The On-Line Resolution of the Sentence Complement/Relative Clause Ambiguity: Evidence from Spanish Josep Demestre and José E. Garcı́a-Albea Department of Psychology, University Rovira i Virgili, Tarragona, Spain Abstract. Two self-paced reading experiments investigated syntactic ambiguity resolution in Spanish. The experiments examined the way in which Spanish subjects initially interpret sentences that are temporarily ambiguous between a sentence complement and a relative clause interpretation. Experiment 1 examined whether the sentence complement preference found in English is observed in Spanish speaking subjects. In Experiment 2, verbal mood was manipulated in order to study the influence of verb-specific information on sentence processing. Since subcategorization for a subjunctive complement clause is generally assumed to be a lexical property of some verbs, the manipulation of the mood of the embedded verb affords us an interesting and novel way to examine the influence of lexical information on syntactic ambiguity resolution. Experiment 1 showed that Spanish speakers initially interpret the ambiguous that-clause as a sentence complement. Experiment 2 showed that verb-specific information, in particular, the information that specifies that a verb subcategorizes for a subjunctive complement, is accessed and used rapidly and affects the ambiguity resolution process. The results are discussed in relation to current models of sentence processing. Key words: parsing mechanisms, syntactic ambiguity resolution, cross-linguistic research, Spanish Introduction One of the central goals of sentence processing research is to provide a systematic account of how people interpret structurally ambiguous sentences. The study of how the language comprehension system deals with ambiguous constructions can reveal the nature of the underlying mechanisms that are responsible for building a parse and for recovering from an initially incorrect interpretation. The sentence complement/relative clause (hereafter, SC/RC) ambiguity in English has been studied This research was supported by DGES PB96Ð1021 and BSO2000Ð1252 project grants. Josep Demestre was supported by DGICYT pre-doctoral fellowship FP94Ð 46230064. We would like to thank Christoph Klauer and three anonymous reviewers for their insightful comments on the manuscript. DOI: 10.1027//1618-3169.51.1.59 ” 2004 Hogrefe & Huber Publishers in a number of papers (see Altmann, 1988; Altmann, Garnham, & Dennis, 1992; Altmann, Garnham, & Henstra, 1994; Crain & Steedman, 1985; Mitchell, Corley, & Garnham, 1992). In a study of this ambiguity, Mitchell et al. (1992) presented evidence that English-speaking subjects are initially biased in favor of the complement interpretation of an ambiguous that-clause. In the first experiment, subjects read temporarily ambiguous garden-path sentences that were presented in isolation. They compared the reading times for a critical display (the two words following the ambiguous word that) in five different conditions: a subject-relative clause, an object-relative clause, a sentence complement, and two control conditions. The crucial finding of the Mitchell et al.’s first experiment is that they observed increased reading times in the critical display for the subject-relative compared to the object-relative and the SC versions. They observed no differences between the SC and Experimental Psychology 2004; Vol. 51(1): 59Ð71 60 Josep Demestre & José E. Garcı́a-Albea the object-relative clause versions. This indicates that immediately after reading the word that, people must have been expecting something other that the subject-relative continuation. Since this effect disappeared in the two control conditions, in which the prior sentential context ruled out a complement interpretation of the that-clause, the authors inferred that the phenomenon must be related in some way to the availability of the complement reading with SCverbs when the word that is encountered. The authors concluded that on reading the word that in sentences headed by a SC-verb, readers were expecting a complement, and that they were garden-pathed (in the subject-relative version) when the next two words could not be reconciled with this structure. Mitchell et al.’s results seem to indicate that English-speaking subjects have an initial preference for attaching the ambiguous that-clause as a SC. These findings have been interpreted as supporting models of sentence processing in which initial attachment decisions are guided by structurally-based principles (see below for a detailed description of theses models). Until recently, sentence-processing research has focused mainly on English constructions. Cross-linguistic studies can play an important role in sentence-processing research, since different parsing models make substantially different claims regarding the universality of the mechanisms governing syntactic processing. To our best knowledge, the SC/RC ambiguity has not yet been studied in Spanish. Syntactic ambiguity resolution research conducted in Spanish has mainly investigated the attachment of RCs to complex NPs (Carreiras, 1992; Carreiras & Clifton, 1993; Cuetos & Mitchell, 1988; Gilboy, Sopena, Clifton, & Frazier, 1995; Gibson, Pearlmutter, Canseco-González, & Hickok, 1996). Other ambiguous structures have been studied but to a lesser degree (Igoa, 1995, 1996; Meseguer, 1995; Meseguer, Carreiras, & Clifton, 2002). The goal of the work presented here was twofold: (1) we aimed to study the on-line resolution of SC/ RC ambiguous sentences in Spanish in order to examine whether the preference in resolving such ambiguity in Spanish is the same as the one observed in English, and (2) to exploit the richness of the Spanish mood system to examine, in a way that is not possible in English, the on-line influence of verbspecific information on sentence processing. Experiment 1 investigated whether the initial SC-preference reported in the literature is replicated in Spanish. Experiment 2 examined the influence of lexically specific information in syntactic ambiguity resolution; in particular, we investigated the SC/RC ambiguity in sentences that are disambiguated by verbal mood information. In Experiment 1, we studied temporarily ambiguous sentences such as those in (1): Experimental Psychology 2004; Vol. 51(1): 59Ð71 (1a) El fontaneroi le contó al cocinero que (proi) pasaba unos dı́as en la playa y que (proi) buscaba una casa cerca del mar Literal Translation: The plumberi told the cook that (proi) was spending some days on the beach and that (proi) was looking for a house near the sea. Translation: The plumberi told the cook that hei was spending some days on the beach and that hei was looking for a house near the sea. (1b) El fontaneroi le contó al cocineroj quej (tj) pasaba unos dı́as en la playa que (proi) buscaba una casa cerca del mar Literal Translation: The plumberi told the cookj thatj (tj) was spending some days on the beach that (proi) was looking for a house near the sea Translation: The plumberi told the cookj thatj (tj) was spending some days on the beach that hei was looking for a house near the sea. In these sentences, the underlined that-clause (que pasaba unos dı́as en la playa) is temporarily ambiguous since it can be either a SC (1a) or a RC (1b). Whereas sentence (1a) contains a complex conjoined SC (the first that-clause, the conjunction, and the second that-clause), sentence (1b) contains a RC (the first that-clause) and a simple SC (the second thatclause). It is important to note that the presence/absence of the conjunction y preceding the second thatclause is critical for the interpretation of the first that-clause. Whereas its absence makes the RC-interpretation the only grammatical interpretation of the first that-clause, its presence does not disambiguate the first that-clause, since both the SC and the RCinterpretation are allowed by the grammar. Sentence (1a) could be temporarily interpreted as a conjoined RC, but the fact that the second that-clause is the last clause of the sentence makes the conjoined SCinterpretation the only grammatical interpretation. Hence, whereas the absence of the conjunction (1b) forces the interpretation toward the less preferred structure, its presence (1a) is in accordance with the preferred interpretation. The ambiguity in (1) arises because que can be either a sentence complementizer or a relative pronoun. Such a lexical category ambiguity produces a local syntactic ambiguity since, at this point (the first que), two possible continuations are allowed by the grammar. If que is treated as a sentence complementizer, then the resulting interpretation will be a SC. In contrast, a RC interpretation will be pursued if the ambiguous word is assumed to be a relative pronoun. ” 2004 Hogrefe & Huber Publishers Ambiguity Resolution in Spanish In NP-V-PP-que contexts in which the V is a SCverb, que is more likely to be a sentence complementizer than a relative pronoun. The construction NP-V-PP-SC is the more frequent pattern in Spanish, and the NP-V-PP-RC the less frequent one. A corpus analysis1 of NP-V-PP-que contexts revealed that for the verbs used in Experiment 1 and 2, in 90.08 & and 97.16 % of the sentences, respectively, que is treated as a sentence complementizer. That is, the ambiguous word que is used much more frequently as a sentence complementizer than as a relative pronoun when following a NP-V-PP sequence, and this frequency asymmetry is just as marked with the verbs used in the first experiment as it is with the verbs of the second experiment. This situation is the same as the one in English, where the SC/RC ambiguity has a high degree of asymmetry in initial availability of the two syntactic alternatives (Spivey-Knowlton & Tanenhaus, 1994). In the literature, there is an interesting debate regarding the nature of the mechanisms underlying syntactic ambiguity resolution. The debate concerns the sources of information the sentence processing system uses in initial structure assignment. On the one side, there are the so-called two-stage models, which assume that initial syntactic commitments are made on the basis of a highly restricted domain of purely syntactic information. On the other side, there are the so-called constraint-based models, which assume that multiple alternatives are at least partially available, and that ambiguity resolution is accomplished by the use of multiple sources of probabilistic information from other nonsyntactic domains. One of the most influential two-stage models is the garden-path theory proposed by Frazier and colleagues (Ferreira & Clifton, 1986; Frazier, 1987, 1990; Frazier & Rayner, 1982; Rayner, Carlson, & Frazier, 1983). The garden-path model assumes that parsing proceeds serially and that the sentence processor operates in two functionally distinct processing stages. During the first stage, a restricted domain of syntactically relevant information is used to posit an initial structure. In the second stage, the thematic processor (Frazier, 1987; Rayner et al., 1983) evalu1 A corpus study was conducted to obtain a record of the relative frequency with which the ambiguous word que is used as a sentence complementizer and as a relative pronoun when following NP-V-PP sequences. We employed a one million word corpus extracted from a Spanish journals database. A search procedure was used to find all the verbs used in both experiments. We then searched for NP-V-PP sequences in which the PP was headed by a (which indicates the presence of an explicit indirect object). All sentences containing no explicit indirect object following the verb were excluded. This decision was made because the RC option is not available when the indirect object is not specified. ” 2004 Hogrefe & Huber Publishers 61 ates and determines the semantic well-formedness or plausibility of the initial parse. If the initial analysis submitted by the syntactic parser is deemed to be semantically deviant, the thematic processor rejects the initial structure and guides a subsequent reanalysis of the ambiguous string. According to the garden-path model, minimal attachment, which is considered to be a universal parsing principle (Frazier, 1987, 1990), will be used during the first stage to select the simplest structure over the set of alternatives. In SC/RC ambiguous sentences, minimal attachment will initially select the SC reading of the that-clause because it is syntactically simpler than the RC alternative. The garden-path model has been most directly contrasted with a class of models known as constraint-based parsing models (MacDonald, 1994; MacDonald, Pearlmutter, & Seidenberg, 1994a, 1994b; Tabossi, Spivey-Knowlton, McRae, & Tanenhaus, 1994; Trueswell & Tanenhaus, 1994). Constraint-based models conceive syntactic ambiguity resolution as a dynamic, continuous constraint satisfaction process, in which different constraints provide evidence in support of partially activated alternatives. Ambiguity resolution is viewed as the competition and interaction of multiple constraints that are relevant to sentence processing. Following interactive activation models (Elman & McClelland, 1984; McClelland & Rumelhart, 1981), constraintbased theorists assume that the alternative interpretations become activated to differing degrees and that they compete for activation; strongly activated interpretations strongly inhibit competitors, whereas weaker interpretations can exert only weak inhibition on competitors. One interpretation eventually “wins” the competition when its activation exceeds some threshold of activation, and all alternative interpretations are completely inhibited. Garden-paths occur when an alternative that is strongly supported by initial evidence later turns out to be incorrect and the correct alternative is no longer active. The two approaches make similar predictions about the processing of fragments that are temporarily ambiguous between a SC and a RC. However, their predictions rely on very different underlying assumptions. The garden-path model predicts that the SC will be the only structure initially computed, because it is the structurally simplest one among the alternatives. The garden-path theory predicts that readers should be garden-pathed whenever the SCinterpretation turns out to be inappropriate. Accordingly, we should find an increased processing load at the second that-clause for those sentences that are finally interpreted as containing a RC. Constraintbased models predict that multiple constraints will determine the initial level of activation of the competing alternatives. In the SC/RC ambiguity two conExperimental Psychology 2004; Vol. 51(1): 59Ð71 62 Josep Demestre & José E. Garcı́a-Albea straints will have particular influence on initial activation: the relative frequency with which a verb is used with a SC, and the relative frequency with which que is used as a complementizer or a relative pronoun. The target verbs used in the following experiment subcategorize for a sentence complement, and in a NP-V-PP-que context, there is a large frequency asymmetry that makes the SC the dominant alternative and the RC the subordinate alternative. Thus, constraint-based models predict a high activation level for the SC and a low activation level for the RC-alternative, resulting in an increased processing difficulty when the disambiguating information is consistent with the less active RC alternative. Experiment 1 In this experiment we used sentences like the examples in (1) to investigate the initial preference in interpreting a temporarily ambiguous that-clause. We aimed to examine whether the initial SC-preference observed in English is replicated in Spanish. Contrary to most of the experiments conducted in English to study the SC/RC ambiguity (with the exception of Mitchell et al.’s (1992) first experiment), we decided to present sentences in isolation, with no previous biasing context. We made this decision because our main goal was not to study the influence of discourse information in initial syntactic parsing decisions, but to explore the interpretation Spanish subjects pursue when faced with such ambiguous constructions. Method Participants Thirty students from the University Rovira i Virgili participated in the experiment for course credits. All participants were native speakers of Spanish with normal or corrected vision. All were naive with respect to the aims of the experiment. The participants were randomly assigned to one of the two experimental groups. Materials Twenty-four SC verbs were selected for the experiment. For each verb, two sentence versions were prepared: a SC version and a RC version. The only surface difference between the two versions was that in the SC condition a conjunction (y) was included between the ambiguous that-clause and the following Experimental Psychology 2004; Vol. 51(1): 59Ð71 that-clause. In all experimental items, the main verb was preceded by the indirect object clitic pronoun le (him/her). In many sentences with verbs that take two complements in Spanish, it is customary to do “clitic doubling” in order to keep the indirect object in focus. Two presentation lists were constructed by combining the 24 experimental sentences with 80 filler sentences and 15 practice items. Within each list, there were 12 SC items and 12 RC items. One version of each item was assigned to one of the two lists. Items were rotated across lists such that each item occurred in each of its conditions. The versions were assigned such that an equal number of items in each condition appeared on each list and no participant saw more than one version of each item. The 80 filler items were of varying syntactic types. The full set of experimental sentences and the different presentation displays is given in the Appendix. Procedure Participants were instructed to read at a normal, comfortable pace in a manner that would enable them to answer a comprehension question. Sentences were presented with a word-by-word self-paced moving-window procedure using PC software. Each trial began with a series of dashes on the computer screen in place of the letters of the words. The first press of a specific key replaced the first set of dashes with the first word in the sentence. With subsequent key presses, the next set of dashes was replaced by the next word, and the preceding word was replaced by dashes. This was repeated until the end of the sentence. The computer recorded the time from when a word was first displayed until the next press of the specified key. When readers finished reading each item, they pressed a key, and the computer displayed a comprehension question. When readers finished reading the question, they pressed a key, and the computer displayed two response alternatives. One alternative was displayed on the left side of the screen, and the other on the right side. For example, the test sentence El fontanero le contó al cocinero que pasaba unos dı́as en la playa y que buscaba una casa cerca del mar was followed by the question ¿Quién habı́a pasado unos dı́as en la playa? (translation: Who had spent some days in the beach?). The question was followed by two alternatives El fontanero Ð El cocinero (translation: The plumber Ð The cook). Readers responded by pressing one of two keys. Readers did not receive feedback on their answers. The computer displayed all sentences on two lines. The line breaks always occurred after the first ambiguous that-region. The experiment lasted for about 30 minutes. ” 2004 Hogrefe & Huber Publishers Ambiguity Resolution in Spanish 63 Results Analysis by Word Position Mean reading times were calculated across subjects and items for twelve word positions in the two experimental conditions. This twelve word positions are shown in Table 1. Reading times more than 2.5 standard deviations away from a subject’s mean reading time for a given word position were considered outliers and were discarded. A summary of the differences between average reading time per word position is given in Figure 1. Mean reading times and ANOVA results for each word position are shown in Table 2. The ANOVA showed no reliable differences for all word positions preceding 2ndQUE. For 2ndQUE reading times to RC sentences were significantly longer than to SC sentences. The 29 ms difference was statistically significant (F1(1, 29) = 20.05, p ⬍ .001; F2(1, 23) = 16.05, p ⬍ .001). For 2ndQUE+1, reading times to RC sentences were on average 35 ms longer than to SC sentences. This difference resulted in a main effect of sentence type (F1(1, 29) = 18.33, p ⬍ .001; F2(1, 23) = 7.58, p ⬍ .01). For 2ndQUE+2, the 24 ms difference between the two conditions resulted in a main effect of sentence type (F1(1, 29) = 4.43, p ⬍ .05; F2(1, 23) = 4.39, p ⬍ .05). No significant effects were found for posterior word positions. Table 1. The Word Positions that Were Analyzed in Experiment 1 Name Description Word in Example Sentence SUB Main Clause‘s Subject MAINV Main Verb OBJ Main Clause’s Indirect Object 1stQ First QUE 1stQ+1 1st word following 1stQ 1stQ+2 2nd word following 1stQ PRE Word preceding 2ndQ * Y The conjunction preceding 2ndQ Second QUE 2ndQ 2ndQ+1 1st word following 2ndQ nd 2 Q+2 2nd word following 2ndQ nd 2 Q+3 3rd word following 2ndQ nd 2 Q+4 4th word following 2ndQ fontanero contó cocinero Discussion que pasaba unos playa y que buscaba una casa cerca Note. * In SC sentences, the 2ndQ was preceded by the conjunction y, which was preceded by PRE. In RC sentences the conjunction y was not present. 50 40 30 20 10 Experiment 2 0 -10 -20 2ndQ+4 2ndQ+3 2ndQ+2 2ndQ+1 PRE 2ndQ 1stQ+2 1stQ+1 1stQ OBJ MAINV -30 SUBJ Mean Reading Time Differences in ms 60 This experiment indicates that Spanish speakers have difficulty in processing the second that-clause when it is consistent with the RC interpretation. The results clearly reveal that reading times to the first three words of the second that-clause were significantly shorter in the SC-condition than in the RCcondition. This experiment shows a clear SC-preference, and thus replicates the results found in English with SC/RC ambiguous constructions when presented in neutral contexts or in isolation, that is, with no previous context. These findings are compatible with both the garden-path model and constraint-based models. The garden-path model (Frazier, 1987; Frazier & Rayner, 1982; Rayner et al., 1983) explains the initial SC preference because of it being the simplest structure over the two possible alternatives. Minimal attachment will select the simplest structure. According to constraint-based models, misanalyses occur because the alternative that is strongly supported by initial evidence later turns out to be incorrect and the correct alternative is no longer (or weakly) active. Word Position Figure 1. The differences (RC minus SC) between the average reading time for each word position in Experiment 1. ” 2004 Hogrefe & Huber Publishers In this experiment, we examined the time course of syntactic ambiguity resolution in SC/RC ambiguous sentences that are disambiguated by verbal mood information. The manipulation of verbal mood in Spanish affords us a new way to study the influence of verb-specific information on sentence processing; by manipulating the mood of the embedded verb we Experimental Psychology 2004; Vol. 51(1): 59Ð71 64 Josep Demestre & José E. Garcı́a-Albea Table 2. Experiment 1: Mean Reading Times in ms and ANOVA Results by Position Word Position SUB MAINV OBJ 1stQ 1stQ+1 1stQ+2 PRE Y 2ndQ 2ndQ+1 2ndQ+2 2ndQ+3 2ndQ+4 RC sentences SC sentences Difference F1(1, 29) 383.19 392.67 389.74 375.96 374.21 388.33 424.60 378.38 394.47 385.70 371.80 370.18 386.38 428.08 275.02 363.71 362.34 394.72 393.19 426.56 4.80 Ð1.79 4.04 4.16 4.03 1.95 Ð3.47 0.37 0.06 0.28 0.33 0.25 0.05 0.07 29.12 35.61 24.78 Ð3.12 5.18 20.05 18.33 4.43 0.17 0.15 392.83 397.95 419.50 390.06 431.74 can examine the influence of lexical information on the early stages of sentences processing. In Spanish, there is a class of three-argument verbs, such as aconsejar (to advise), pedir (to ask for), ordenar (to order), that subcategorize for an NP and a clause. With these verbs there exists the choice between an infinitival and a subjunctive complement clause. Whenever their embedded clause is finite, these verbs obligatorily require the subjunctive mood. For example, with a verb such as aconsejar, the complement clause, which is specified in the verb’s subcategorization frame, whenever it is finite, has to be in the subjunctive mood. In languages such as Spanish, subjunctive is marked by a set of inflectional endings on the verb, which are distinct from those of the indicative paradigm; that is, verbal mood is explicitly indicated by inflectional morphology. Through the manipulation of verbal mood, the influence of verb-specific information can be tested in Spanish in a way that is not possible in English. Consider the examples in (2), in which the subjunctive/indicative subordinate verb is underlined: (2a) El profesor les aconsejó a los alumnosi que (proi) fueran (subjunctive) a la presentación del libro y que (proi) leyeran la obra del autor Literal Translation: The teacher advised the studentsi that (proi) should go to the book’s presentation and that (proi) should read the author’s works Translation: The teacher advised the studentsi (PROi) to go to the book’s presentation and (PROi) to read the author’s works (2b) El profesor les aconsejó a los alumnosi quei (ti) fueron (indicative) a la presentación del libro que (proi) leyeran la obra del autor Experimental Psychology 2004; Vol. 51(1): 59Ð71 p F2(1,23) ⬎ ⬎ ⬎ ⬎ ⬎ ⬎ ⬎ 0.5 0.8 0.5 0.5 0.6 0.8 0.7 0.01 0.75 0.50 0.19 0.86 0.67 0.008 ⬍ ⬍ ⬍ ⬎ ⬎ 0.001 0.001 0.05 0.6 0.6 16.05 7.58 4.39 1.30 1.26 p ⬎ ⬎ ⬎ ⬎ ⬎ ⬎ ⬎ 0.9 0.3 0.4 0.6 0.3 0.4 0.9 ⬍ ⬍ ⬍ ⬎ ⬎ 0.001 0.01 0.05 0.2 0.2 Literal Translation: The teacher advised the studentsi thati (ti) went to the book’s presentation that (proi) should read the author’s works) Translation: The teacher advised the studentsi thati (ti) went to the book’s presentation (PROi) to read the author’s works For sentence (2a), the SC interpretation is available since the subordinate verb is in the subjunctive mood. It is important to note that the subjunctive mood is also compatible with the RC interpretation; that is, in sentence (2a), the first that-clause is syntactically ambiguous since it is compatible with both the SC and the RC interpretation. In sentence (2b), the subordinate verb is in the indicative mood, making the SC interpretation ungrammatical, leaving the RC interpretation the only grammatical interpretation. The two verbal forms differ orthographically only in the last vowel of the word: the verbs in subjunctive have an -an ending and indicative verbs an -on ending. Thus, the two verbal forms are maximally similar in their orthographic form. This experiment examines whether Spanish speakers are sensitive to “violations” in the requirements of the main verb. The detection of such a verbal mood violation (violation of the main verb’s requirements on the mood of the subordinate verb) would not only indicate that verb-specific information is available, but also that the main verb’s finegrained information regarding the verbal mood of the subordinate verb is rapidly used. Therefore, the language processor would have to access information about the possible complements for a particular verb, and information about the mood requirements the subordinate verb has to satisfy. ” 2004 Hogrefe & Huber Publishers Ambiguity Resolution in Spanish The garden-path model (Frazier, 1987; Ferreira & Clifton, 1986; Frazier & Rayner, 1982; Rayner et al., 1983) assumes lexical information, other than category membership, does not play a role in determining the initial parse. If the main verb’s lexical information is not available until a second stage of processing, then we should expect to find no differences in reading times at the subordinate verb. Constraint-based models (MacDonald et al., 1994a, 1994b; Trueswell & Tanenhaus, 1994) assume lexical information plays an important role in sentence processing. According to these models, lexically specific information is activated as soon as a lexical item is encountered in the input string, and is used as a source of information in the parsing process. Constraint-based models, therefore, predict an increased processing load at the indicative subordinate verb, because the incoming input is inconsistent with the most activated alternative. The question that arises at this point is whether the detection of a subordinate verb in the indicative mood will strongly activate the RC interpretation and block the SC interpretation, or the initially weakly activated RC interpretation will not inhibit the highly activated SC. If the parser, when faced with the indicative subordinate verb, pursues the RC interpretation and blocks the SC interpretation, one would expect to find no differences at the second that-clause, since the second que would be treated as a sentence complementizer in order to fulfill the main verb’s subcategorization requirements. If the detection of a cue (the indicative mood) that clearly indicates the infelicity of the SC alternative does not produce a strong inhibition of the more frequent competitor, one would expect slower reading times at the second that-clause for sentences that are finally interpreted as a RC. To sum up, constraint-based models predict that the verbal mood anomaly will be rapidly detected, since lexical information is activated as soon as a lexical item is encountered in the input string. However, it is not clear whether its influence will be strong, that is, activate the RC interpretation and block the alternative interpretation, or weak due to the strength of the dominant alternative, observing also an increased processing load at the second that-clause. Method Participants Thirty-two students from the same pool as the previous experiment participated in the experiment. The participants were randomly assigned to one of the two experimental groups. ” 2004 Hogrefe & Huber Publishers 65 Materials Sixteen sets of two sentences each were constructed. All the verbs used in this experiment belong to a class of verbs that require subjunctive verb forms in sentential complements. The two versions of a target sentence differed for the mood of the subordinate verb and for the presence of a conjunction between the two that-clauses in the SC condition. Two presentation lists were constructed by combining the 16 experimental sentences with 80 filler sentences and 10 practice items. Items were rotated and assigned to each list as in Experiment 1. As can be seen in Table 3, all experimental sentences were presented in nine (RC items) and ten (SC items) different sentence regions. The segmentation criterion was the following: each of the ten different regions was composed of one of the major grammatical phrases of the sentence (subject, main verb, object, complementizer, subordinate verb, subordinate verb’s object, conjunction, complementizer, subordinate verb, and subordinate verb’s object). This segmentation was chosen because it sounded more natural to present the materials in such phrase-like chunks. A true/false question was prepared for each sentence. The full set of experimental sentences is given in the Appendix. Table 3. The Sentence Regions That Were Analyzed in Experiment 2 Name Description SUB Main Clause’s Subject MAINV Main Verb OBJ Main Clause’s Indirect Object 1stQ First QUE Verb of the 1st subordi2ndV nate clause 2ndV+1 The argument following 2ndV * Y The conjunction preceding 2ndQ nd 2 Q Second QUE ** Verb of the 2nd subor3rdV dinate clause 3rdV+1 The argument following 3rdV *** Word in Example Sentence El profesor les aconsejó a los alumnos que fueran a la presentación del libro y que leyeran la obra del autor Note. * 2ndV was always followed by an argument, which, in turn, was modified by a simple PP. ** In SC sentences, the 2ndQ was preceded by the conjunction y, which was preceded by 2ndV+1. In RC sentences the conjunction y was not present. *** 3ndV was always followed by an argument, which in turn, was modified by a simple PP. Experimental Psychology 2004; Vol. 51(1): 59Ð71 66 Josep Demestre & José E. Garcı́a-Albea Figure 2 Procedure Results Mean reading times were calculated across subjects and items for nine sentence regions in the two experimental conditions. The same procedure as in Experiment 1 was used to discard outlier reading times. A summary of the differences between average reading time per region is given in Figure 2. Analysis by Sentence Region Mean reading times and ANOVA results for the different regions are shown in Table 4. The ANOVA showed no reliable differences for all sentence regions preceding 2ndVERB. Sentences containing a subjunctive verb were read faster in the 2ndVERB region than sentences containing an indicative verb (F1(1, 31) = 9.43, p ⬍ .004; F2(1, 15) = 5.15, p ⬍ .04). The statistical analysis revealed no reliable differences for the 2ndVERB+1 region. In the 2ndQUE region, reading times to RC sentences were 60 Mean Reading Time Differences in ms The experimental procedure was identical to Experiment 1 with two exceptions: (1) Sentences were presented with a self-paced region-by-region procedure, and (2) subjects had to decide whether a comprehension sentence was true or false. When readers finished reading the comprehension sentence, they pressed one of two keys (TRUE/FALSE). For example, the test sentence El profesor les aconsejó a los alumnos que fueron a la presentación del libro que leyeran la obra del autor was followed by the sentence Los alumnos fueron a la presentación del libro (translation: The students went to the book’s presentation). The experiment lasted for about 30 minutes. 50 40 30 20 10 0 -10 -20 -30 SUB MAINV OBJ 1stQ 2ndV 2ndV+1 2ndQ 3rdV 3rdV+1 Sentence Region Figure 2. The differences (RC minus SC) between the average reading time for each sentence region in Experiment 2. reliably longer than to SC sentences (F1(1, 31) = 14.02, p ⬍ .001; F2(1, 15) = 6.55, p ⬍ .02). For the 3rdVERB region, reading times were on average 33 ms shorter to SC sentences than to RC sentences. This 33 ms difference resulted in a main effect of sentence type (F1(1, 31) = 6.28, p ⬍ .01; F2(1, 15) = 6.27, p ⬍ .02). No significant differences were found for the 3rdVERB+1 region. In order to examine whether the differences observed in the 2ndVERB region were due to frequency differences between the indicative and subjunctive verbal forms, a t test was conducted. To minimize the variance observed in the frequency of the different verbs, a square root transformation was applied to all verb frequencies. The statistical test revealed reliable differences between the two verbal forms (t = 3,507; p ⬍ .005). The verbs in the indicative mood had an average frequency of 136 occurrences per million (maximum 1447, minimum 1), and the same verbs in the subjunctive mood had an average frequency of 28 occurrences per million (maximum 278, minimum 0; Sebastián-Gallés, Martı́, Carreiras, & Cuetos, 2000). Table 4. Experiment 2: Mean Reading Times in ms and ANOVA Results by Region Sentence Region SUB MAINV OBJ 1stQ 2ndV 2ndV+1 Y 2ndQ 3rdV 3rdV+1 RC sentences SC sentences Difference F1(1, 31) 708.74 874.60 858.42 548.84 675.86 946.89 718.38 867.12 842.50 536.30 626.27 925.39 305.56 484.04 570.43 1081.36 Ð9.63 7.48 15.91 12.53 49.58 21.50 0.12 0.07 0.45 0.53 9.43 0.69 48.86 33.37 Ð18.78 14.02 6.28 0.26 532.90 603.81 1062.58 Experimental Psychology 2004; Vol. 51(1): 59Ð71 p ⬎ ⬎ ⬎ ⬎ ⬍ ⬎ 0.7 0.7 0.5 0.4 0.004 0.4 ⬍ 0.001 ⬍ 0.01 ⬎ 0.6 F2(1, 15) p 0.03 0.55 0.28 0.36 5.15 0.05 ⬎ ⬎ ⬎ ⬎ ⬍ ⬎ 0.8 0.5 0.6 0.5 0.04 0.8 6.55 6.27 0.001 ⬍ 0.02 ⬍ 0.02 ⬎ 0.9 ” 2004 Hogrefe & Huber Publishers Ambiguity Resolution in Spanish Discussion The reading time pattern for the indicative/subjunctive verb suggests that the information that specifies that a verb subcategorizes for a subjunctive complement is available when the subordinate verb is recognized. Reading times to the indicative forms were significantly longer than to the subjunctive forms, showing that subjects are sensitive to the subjunctive/ indicative manipulation. This sensitivity indicates that the main verb’s lexical information is available at this point. The t test showed reliable differences in frequency, the indicative forms being significantly more frequent as compared to the subjunctive forms. Thus, the reading time differences cannot be explained by frequency differences, since the less frequent verbal forms were read reliably faster than the more frequent forms. Despite such frequency asymmetries, the differences observed in the 2ndVERB region were reliable. The data are difficult to explain from a two-stage approach, even if it is assumed that the thematic processor starts its computations very quickly. In the first stage, the parser will initially pursue the SC structure on the basis of purely structural mechanisms. At the second stage, the thematic processor will evaluate the felicity of the initial parse on the basis of lexical, semantic, and pragmatic information. For the parser to detect such a mood anomaly, verb-specific information would have to be available. According to the garden-path theory, verb-specific information is made available in the second stage. It could be argued that reading times are not sensitive enough to allow us to study separately what is being computed in each processing stage. However, it is hard to explain how the structure-based initial stage and the revision stage could take place so rapidly. If there were some temporal parallelism between the two functionally different processing stages, we would then need a clear specification of the temporal overlap of such processing stages, and a detailed description of the communication between them. The garden-path model could try to explain the results for the indicative/subjunctive verb by appealing to a quick reanalysis process. However, even if it is assumed that a quick reanalysis has taken place, this theory would have trouble explaining the data for the second que. It is hard to explain why, having reanalyzed the ambiguous material as a RC, reading times for the second that-clause are longer for RC items than for SC items. The results of the indicative/subjunctive region are more easily explained from the constraint-based approach. According to these models, as soon as the processor recognizes the matrix verb, information regarding the mood of the verb that heads the complement clause will become available. Hence, the ” 2004 Hogrefe & Huber Publishers 67 parser’s sensitivity to the mood anomaly would be due to the fact that it has already accessed the information stored in the matrix verb’s lexical entry. The reading time pattern for the second thatclause seems to indicate that the parser, although having previously detected a mood anomaly, has pursued a SC interpretation. Increased reading times at 2ndQUE and 3rdVERB regions in RC sentences shows that subjects have interpreted the first thatclause as a SC.2 It seems, then, that the processor, despite having detected an anomaly, is pushed to follow the SC interpretation. An important question arises at this point: Why the parser, having detected a cue that indicates the infelicity of the SC-interpretation, is being led up the garden-path (as indicated by increased reading times at 2ndQUE and 3rdVERB regions)? We will try to offer a plausible explanation. The differences in processing difficulty observed in the second that-clause might be explained by the difference in activation level between the two competing alternatives. Initially, the more frequent and less complex SC alternative is much more activated than the RC alternative, since both subcategorization information and relative frequency favor the SC-interpretation. When the processor encounters the indicative verb, the RC alternative receives additional activation, but its activation level is not high enough to strongly inhibit the highly active SC alternative. The SC alternative remains active since the matrix verb requires a complement clause, and for this reason the parser keeps the SC interpretation available until the main verb’s requirements are fulfilled. The competition process between the two interpretations is temporarily dominated by the SC interpretation, since it is favored by subcategorization information and frequency, and because the syntactically relevant cue seems to be not strong enough to override the SC interpretation. The second que triggers the reinterpretation of the first that-clause as a RC, and pushes the system to settle down in a final interpretation of the sentence. 2 A tentative explanation for the pattern observed at the 3rdVERB would be that the differences could be due, not to the experimental manipulation, but to some priming effects. It could be the case that the presence of a previous verb (2ndVERB) in the subjunctive mood and with a similar phonological form would exert some priming effect over the 3rdVERB, decreasing the reading times. Meseguer (1995) examined whether the presence of a previous verb in subjunctive and with a similar phonological form facilitates or reduces the reading times for a posterior verb. He found no evidence of such a facilitation, since no sign of phonological priming nor verbal mood priming was observed. Experimental Psychology 2004; Vol. 51(1): 59Ð71 68 Josep Demestre & José E. Garcı́a-Albea General Discussion and Conclusions The experiments reported in this paper study the online resolution of SC/RC temporarily ambiguous sentences in Spanish. The results of Experiment 1 and 2 show that Spanish speakers have a clear preference for initially interpreting an ambiguous that-clause as a SC, as shown by the difficulty in processing ambiguous sentences that turn out to be inconsistent with the SC interpretation. This preference replicates the results of previous experiments conducted in English (Altmann, 1988; Altmann, et al., 1992; Altmann et al., 1994; Mitchell et al., 1992). The results of Experiment 1 are compatible with both two-stage models and constraint-based models. The garden-path model (Frazier, 1987; Frazier & Rayner, 1982; Rayner et al., 1983) explains the initial SC preference because of it being the simplest structure over the two possible alternatives. Minimal attachment will select the simplest structure on the basis of purely structural information. Constraint-based models (MacDonald, 1994; MacDonald et al., 1994a, 1994b; Trueswell & Tanenhaus, 1994) predict an initial SC preference on the basis of the main verb’s subcategorization information, and the relative frequency with which que appears to be a complementizer in NP-V-PP contexts. For the verbs we have studied, both subcategorization information and the more likely use of que as a complementizer favor the SC reading. Experiment 2 addressed the question of whether verb-specific lexical information is used rapidly in sentence processing. We used verbs that belong to a class of three-argument verbs that subcategorize for an NP and a clause. With these verbs there exists the choice between an infinitival and a subjunctive complement clause. Whenever their embedded clause is finite, these verbs obligatorily require the subjunctive mood. In sentences in which the subordinate verb is not in subjunctive mood, the SC interpretation is no longer correct, so the RC interpretation is the only grammatical continuation. However, in order for the sentence to be grammatical it must contain three arguments, since the main verb obligatorily requires three arguments. Hence, at some point after the RC a third argument must appear in the input string. By using these three-argument verbs, the influence of verb-specific information can be tested in Spanish in a way that is not possible in English. Through the manipulation of verbal mood, we aimed to examine whether such fine-grained verb-specific lexical information is used when the subordinate verb is recognized. Constraint-based models and two-stage models make different predictions regarding the availability of such lexically specific information. While conExperimental Psychology 2004; Vol. 51(1): 59Ð71 straint-based models assume lexical information is made available as soon as the main verb is recognized, two-stage models assume information specific to each lexical item is made available in the second stage. The results of Experiment 2 indicate that subjects are sensitive to the verbal mood manipulation, since reading times to the subordinate verb were reliably longer in the indicative condition as compared to the subjunctive condition. To explain these differences we have to assume that the processor has accessed the matrix verb’s information which specifies the requirement that the subordinate verb be in the subjunctive mood. Thus, the results indicate that verbspecific information is accessed and used rapidly. The fact that we observed increased reading times at the second that-clause could be explained as follows: Although a syntactically relevant cue has been detected, its influence on the parsing process is weak because the competing alternative is strong and the parser tries to fulfill the matrix verb’s requirements as soon as possible. In summary, the results of the two experiments clearly show that Spanish speakers have an initial preference for interpreting the ambiguous that-clause as a sentence complement. As shown by the Spanish corpus analysis, the SC/RC temporary ambiguity presents an asymmetry in initial availability of the two syntactic alternatives that results in a processing cost for sentences that are finally interpreted as the less frequent and more complex RC. Most importantly, we have shown a new way, which is not possible in English, to examine the influence of verb-specific information on sentence processing. The use of languages other than English in sentence processing research can be very informative to understand the underlying mechanisms governing the interpretation of sentences. References Altmann, G. T. (1988). Ambiguity, parsing strategies, and computational models. Language and Cognitive Processes, 3(2), 73Ð97. Altmann, G. T., Garnham, A., & Dennis, Y. (1992). Avoiding the garden path: Eye movements in context. Journal of Memory and Language, 31, 685Ð712. Altmann, G. T., Garnham, A., & Henstra, J. A. (1994). Effects of syntax in human sentence parsing: Evidence against a structure-based proposal mechanism. Journal of Experimental Psychology: Learning, Memory and Cognition, 20, 209Ð216. Carreiras, M. (1992). Estrategias de análisis sintáctico en el procesamiento de frases: Cierre Temprano vs. Cierre Tardı́o [Parsing strategies in sentence processing: Late closure vs. early closure]. Cognitiva, 4, 3Ð27. Carreiras, M., & Clifton, C., Jr. (1993). Relative clause interpretation preferences in Spanish and English. Language and Speech, 36, 353Ð372. ” 2004 Hogrefe & Huber Publishers Ambiguity Resolution in Spanish Crain, S., & Steedman, M. (1985). On not being led up the garden path: The use of context by the psychological parser. In D. R. Dowty, L. Karttunen, & A. M. Zwicky (Eds.), Natural language parsing: Psychological, computational, and theoretical perspectives (pp. 320Ð358). New York: Cambridge University Press. Cuetos, F., & Mitchell, D. C. (1988). Cross-linguistic differences in parsing: Restrictions on the use of the Late Closure strategy in Spanish. Cognition, 30, 83Ð106. Elman, J. L., & McClelland, J. L. (1984). Speech perception as a cognitive process: The interactive activation model. In N. Lass (Ed.), Speech and Language (Vol. 10, pp. 337Ð374). New York: Academic Press. Ferreira, F., & Clifton, C., Jr. (1986). The independence of syntactic processing. Journal of Memory and Language, 25, 348Ð368. Frazier, L. (1987). Sentence processing: A tutorial review. In M. Coltheart (Ed.), Attention and Performance XII. Hillsdale, NJ: Erlbaum. Frazier, L. (1990). Exploring the architecture of the language-processing system. In G. Altmann (Ed.), Cognitive models of speech processing: Psycholinguistic and computational perspectives. Cambridge, MA: MIT Press. Frazier, L., & Rayner, K. (1982). Making and correcting errors during sentence comprehension: Eye movements in the analysis of structurally ambiguous sentences. Cognitive Psychology, 14, 178Ð210. Gibson, E., Pearlmutter, N. J., Canseco-González, E., & Hickok, G. (1996). Recency preference in the human sentence processing mechanism. Cognition, 59, 23Ð59. Gilboy, E., Sopena, J. M., Clifton, C., Jr., & Frazier, L. (1995). Argument structure and association preferences in Spanish and English complex NPs. Cognition, 54, 131Ð167. Igoa, J. M. (1995, April). Parsing decisions and the construal hypothesis: Attachment preferences in primary phrases in Spanish. Paper presented at the Second Symposium on Psycholinguistics, Tarragona, Spain. Igoa, J. M. (1996, March). A study on late closure in Spanish: Computationally-grounded vs. Frequency-based accounts of attachment preferences. Poster presented at the Ninth Annual CUNY Sentence Processing Conference, New York, NY. MacDonald, M. C. (1994). Probabilistic constraints and syntactic ambiguity resolution. Language and Cognitive Processes, 9, 692Ð715. MacDonald, M. C., Pearlmutter, N. J., & Seidenberg, M. S. (1994a). The lexical nature of syntactic ambiguity resolution. Psychological Review, 101, 676Ð703. MacDonald, M. C., Pearlmutter, N. J., & Seidenberg, M. S. (1994b). Syntactic ambiguity resolution as lexical ambiguity resolution. In C. Clifton, L. Frazier, & K. Rayner (Eds.), Perspectives on sentence processing. Hillsdale, NJ: Erlbaum. ” 2004 Hogrefe & Huber Publishers 69 McClelland, J. L., & Rumelhart, D. E. (1981). An interactive activation model of context effects in letter perception: Part 1. An account of basic findings. Psychological Review, 88, 375Ð407. Meseguer, E. (1995). Estrategias de procesamiento sintáctico en Español [Parsing strategies in Spanish]. Unpublished Doctoral Dissertation. Cognitive Psychology Department. University of La Laguna, Tenerife, Spain. Meseguer, E., Carreiras, M., & Clifton, C. Jr (2002). Overt reanalysis strategies and eye movements during the reading of mild garden path sentences. Memory & Cognition, 30 (4), 551Ð561. Mitchell, D. C., Corley, M. M. B., & Garnham, A. (1992). Effects of context in human sentence parsing: Evidence against a discourse-based proposal mechanism. Journal of Experimental Psychology: Learning, Memory and Cognition, 18, 69Ð88. Rayner, K., Carlson, M., & Frazier, L. (1983). The interaction of syntax and semantics during sentence processing: Eye movements in the analysis of semantically biased verbs. Journal of Verbal Learning and Verbal Behavior, 22, 358Ð374. Sebastián-Gallés, N., Martı́, M. A., Carreiras, M., & Cuetos, F. (2000). LEXESP: Léxico informatizado del español [LEXESP: Computerized Spanish lexicon]. Barcelona, Spain: Edicions Universitat de Barcelona. Spivey-Knowlton, M. J., & Tanenhaus, M. K. (1994). Referential context and syntactic ambiguity resolution. In C. Clifton, L. Frazier, & K. Rayner (Eds.), Perspectives on sentence processing. Hillsdale, NJ: Erlbaum. Tabossi, P., Spivey-Knowlton, M. J., McRae, K., & Tanenhaus, M. K. (1994). Semantic effects on syntactic ambiguity resolution. In C. Umilta & M. Moscovitch (Eds.), Attention and Performance XV. Hillsdale, NJ: Erlbaum. Trueswell, J. C., & Tanenhaus, M. K. (1994). Toward a constraint-based lexicalist approach to syntactic ambiguity resolution. In C. Clifton, L. Frazier, & K. Rayner (Eds.), Perspectives on sentence processing. Hillsdale, NJ: Erlbaum. Josep Demestre Department of Psychology University Rovira i Virgili E-43007 Tarragona Spain Tel.: +34 77 558567 Fax: +34 77 555088 E-mail: jdv@fcep.urv.es Received October 15, 2002 Revision received March 28, 2003 Accepted May 20, 2003 Experimental Psychology 2004; Vol. 51(1): 59Ð71 70 Josep Demestre & José E. Garcı́a-Albea Appendix Note that “/” marks indicate display boundaries. In brackets, the words of the SC condition that differed from the RC condition. In Experiment 1, the only difference between conditions was the presence of the conjunction in SC items. In Experiment 2, there were two differences: the presence of the conjunction and the mood manipulation. Line breaks occurred at the points marked “(%)”. ITEMS IN EXPERIMENT 1 [y/]que/le/faltaba/abonar/el/importe/de/las/ comidas 1. El/fontanero/le/contó/al/cocinero/que/pasaba/ unos/dı́as/en/la/playa/(%) [y/]que/buscaba/una/casa/cerca/del/mar 15. El/cámara/le/gritó/al/actor/que/estaba/cansado/ de/esperar/(%) [y/]que/no/podı́an/perder/más/tiempo 2. El/profesor/le/dijo/al/alumno/que/debı́a/repetir/ curso/(%) [y/]que/hablase/con/el/director 16. El/ciclista/le/recriminó/al/espectador/que/habı́a/ lanzado/agua/(%) [y/]que/no/sabı́a/comportarse/adecuadamente 3. El/encargado/le/indicó/al/aprendiz/que/habı́a/ estropeado/la/máquina/(%) [y/]que/deberı́a/ir/con/más/cuidado 17. El/cantante/le/comentó/al/guitarrista/que/habı́a/ estado/muy/acertado/(%) [y/]que/estaba/muy/contento/de/la/gira 4. El/entrenador/le/comentó/al/gimnasta/que/habı́a/ cometido/un/fallo/(%) [y/]que/aún/podı́a/ganar/la/prueba 18. El/cliente/le/afirmó/al/camarero/que/vivı́a/cerca/ del/puerto/(%) [y/]que/volverı́a/más/a/menudo 5. El/corresponsal/le/expuso/al/redactor/que/habı́a/ sido/censurado/(%) [y/]que/pondrı́a/una/denuncia 19. El/goleador/le/aclaró/al/técnico/que/querı́a/ abandonar/el/club/(%) [y/]que/pensaba/jugar/en/el/extranjero 6. El/capataz/le/insistió/al/obrero/que/habı́a/ avisado/a/los/bomberos/(%) [y/]que/el/accidente/habı́a/sido/fortuito 20. La/fotógrafa/le/confesó/a/la/modelo/que/estaba/ muy/angustiada/(%) [y/]que/esta/podı́a/ser/una/gran/oportunidad 7. El/bombero/le/insinuó/al/propietario/que/no/ poseı́a/seguro/contra/incendios/(%) [y/]que/el/fuego/parecı́a/intencionado 21. La/modelo/le/explicó/a/la/telefonista/que/querı́a/ estudiar/farmacia/(%) [y/]que/conoció/a/un/chico/muy/simpático 8. El/delincuente/le/manifestó/al/periodista/que/ seguı́a/el/caso/con/interés/(%) [y/]que/se/arrepentı́a/de/todo 22. El/esquiador/le/declaró/al/periodista/que/habı́a/ estado/en/tres/olimpiadas/(%) [y/]que/la/pista/parecı́a/muy/rápida 9. La/abogada/le/demostró/a/la/juez/que/habı́a/sido/ investigada/(%) [y/]que/sus/pruebas/eran/verdaderas 23. El/guardia/le/advirtió/al/conductor/que/se/habı́a/ saltado/un/semáforo/(%) [y/]que/le/inmovilizarı́a/el/coche 10. El/jugador/le/respondió/al/árbitro/que/habı́a/ recibido/un/pelotazo/(%) [y/]que/no/pensaba/abandonar/el/campo 24. El/peletero/le/recordó/a/la/modista/que/tenı́a/ una/reunión/de/trabajo/(%) [y/]que/habı́an/quedado/para/verse/el/sábado 11. El/niño/le/juró/al/profesor/que/preparaba/las/ clases/cada/dı́a/(%) [y/]que/no/volverı́a/a/llegar/tarde ITEMS IN EXPERIMENT 2 12. El/responsable/le/prometió/al/fı́sico/que/estaba/ preparando/un/proyecto/(%) [y/]que/intentarı́a/encontrarle/un/trabajo 13. El/vendedor/le/aseguró/a/la/clienta/que/ compraba/en/las/mejores/tiendas/(%) [y/]que/no/encontrarı́a/mejor/pescado 14. El/hotelero/le/argumentó/al/turista/que/habı́a/ pagado/tres/noches/(%) Experimental Psychology 2004; Vol. 51(1): 59Ð71 1. El profesor/les aconsejó/a los alumnos/que/fueron[fueran]/a la presentación del libro/(%) [y/]que/leyeran/la obra del autor 2. La cantante/les indicó/a los músicos/que/recibieron[recibieran]/a la prensa especializada/(%) [y/]que/estuvieran/en la rueda de prensa 3. El encargado/les mandó/a los trabajadores/que/ atendieron[atendieran]/a los clientes rusos/(%) [y/]que/fueran/más amables por teléfono ” 2004 Hogrefe & Huber Publishers Ambiguity Resolution in Spanish 4. El ingeniero/les encargó/a los delineantes/que/dibujaron[dibujaran]/los planos de la obra/(%) [y/]que/presentaran/el proyecto del trazado 5. El escritor/les sugirió/a los jóvenes/que/compraron[compraran]/la última novela/(%) [y/]que/leyeran/los grandes clásicos americanos 6. La encargada/les ordenó/a las azafatas/que/estuvieron[estuvieran]/en la recepción del hotel/(%) [y/]que/acompañaran/a los invitados al convite 7. El cocinero/les recomendó/a las chicas/que/compraron[compraran]/carne de buey/(%) [y/]que/hicieran/un estofado con verduras 8. La delegada/les rogó/a los abogados/que/pagaron[pagaran]/las deudas del bufete/(%) [y/]que/guardaran/los resguardos del pago 9. El médico/les pidió/a los enfermos/que/limpiaron[limpiaran]/las habitaciones de la planta/(%) [y/]que/vaciaran/los cubos de basura 10. La dependienta/les suplicó/a los jóvenes/que/salieron[salieran]/de los probadores del local/(%) [y/]que/no tocaran/las prendas del escaparate ” 2004 Hogrefe & Huber Publishers 71 11. La modista/les solicitó/a los periodistas/que/asistieron[asistieran]/al desfile de verano/(%) [y/]que/fotografiaran/a la modelo de color 12. El novio/les propuso/a los camareros/que/sirvieron[sirvieran]/el aperitivo del banquete/(%) [y/]que/cenaran/en la cocina del hotel 13. El carpintero/les dijo/a los ayudantes/que/arreglaron[arreglaran]/los muebles de la cocina/(%) [y/]que/estudiaran/el presupuesto de la reforma 14. La portera/les recomendó/a las vecinas/que/dejaron[dejaran]/las botellas de cristal/(%) [y/]que/salieran/por la puerta del garaje 15. El jubilado/les prohibió/a los chavales/que/vieron[vieran]/la pelı́cula de terror/(%) [y/]que/salieran/al patio del edificio 16. El pintor/les exigió/a los trabajadores/que/aparcaron[aparcaran]/el coche del gerente/(%) [y/]que/subieran/la escalera de madera Experimental Psychology 2004; Vol. 51(1): 59Ð71