task. In both tasks the L2 learners and the native speakers discriminated between grammatical and ungrammatical sentences with differences in clitic placement in Spanish and French. Not only did the sentence-matching task detect differences in grammaticality, but it also provided corroborating evidence for a particular syntactic approach to clitic placement in Romance languages.

To see whether L2 learners of Spanish distinguish syntactically between unaccusative and unergative verbs and whether this difference would be reflected in processing, in this study I extended the methodology used by Bever and Sanz (1997) to another population of native Spanish speakers from different countries and to a group of English-speaking L2 learners of Spanish. By doing a replication with some modifications that I discuss below, the reliability problems found in the original Bever and Sanz study could perhaps be resolved. In this way, it is possible to tease apart, for example, whether some of the problems of the original study have their roots in the particular selection of subjects and/or stimuli employed in the methodology, or even in the way in which the results were analyzed.

Adopting an event structure approach couched in minimalist terms, Sanz (2000) and Sanz and Bever (2001) have proposed that in English the distinction between unaccusative and unergative verbs is merely semantic, and that unlike Spanish unaccusative verbs, English unaccusative verbs do not have traces. More specifically, Sanz proposes that the syntactic difference between unaccusative verbs in Spanish and in English is that in Spanish the argument of unaccusatives is internal to the verb (sister) and moves to a projection Event Phrase to check [±interpretable] telicity features. By contrast, the argument of unaccusative verbs in English originates in a position above the verb (in a functional category called Transitive Phrase, where the features [±measure] are checked) and does not move overtly to Event Phrase. Evidence for this analysis comes from the behavior of the aspectual reflexive se in Spanish, which Sanz (2000) takes as the morphologically overt expression of the head of Event Phrase in this language, and the fact that the few syntactic reflexes identified in English (there insertion, the locative inversion, and the resultative constructions) either do not readily discriminate between unaccusative and unergative verbs or do not work for all unaccusative verbs (see also Levin & Rappaport Hovav, 1995, for a similar discussion). Sanz (2000, p. 140) therefore concludes that unaccusatives and unergative verbs in English vary in interpretation (i.e., unaccusatives tend to be telic but unergatives are atelic) but not in their syntax, whereas in Spanish they vary both in interpretation and syntax. In a similar psycholinguistic test conducted with English native speakers, Sanz (2000) found no trace-facilitation effect with unaccusative verbs.

Assuming for the sake of argument that Sanz's (2000) proposal is correct for both Spanish and English, if English-speaking L2 learners of Spanish do not respond faster to sentences with unaccusative verbs than to sentences with unergative verbs, there may be at least two reasons for this response: either because English-speaking L2 learners of Spanish transfer their parsing strategies from English into Spanish and L2 processing masks L2 linguistic competence, or because adopting their L1 analysis, English-speaking learners of Spanish do not discriminate between unaccusative and unergative verbs syntactically in Spanish either, and processing reveals linguistic competence.⁵

A second aim of this study, which differs from Bever and Sanz (1997), was to investigate whether the semantic hierarchy proposed by Sorace (2000) plays a role in the processing of these verbs. If the unaccusative hierarchy is psycholinguistically real, we expect to find shorter RTs for “core” unaccusatives than for “core” unergatives but variability in RTs with “less core” and “peripheral” unaccusative and unergative verbs for both the native speakers and the L2 learners. According to an anonymous reviewer, this particular hypothesis lacks predictability because it suggests that any result is acceptable with less core verbs. I am very sympathetic with the reviewer's concern about the lack of predictability in general of the concepts of gradience, continua, and prototypicality in grammatical analysis, concepts that are fundamental to functionalist approaches to linguistics. The problem with hierarchies of this sort is that it is difficult to find counterexamples and to make sharp and discrete distinctions among categories of grammatical elements (see Baker, in press, and Newmeyer, 1998, for more extensive arguments along these lines). However, I do not think that Sorace's (2000) account lacks predictability at all. What she says is that peripheral verbs may be susceptible to event-type shifts (sometimes they can be construed as telic or atelic or change from activities, as in “I run for an hour.” to accomplishments, as in “I ran to the store.” or “I ran a mile.”) or they can receive multiple argument realization, depending on other elements in the sentence, like arguments, adverbials, clitics, or prepositional phrases. In turn, this grammatical variability is reflected in how individual speakers may conceptualize these verbs. By contrast, core verbs are not open to such variation, and this is the stronger part of the prediction. Syntactically, this means that sometimes less core and peripheral unaccusatives and unergatives will have a trace and sometimes they will not. Thus, the hypothesis is that with core verbs we should definitely observe the presence of a trace in unaccusative but not in unergatives, but with the more peripheral verbs speakers may vary greatly (see also Sorace, 2000; and Sorace & Shomura, 2001, for elaborations of this hypothesis). Counterevidence for the hierarchy would be if we find that core unergatives are responded to faster than core unaccusatives, we find no differences between these two, or we find this same two patterns of response with all verbs, regardless of their position in the hierarchy.

METHODOLOGY

Thirty-five Spanish native speakers from different countries participated in the study. Twelve of these native speakers were tested in Argentina, and the rest were tested in the United States. Their mean age was 31.03 (range=23–59). The nonnative speakers were 44 English-speaking learners of Spanish who were taking Spanish classes at a major research university in the United States. Their mean age was 23.13 (range=20–35) and their mean age of exposure to Spanish as an L2 was 13.98 (range=11–18).

The main task was an on-line visual probe recognition task, the same used in Bever and Sanz's (1997), with some modifications. In the present study, nine unaccusative verbs and nine unergative verbs were used. These verbs, shown in Table 2, were selected on the basis of semantic subclasses in order to test Sorace's unaccusativity hierarchy (2000).

Because each verb appeared twice, the stimuli consisted of 36 test sentences of the exact form and length used in Bever and Sanz (1997). For the verbs that were also used in Bever and Sanz's study, desaparecer (disappear), morir (die), caer (fall), llegar (arrive), caminar (walk), correr (run), hablar (talk), and cantar (sing), their exact sentences were used. Similar sentences were constructed for the other verbs selected for this study that did not appear in Bever and Sanz's study. However, in this study only concrete adjectives were used as probes; all verbs appeared first with preverbal, and then with postverbal subjects; and in order to be able to perform an analysis by verb, all subjects saw all the verbs. All the distracter sentences used in this task (72 total) were the same sentences used by Bever and Sanz (1997). All the critical sentences were followed by concrete adjective probes and by a content question. The answer to both questions was yes. For the distracter sentences, verbs, nouns, adjectives, and adverbs were used as probes. In 70% of the distracter sentences the probe word was not in the sentence and required a negative response. Only 40% of distracter sentences were followed by a content question. All this was done to divert the participant's attention from the main objective of the experiment. The appearance of each fragment was self-paced (i.e., controlled by the experimental subject), but if no response was made within a 3000-ms span, the fragment disappeared and the next fragment appeared.

The program used to run the experiment was DMDX.⁶

This program was developed and maintained by Kenneth Forster and Jonathan Forster from the University of Arizona and it can be downloaded from www.u.arizona.edu/~kforster/dmastr/dmastr.htm.

The presentation of sentences was pseudorandomly scrambled. That is, the stimuli were prepared in 9 blocks of 12 items each, containing a sentence with an unaccusative verb with preverbal subject, a sentence with an unaccusative verb with postverbal subject, a sentence with an unergative verb with preverbal subject, and eight distracter and filler items. The computer program scrambled both the sequence of 12 items within each block and the 9 blocks. In this way, the items were presented in a different order for each subject, yet the items in each condition were evenly distributed throughout the experiment. This guarded against the possibility of having all items from one condition (say, unaccusatives with preverbal subjects) being randomly clustered together at some point during the experiment. The experiment started with a training session containing 10 trial items. After the training session, each participant was given the opportunity to ask any clarification questions. When the participant felt ready, he or she continued with the experimental session. The session lasted 30 min and was divided into three 10-min parts. Each subject controlled how much time he or she wanted to take for each break and was instructed to press the ENTER key when ready to continue.

OVERALL RESULTS

Only correct responses to experimental trials (probes and content questions) were kept for the statistical analysis (around 84.65% total). The RTs for each subject were then truncated to 2.5 standard deviations based on all the correct recognition of positive probes. Responses over 2000 ms were discarded. Because there was great variability even between the native speakers in RTs (some reacted in as fast as 600 ms, while others averaged 1200 ms), all data were transformed to z scores. Converted scores were then submitted to a factorial ANOVA with verb (unaccusative vs. unergative) and subject position (preverbal vs. postverbal) as the within variables, and group (native speakers vs. L2 learners) as the between variables. There was no main effect for verb, but there was a main effect for subject position, postverbal was responded to faster than preverbal, F (1, 77)=33.006, p<.0001, and for group, native speakers were faster than L2 learners, F (1, 77)=1879.463, p<.0001; as well as a subject position per verb interaction, F (1, 77)= 4.252, p<.0001. The alpha level for significance was set at .05.

Figure 1 presents the results of the native speakers. When the native speaker group was analyzed as a whole, there were no significant differences between unaccusative and unergative verb sentences with preverbal subjects. Following Bever and Sanz's (1997) procedure, the correlation between sentence length with distracter items and reaction times was calculated for each subject. Of the 35 subjects, 16 (46%) showed a weak positive correlation between sentence length and reaction time (r²=.29, p<.01), and the remaining 19 subjects (56%) showed a negative correlation (r²=−.14,ns). These weak correlations are almost identical to those found by Bever and Sanz (1997). To see whether the effect for subject position and verb would replicate, the difference between sequence-sensitive and sequence-insensitive individuals was taken into account for the native speakers in this experiment as well. Once subjects were divided into these groups, then the results replicated those of Bever and Sanz's (1997). A factorial ANOVA showed a verb by position by comprehender interaction, F (1, 33)=32.045, p<.0001.

The mean reactions times for Spanish native speakers.

As can be seen in Figure 1, sequence sensitive individuals reacted faster to unaccusative verbs with preverbal subjects (mean RT=987) than to unergative verbs with preverbal subjects (mean RT=1091). The 104-ms difference between the two verbs was statistically significant, F (1, 15)=14.325, p<.001. By contrast, the sequence-insensitive individuals showed an effect in the opposite direction: they reacted slower to unaccusative verbs (990 ms) than to unergative verbs (904 ms), and the difference was also significant, F (1, 17)=8.546, p<.002. Also replicating Bever and Sanz's (1997) results, sentences with postverbal subjects were responded to faster than sentences with preverbal subjects, and there were no significant differences between unaccusative and unergative verbs when the subject NP was in postverbal position.⁷

A reason why postverbal subjects were responded to faster than preverbal subjects in this study may be related to the fact that each verb appeared twice and sentences with postverbal subjects always appeared after sentences with preverbal subjects in the experimental blocks, so the shorter latency for sentences with postverbal subjects could just be a priming effect. However, Bever and Sanz (1997), who did not present each verb twice and did not control for order of presentation of sentences, also found the same pattern. Therefore, recency (the fact that the subject NP appearing last is closer in time and in the string to the question with the probe) might be the explanation for this subject position effect.

Let us now turn to the results of the L2 learners, illustrated in Figure 2.

The mean reaction times for L2 learners.

The L2 learners show the same pattern of response as the sequence-sensitive native speakers. First, sentences with postverbal subjects were responded to faster than sentences with preverbal subjects, but the difference between the two verb classes was not significant. Second, with preverbal subjects, the L2 learners patterned with the sequence-sensitive native speakers, reacting faster to unaccusative verbs (1053 ms) than to unergative verbs (1097 ms), and the 44-ms difference was significant, F (1, 43)=4.558, p<.038. When the sequence-sensitive versus sequence-insensitive distinction was applied to the L2 learners on the basis of the correlation between reaction time and syllable length of distracter items, the results patterned with those of the native speakers. Sequence-sensitive L2 learners displayed shorter latencies for unaccusative verbs (913 ms) than for unergative verbs (1011 ms), whereas sequence-insensitive L2 learners displayed the opposite pattern: unaccusatives were responded to slower (1070 ms) than unergative verbs (999 ms), and these differences were significant, F (1, 43)=13.423, p<.0001.

To summarize, sequence-sensitive native speakers and L2 learners recognized probe words in the antecedent faster with unaccusative verb sentences with preverbal subjects than with unergative verb sentences with postverbal subjects. This result suggests that there is a syntactic trace or copy in unaccusative-verb sentences, that the NP trace/copy facilitates recognition of the antecedent, and that L2 learners process unaccusative verbs like native speakers.

RESULTS BY SEMANTIC SUBCLASSES

Let us now turn to our other research question: whether native speakers and L2 learners respond differently to different subclasses of unaccusative and unergative verbs depending on their relative position in the unaccusativity hierarchy proposed by Sorace (2000). It was predicted that core unaccusatives should be responded to faster than core unergatives. However, because less core and peripheral verbs are supposed to display variable syntactic behavior within and across languages, differentresponses are expected here: namely, faster RTs for unaccusatives than unergatives, opposite RTs, or no difference between RTs between unaccusatives and unergatives. Crucially, core unergatives should not display shorter latencies than core unaccusatives.

The statistical analysis applied to all the verbs confirmed these trends for the two groups. Note that for this analysis I did not take into account the sequence-sensitive or -insensitive distinction for the native speakers. There was a verb by semantic subclass interaction, F (1, 77)=782.241, p<.0001. The native speakers and the L2 learners responded faster to core and peripheral unaccusatives than to core and peripheral unergatives, and the two groups showed the opposite pattern of response with the less core classes, where the unergatives were responded to faster than the unaccusatives. The results are shown in Figures 3 and 4, respectively.

The mean reaction times per verb class for native speakers.

The mean reaction times per verb class for L2 learners.

This verb analysis shows that the lack of effect for the native speakers in favor of unaccusative verbs shown in Figure 1 are explained by the fact that within the unaccusative and unergative class, the less core verbs display the opposite RT pattern from the other two verb classes, canceling out the effect in favor of unaccusatives. Therefore, an explanation in terms of different types of comprehenders is not necessary.

Recall that the verbs for this study were selected on their lexical semantic meaning and not on the basis of word frequency.⁸

Bever and Sanz (1997) checked verb frequencies by conducting a survey with native speakers.

This presents a challenge to the present experiment because in processing studies word frequency is directly related to speed of word recognition (the more frequent, the faster a word is recognized, and vice versa). To work around this problem, the frequency of all the verbs used in the experiment were checked in a manual of Spanish word frequencies (Alameda & Cuetos, 1995). The results are shown in Table 3.

What emerges from Table 3 is that core and less core unaccusative verbs are more frequent than core and less core unergative verbs. The peripheral verbs are comparably infrequent. To rule out the possibility that the results obtained with unaccusative verbs were not due to a frequency effect, correlations between RTs for individual verbs and frequencies were calculated. For the native speakers, there was a very weak positive correlation (r²=.007), and for the L2 learners, there was a very weak negative correlation (r²=−.015). In both cases, the correlations turned out to be nonsignificant. Nonetheless, to compare the results of different semantic subclasses and their position in the hierarchy, only verbs of comparable frequency were used. That is, for the core unaccusative and core unergative class, only the verbs fall (freq.=92) and speak (freq.=106) were taken into account; for the less core classes, run (freq.=45) and disappear (freq.=54) were contrasted; and because all the peripheral unaccusative and unergative verbs were comparably infrequent (freq.<5), the 3 verbs in each class were compared.

The results of the native speakers displayed in Figure 5 showed an effect in the predicted direction with the core verbs: unaccusatives were responded to faster than unergatives, and the 68-ms difference was significant according to paired sample t tests, t (34)=−3.275, p<.002. With respect to the noncore verbs, there was no effect for the less core verbs, t (34)=−.954, p<.347, but there was an effect in favor of unaccusatives for the periphery verbs, t (34)=3.829, p<.001.

The mean reaction times for frequency matched unaccusative and unergative verbs per semantic subclass for native speakers.

The results of the L2 learners shown in Figure 6 follow the pattern of response of the native speakers, except that the RTs differences between the verbs were even more marked. The difference between RTs for core unaccusatives and unergatives was 238 ms, t (43)=−3.241, p<.002, and the difference between periphery unaccusatives and unergatives was 197 ms, t (43)=4.196, p<.001. The difference between the less core verbs, of 217 ms, resulted in faster latencies for unergatives, t (43)=3.987, p<.001. Therefore, it seems that verb frequency does not determine how unaccusative and unergative verbs are processed by native and nonnative speakers. Furthermore, by looking at verbs in this way, the distinction between sequence-sensitive and sequence-insensitive individuals is not called for to explain the results.

The mean reaction times for frequency matched unaccusative and unergative verbs per semantic subclass for L2 learners.