Relative clause attachment in German, English, Spanish and French: Effects of position and length

Available online at www.sciencedirect.com

ScienceDirect Lingua 166 (2015) 43--64 www.elsevier.com/locate/lingua

Relative clause attachment in German, English, Spanish and French: Effects of position and length Barbara Hemforth a,*, Susana Fernandez b, Charles Clifton Jr.c, Lyn Frazier d, Lars Konieczny e, Michael Walter e a

Laboratoire de Linguistique Formelle, CNRS, Université Paris Diderot, 5 rue Thomas Mann, F-75205 Paris Cedex 13, France b Facultad de Psicologia, Universidada Complutense de Madrid, Campus de Somosaguas, 28223 -- Pozuelo de Alarcon, Madrid, Spain c Department of Psychological and Brain Sciences, University of Massachusetts, Amherst, MA 01003, USA d Department of Linguistics, University of Massachusetts, Amherst, MA 01003, USA e Center for Cognitive Science, University of Freiburg, Friedrichstr. 50, D-79098 Freiburg, Germany Received 24 July 2014; received in revised form 18 August 2015; accepted 20 August 2015 Available online 29 September 2015

Abstract German, English, Spanish, and French versions of short (ia) and long (ib) relative clauses were tested in written questionnaire studies. The noun phrases containing the clauses appeared in either object position (as in i) or in subject/topic position, and the relative clauses ambiguously modified the first or the second noun of the noun phrase. (i).

a. b.

The doctor met the son of the colonel who died. The doctor met the son of the colonel who tragically died of a stroke.

In the first experiment, German, English and Spanish were tested. In all three languages, more high attachment (first noun modification) interpretations were observed for long relative clauses than for short ones, perhaps reflecting differences in implicit prosodic phrasing provided by participants when they read the questionnaire. Across languages, more high attachment was observed in object position than in subject/topic position, but this effect was larger for German and Spanish than for English. In addition, although more low attachment than high attachment was observed in every case except German object position, German showed more high attachment than Spanish or English. A second questionnaire indicated that the preference for high attachment in German object position relative clauses cannot be attributed to readers taking these relative clauses to be extraposed. In a third questionnaire in French, we found a generally higher preference for high attachments as in German, but no position effect, as observed in English. We provide an account of these data, which does not require any special theory of relative clause attachment or parameterization for languages. It requires only general processing principles together with independently required grammatical differences among the languages studied involving the focus properties of object vs. non-object position and the availability in Spanish and German, but not English and French, of separate positions for topical vs. nontopical subjects. © 2015 Published by Elsevier B.V.

Keywords: Relative clause attachment; Cross linguistic comparison; English; French; German; Spanish

* Corresponding author. Tel.: +33 01 57 27 57 85. E-mail addresses: [email protected] (B. Hemforth), [email protected] (S. Fernandez), [email protected] (C. Clifton Jr.), [email protected] (L. Frazier), [email protected] (L. Konieczny), [email protected] (M. Walter). http://dx.doi.org/10.1016/j.lingua.2015.08.010 0024-3841/© 2015 Published by Elsevier B.V.

44

B. Hemforth et al. / Lingua 166 (2015) 43--64

1. Introduction 1.1. RC attachment across languages Cuetos and Mitchell (1988) studied the processing of ambiguous relative clause (RC) attachment sentences like (1) in Spanish and English. (1)

The journalist interviewed the son of the colonel who was standing on the balcony.

They argued that Spanish readers prefer to interpret the relative clause who was standing on the balcony as modifying the head noun, son, of the noun phrase (NP) that contained it rather than modifying the second noun in the NP, colonel, while English readers prefer to interpret it as modifying the second noun. They took this observation as evidence against the crosslanguage universality of parsing principles such as late closure (Frazier, 1978, 1990) or recency (Gibson, 1991), which would seem to favor modification of the second noun (but cf. Frazier and Clifton, 1996). Since their pioneering study, a great number of studies have examined RC attachment ambiguities in a wide array of languages (e.g., Brysbaert and Mitchell, 1996; Carreiras and Clifton, 1993, 1999; Cuetos and Mitchell, 1988; De Vincenzi and Job, 1993; Gibson et al., 1996; Gilboy et al., 1995; Hemforth et al., 2000a,b; Hemforth et al., 2000; Mitchell et al., 1995; Mitchell and Brysbaert, 1998; Zagar et al., 1997). In addition to RC attachment studies focused on cross-language differences, various studies have examined the role of individual differences in working memory capacity in determining RC attachment (Swets et al., 2007; Traxler, 2007, 2009), or investigated the priming of RC attachment (Desmet and Declerc, 2006; Scheepers, 2003; Scheepers et al., 2011). Most research on sentence comprehension has searched for language-universal processes. The research on RC interpretation has focused instead on differences between languages, especially, the possibility that a preference to relate new material to recently perceived material (e.g., late closure; Frazier, 1978) is not universal across languages. Though the literature on RC attachment has addressed a great many interesting issues, firm answers have not been easy to establish. The literature is sufficiently rich and varied that one may find support for whatever position one favors, especially with respect to the basic question of whether cross-language differences exist and, if so, what underlies them. Answers to these basic questions have not been firmly established, we think, because there have been few studies where the materials are held constant across languages (e.g., Gilboy et al., 1995; Cuetos and Mitchell, 1988; Carreiras and Clifton, 1993, 1999, for the Spanish--English comparison). The need to hold the materials constant across languages is particularly acute because it is known that there is a great deal of RC attachment variation within languages, e.g., ranging between roughly 10% and 80% low attachment of RCs in Spanish in Gilboy et al. (1995) and ranging from roughly 20% to 70% low attachment in English for translation equivalents of the same materials, with the amount of low attachment depending on the type of head noun and the type of preposition in the head nominal. The goal of the present paper is to directly compare languages that have been claimed to differ on RC attachment, using materials that are comparable across languages (Spanish, English, German in Experiment 1; German with or without extraposition in Experiment 2, French in Experiment 3), and using items in the range where cross-language variation would be expected to emerge if indeed it exists. The present work addresses cross-language differences in the final interpretation of RCs and explicitly challenges the suggestion that there are fundamental differences among languages in how RCs are comprehended. It suggests that the differences that have been observed are actually reflections of deeper factors that are common to all languages that have been studied. We discuss both these points before reporting the experiments.1 1.2. The bases of interpretation preferences English has been claimed to show a low attachment (recency, late closure) preference for interpreting RCs. Spanish, Dutch and German (and to a possibly even greater extent, French) have been claimed to show a high attachment

1

The present studies are off-line judgment studies, like most studies of relative clause attachment. There are rather few on-line studies, and most confirm the off-line data (e.g., Carreiras and Clifton, 1993, 1999; Carreiras et al., 2004). The initial report of a preference for high attachments of RCs in Spanish (Cuetos and Mitchell, 1988) contained three on-line self-paced reading studies, but these studies did not directly compare high vs. low attachments (see also Traxler et al., 1998). Rather, they compared ambiguous and unambiguous attachments, finding that ambiguous attachments were faster. Later research, however, has shown faster reading of RC sentences that are disambiguated in favor of the off-line preference for the language in which the study was conducted (including, for Spanish and English, Carreiras and Clifton, 1993, 1999, and at least suggestively, Thornton et al., 1999; for German, Konieczny and Hemforth, 2000; Hemforth et al., 2000a, 2000b; for Dutch et al., 1996, and Desmet et al., 2002a; for French et al., 1997). One arguable exception appears in De Vincenzi and Job (1993, 1995), who found an early reading time penalty in Italian for RC sentences that were disambiguated toward high attachment, followed by a later penalty for low attachment sentences, but in general, off-line data do seem to reflect initial parsing preferences as reflected in on-line data.

B. Hemforth et al. / Lingua 166 (2015) 43--64

45

preference. One line of explanation for these apparent differences emphasizes the differences among languages. This approach began with the denial that a preference for recency is universal across all languages (Cuetos and Mitchell, 1988) and has developed into various ways of accounting for presumed language-specific preferences. Some accounts have appealed to language statistics, suggesting that languages in which one attachment is more frequent favor such an attachment (Cuetos and Mitchell, 1988). Others have appealed to language differences in the weighting of opposing constraints favoring recency vs. high attachment (Gibson et al., 1996). Still others have attempted to identify underlying grammatical differences, e.g., properties that the relative pronoun has in a given language (Hemforth et al., 2000a, who treat relative pronouns with overt number and case features as part of a binding system and those without as attachment). A second approach emphasizes the similarities in process across all languages and accounts for the apparent differences between languages by appealing to sometimes-subtle differences in how they deploy the same psycholinguistic tools. Some instances of this approach argue that the variability across languages is small compared to the variability observed within each language. As noted above, Gilboy et al. (1995) observed that RC attachment tendencies (high vs. low) vary rather greatly across different RCs in a single language, reflecting pragmatic factors, argument structure, focus structure, etc. (cf. also, Frazier and Clifton, 1996; Schafer et al., 1996). Grillo and Costa (2014) propose that languages differ in terms of whether they permit a ‘‘pseudo-relative’’ structure, and that the availability of such a structure introduces ambiguity, affecting the apparent relative preference for high vs. low attachment. A third instance of this approach appears in Fodor (1998), who attributes differences in RC attachment preferences within and across languages to the properties of individual sentences. Fodor proposed the Balanced Sister hypothesis, which specifies that constituents (RCs, for example) prefer to attach to sisters of the same (possibly prosodic) weight or length. Given a short object of a preposition (e.g., the daughter of the king) and a long RC (e.g., who wore excessive amounts of silk clothing), the RC will prefer to attach high to achieve prosodic balance. A short RC (e.g., who drank) might preferably attach low to achieve balance. Fodor (2002) put forward a more general hypothesis, the ‘Implicit Prosody hypothesis,’ which claims readers in a language impose the default prosody of a structure in that language during silent reading.2 In the present paper, we address three distinct questions: (1) do languages have fundamental differences in their general tendency to favor high vs. low relative clause attachment, or do any apparent differences result from other, more basic, properties of their grammar; (2) do longer relative clauses tend to attach high, and if so, is this true across different languages; and (3) do relative clauses modifying subject (or topic) NPs show different attachment preferences from relative clauses modifying object NPs (where they have been studied in nearly all examinations of relative clause attachment), and are these differences stable across languages. In the present paper, we addressed these questions in German, English, and Spanish, as well as French in a separate study. We examine whether the claimed greater preference for high attachment in German, French, and Spanish than in English holds true generally, or is limited to certain syntactic positions or prosodic lengths. 2. Experiment 1 We report a written questionnaire experiment using very similar sentences in German, English and Spanish. The questionnaire focused primarily on evaluating Fodor’s prosodic account, and assumes that readers impose a prosodic structure on sentences they read. In addition, it explored the possible effects of having the noun phrase (NP) modified by the RC appear in pre-verbal vs. post-verbal position. Each sentence in the experiment appeared in four forms, illustrated in (2) in English. (2)

a. b. c. d.

The The The The

son of son of doctor doctor

the colonel who died had written five books on tropical diseases. the colonel who tragically died of a stroke had written five books on tropical diseases. met the son of the colonel who died. met the son of the colonel who tragically died of a stroke.

Two forms (2a, c) contained a short RC (e.g., who died) and two (2b, d) contained longer RCs (e.g., who tragically died of a stroke.) Two forms (2a, b) appeared in non-object position (which we will analyze as being subject position in English and subject/topic position in Spanish, initial ‘topic’ position preceding the second-position verb in German; see discussion below.). The c--d forms contained a RC in object position in all three languages. One central purpose of the study was to evaluate Fodor’s Balanced Sister Hypothesis. This hypothesis predicts that the longer RCs will result in more high attachments in all three languages. The study was also designed to contrast the

2 We are using the term ‘balance’ as it was introduced in Fodor (1998), who described balanced structures as ones ‘‘in which sister constituents are roughly equal in prosodic weight (a function of length and also stress, etc.)’’ (p. 302).

46

B. Hemforth et al. / Lingua 166 (2015) 43--64

nature of RC attachment when the noun phrase that the RC modifies appears in object position (which has usually been the case in studies of RC attachment) and when it appears in non-object position. In all three languages, the ‘nuclearscope’ -- the position where asserted or focused information is expected to occur -- includes the object but not the (preverbal) subject or initial position (see Carlson et al., 2009, and references therein). The latter is typically associated with topical and often presumably given information. Schafer et al. (1996) have presented evidence that using a pitch accent to focus one of the nouns in a NP of NP RC phase attracts attachment of the RC to the focused noun. Presenting the NP of NP RC phrase in object position will, we assume, by default impart broad focus to the entire phrase (see Selkirk, 1984, for discussion of broad vs. narrow focus and how the latter can be conveyed by marked pitch accents on individual parts of a phrase). We suggest that broad focus on a phrase entails that the head of the phrase -- the initial noun in our materials -- is focused, and thus (following Schafer et al., 1996) should attract RC attachment. Since a NP of NP RC phrase in non-object position does not receive focus by default, neither of its nouns should receive focus. Thus the tendency for high attachment of the RC (attachment to the first noun) expected in object position should be less in nonobject position. Prompted by an anonymous reviewer, we collected data to evaluate the suggestion that the head of a focused phrase is generally focused. We had 16 Mechanical Turk (https://requester.mturk.com/) ‘workers’ choose which of two continuations of a sentence fragment they preferred (see Appendix A for experimental details). The sentence fragments were like (3a), and contained the focus particle only or even to explicitly focus the NP-PP phrase. The two continuations were like (3b). (3)

a. b.

The doctor met only the daughter of the colonel (1) and not the colonel’s son. (2) and not the general’s daughter.

The continuations differed in terms of which word in the sentence fragment they contrasted with. If the first noun in the sentence fragment carries the focus (as we assume), the preferred continuation should be the one that contrasts with it ((3b1, son). The respondents chose this continuation 83% of the time (SE = 0.02%), supporting our assumption that placing the entire modified noun phrase in object position would result in focus being placed on its head. Not all accounts predict a clear difference between modifiers of subject vs. object phrases. For instance, Gibson et al.’s (1996) Predicate Proximity Principle states that a relative clause will preferentially be attached to the phrase that is structurally closer to the head of the predicate (the verb). Both the head of a NP-of-NP subject, and the head of a NP-of-NP object, are structurally closer to the head of the predicate than the noun in the of-NP prepositional phrase is, and each should be favored as an attachment site (modulo the other factor, recency, posited by Gibson et al.). The final question is whether these factors interact with each other and/or with the language tested. 2.1. Methods Materials. Thirty-two sentences were constructed whose translations in English, German, and Spanish were very similar in meaning and acceptability (as judged by the experimenters). Each sentence contained a complex NP of the form NP1 of NP2, followed by a RC modifier, which could ambiguously modify the first or the second NP of the complex NP. The RC was chosen to be plausible both as a modifier of the first and the second NP in the judgment of the experimenters (a judgment which was checked in plausibility norms to be reported below). Twenty-five of the English RCs were in simple past tense, and the remainder in present tense; all matrix predicates were in past tense. For the Spanish version, the RCs of 10 sentences were in the preterit, 20 in the imperfect past, and two in present tense. All matrix predicates were in past tense: 22 in the preterit and the rest in the imperfect past. In the German version, 25 items had a past tense RC and 7 a present tense; all main clauses were past tense. Four versions of each sentence were constructed. In two, the modified NP appeared in preverbal position (topic position in German, subject position in English and subject/topic position in Spanish). In the other two, the NP appeared in object position. Crossed with this preverbal/postverbal factor, the RC was either short or long. The short version averaged two and a half syllables in English, five syllables in Spanish, and three syllables in German. The long version averaged nine, fourteen, and eight syllables in English, Spanish, and German, respectively. One full set of sentences appears in (4) through (6); all sentences used appear in Appendix B. (4)

English Subject RC/Short: The son of the colonel who died wrote five books on tropical disease. Subject RC/Long: The son of the colonel who tragically died of a stroke wrote five books on tropical disease. Object RC/Short: The doctor met the son of the colonel who died. Object RC/Long: The doctor met the son of the colonel who tragically died of a stroke.

B. Hemforth et al. / Lingua 166 (2015) 43--64

(5)

(6)

47

German Subject RC/Short: Der Sohn des Majors, der starb, schrieb fünf Bücher über tropische Krankheiten. Subject RC/Long: Der Sohn des Majors, der tragischerweise an einem Schlaganfall starb, schrieb fünf Bücher über tropische Krankheiten. Object RC/Short: Der Arzt traf den Sohn des Majors, der starb. Object RC/Long: Der Arzt traf den Sohn des Majors, der tragischerweise an einem Schlaganfall starb. Spanish Subject RC/Short: El hijo del coronel que murio´ escribio´ cinco libros sobre enfermedades tropicales. Subject RC/Long: El hijo del coronel que trágicamente murio´ de apoplejía escribio´ cinco libros sobre enfermedades tropicales. Object RC/Short: El doctor conocio´ al hijo del coronel que murio´. Object RC/Long: El doctor conocio´ al hijo del coronel que trágicamente murio´ de apoplejía.

Four counterbalanced forms of a questionnaire were constructed. One quarter of the 32 experimental sentences appeared in each version in each form of the questionnaire, and across the four forms, each experimental sentence appeared once in each version. Each sentence was followed by a sentence fragment, which participants were supposed to complete to indicate their understanding of the sentence. These 32 sentences were combined with 64 sentences of various forms (including ambiguous and unambiguous pronoun interpretation, interpretation of prepositional phrases as modifying a verb or a noun, and simple unambiguous predications). One randomization was made of each form. The same procedures, and same randomization, were used in each language. Participants. Forty-eight undergraduate students at the University of Massachusetts completed the English questionnaire, receiving extra course credit for their participation. Forty-eight undergraduate students at the Complutense University of Madrid, who also received course credit, completed the Spanish version. Thirty-two native German speakers at the University of Freiburg responded to the German version. The English- and German-speaking participants were tested individually or in groups of up to five people. The Spanish participants were tested in a single group. Procedures. The questionnaire began with written instructions in the language of the questionnaire. The instructions indicated that the participant’s task was to read a sentence and then indicate his or her understanding of the sentence by completing a following sentence that contained a blank. For example, participants were told that the complete sentence might be ‘‘The boss of the woman who had a long gray beard was on vacation’’ and following that, they would see ‘‘The _______________________ had a long gray beard.’’ They were to fill in the blank with one word (‘‘boss’’ in this example) that indicated the meaning of the initial sentence. Each participant then completed one counterbalanced form of the questionnaire. In each language, equal numbers of participants were assigned to each form of the questionnaire. 2.2. Results Table 1 contains the mean proportions with which the RC was interpreted as modifying the first NP of the complex NP. Overall, high attachments (N1 modification) were less frequent than low attachments (N2 modification), 43.25 N1 vs. 57.75% N2. High attachment interpretations were more frequent for long than short RCs, 48% vs. 38.5%, F1(1,125) = 48.7; F2(1,93) = 24.6, p < .01. High attachments were also more frequent when the modified NP was in object position than when it was in non-object position, 48% vs. 39%, F1(1,125) = 24.3; F2(1,93) = 22.1, p < .001. The interaction between these two factors approached but did not reach significance (F1(1,125) = 3.49, p = .064; F2(1,93) = 2.794, p = .098). There was a tendency toward a larger length effect for NPs in object than in non-object position, as can be seen in Table 2. However, even the smaller length effect for NPs in subject position was significant F1(1,125) = 12.65, p < .001; F2(1,93) = 7.46, p < .01).

Table 1 Percentages of high attachment choices, Experiments 1 and 3 (with by-subject standard errors). Language

Subject RC Short

Subject RC Long

Object RC Short

Object RC Long

German Spanish English Average French

45 29 36 36 51

48 39 42 42 64

55 41 33 41 53

62 55 48 54 62

(5.7) (3.6) (4.4) (4.7)

(5.7) (4.6) (4.9) (4.26)

(4.4) (3.9) (3.8) (4.2)

(4.3) (4.1) (4.2) (3.9)

Note: German, Spanish, and English were studied together in Experiment 1; French separately, in Experiment 3, and therefore is not included in the Experiment 1 average.

48

B. Hemforth et al. / Lingua 166 (2015) 43--64 Table 2 Effect of RC length (percentages of high attachment, long--short RC), Experiment 1. Language

Subject RC

Object RC

German Spanish English Average

4 10 6 7

7 14 15 12

There was an overall difference between languages, significant by items but marginal by subjects. German had a higher frequency of high attachment interpretations than Spanish and English, though only marginally so across subjects (52%, 41%, and 40%, respectively; F1(2,125) = 2.89, p < .06; F2(2,93) = 12.49, p < .001). There was one significant interaction among languages: Object vs. non-object position of the modified NP interacted with language, F1(2,125) = 4.88, p < .01; F2(2,93) = 4.17, p < .02). The effect of position was significant in Spanish (14% more high attachments in object than in non-object position, F1(1,47) = 24.95, p < .001; F2(1,31) = 14,23; p < .001) and in German (12%, F1 (1,31 = 8.94, p < .01; F2(1,31) = 8.75, p < .01) but not in English (1%, F1(1,47) < 1; F2(1,31) < 1). Interactions involving language and length of the RC were nonsignificant. 2.3. Discussion In all three languages, longer RCs showed more high attachment than shorter ones. This may be explained by Fodor’s Balanced Sister hypothesis, which appeals directly to a tendency to attach a phrase as a sister to a phrase of similar size. This may be a primitive tendency, or it may be based on a tendency for a reader to impose an implicit prosodic break before a long RC. Assuming that implicit prosody mirrors overt prosody, the latter proposal fits with Watson and Gibson’s (2004) finding in an elicited production task that the probability of an intonational boundary increases with the length of the upcoming constituent as well as the length of the preceding constituent, which was not systematically manipulated in the present experiments (see Ferreira, 1993; Gee and Grosjean, 1983, for related analyses). The proposal is also supported by oral production evidence presented by Fernández (2005), who had Spanish and English monolinguals pronounce sentences derived from those used in the present paper under conditions that disambiguated them to modification of the second NP. Fernandez showed that the final noun preceding the relative clause was longer in duration when the relative clause was longer, for both Spanish and English (with a larger effect when the modified NP was in object than in subject position). This is evidence for the existence of more frequent prosodic boundaries before longer relative clauses.3 A prosodic boundary immediately before the RC should favor its high attachment if the break is larger than any other prosodic break after N1 (Carlson et al., 2001; Clifton et al., 2002), and evidence exists that an actual prosodic boundary before a relative clause promotes its high attachment (Maynell, 1999, cited in Fodor, 2002). The present data do not allow us to decide between these two closely related alternatives, the balanced sister hypothesis and the implicit prosody hypothesis, but they do firmly establish that the length of a constituent affects its interpretation. Before providing a more complete interpretation of the data that have been presented, we will examine two possible accounts for some of the data. A reviewer of this paper suggested that some of the materials may confound plausibility with high vs. low attachment. Our argument does not crucially depend on having equated high and low attachment in terms of plausibility. Our goal is not to establish whether or not there is an overall high vs. low attachment preference, but to investigate how factors such as RC length and NP position affect attachment preferences. It is important, however, to establish that plausibility of N1 vs. N2 modification is not systematically affected by RC length, and accordingly, we conducted a post hoc plausibility norming study. A second possible account of one aspect of the data, the greater frequency of high attachments for German than for Spanish or English, appeals to a potential structural difference between the languages, namely, that the German relative clauses in object position may be extraposed. We conducted Experiment 2 to test this possibility.

3 Jun (2010) showed that English oral readers were more likely to place a prosodic boundary after the second noun than after the first noun in the NP of NP RC construction, which she took to be evidence against the implicit prosody hypothesis, given her assumption that English favors low attachment and that low attachment would be associated with the absence of a prosodic boundary. However, we doubt the former assumption, and we note that she also reported that longer relative clauses did result in generally larger prosodic boundaries after the second noun.

B. Hemforth et al. / Lingua 166 (2015) 43--64

49

Table 3 Mean plausibility ratings, normative study English. Length

N1 modification proposition

N2 modification proposition

Short Long

4.00 3.68

4.26 3.84

3. Normative study 3.1. Methods Each of the 32 English RC items of Experiment 1 was rewritten in four versions to be rated for plausibility, as shown in (7): (7)

a. b. c. d.

The The The The

colonel’s son died. colonel’s son tragically died of a stroke. colonel died. colonel tragically died of a stroke.

(7a, b) presented the proposition underlying an N1 interpretation of the RC, and (7c, d) presented the proposition underlying an N2 interpretation. (7a, c) contained the information in short RCs, and (7b, d) the information in long RCs. The resulting sentences were assigned to four counterbalanced questionnaires, one version of each sentence to each form, and were randomly ordered in each questionnaire with 16 foil sentences of moderate to high implausibility (e.g., The salesclerk refused the money, The jockey’s horse came out the starting gate backwards). Each item was followed by a five-point plausibility rating scale, where ‘‘1’’ was defined as ‘‘very unlikely, very implausible, perhaps even impossible’’ and ‘‘5’’ as ‘‘perfectly plausible, unsurprising, a very everyday occurrence.’’ Forty-four University of Massachusetts undergraduates individually completed the questionnaire, 11 completing each of the four forms (another 8 were tested but replaced because they consistently assigned ratings of 1 or 2 to the highly plausible foils). 3.2. Results and discussion The mean plausibility ratings for the experimental items appear in Table 3. Items formed from N2 modifying RCs were rated as more plausible than items formed from N1 modifying RCs, 4.05 vs. 3.84 on the 5-point scale (F1(1,43) = 19.18, p < .001; F2(1,31) = 6.26, p < .02). Further, short items were rated as more plausible than long items, 4.13 vs. 3.76 (F1 (1,43) = 19.74, p < .001; F2(1,31) = 15.38, p < .001). However, there was no interaction between the two factors (F < 1). Although the overall preference for N2 modification in Experiment 1 could be attributed to the plausibility difference observed in the normative study, we consider this of little consequence. We do not propose that any conclusions of interest turn on an absolute preference for N1 vs. N2 modification. A similar conclusion holds for the length difference: no effects in Experiment 1 could be attributed to the greater plausibility of short items (which we assume is simply due to the greater opportunity for long items to contain some information that strikes some raters as odd). The important result is that any effect of length on plausibility was similar for N1 and N2 modification items. Thus, the finding that added relative clause length increased the frequency of N1 interpretations cannot be attributed to differences in plausibility.4 However, since it may be possible that differences in plausibility have a somewhat indirect influence on attachment preferences, we ran a post hoc analysis of the data presented in Experiment 1, eliminating all items that the norming study suggested may have been biased. For 19 of the 32 items, length was not confounded with plausibility according to the norming study (F’s < 1). Table 4 shows the mean number of high attachments along with mean plausibility ratings (in parentheses) for the four experimental conditions. Due to the elimination of items, the design was not balanced across subjects. Therefore, we only ran an analysis across items. The length effect still showed up reliably (F(1,56) = 13.09, p < 0.001) and the position effect marginally (F(1,56) = 3.12, p < 0.09). The effect of RC length on interpretation preferences observed in Experiment 1 cannot be attributed to differences in plausibility.

4 A fully parallel norming study on German (32 participants) showed a highly similar pattern (N1-short: 3.93, N1-long: 3.78, N2-short: 4.27, N2-long: 3.98) with shorter items being judged more plausible than longer items (F(1,124) = 4.26; p < .05), items constructed from N2-modifying RCs being judged more plausible than items constructed from N1-modifying RCs (F(1,124) = 8.14, p < .01), and no interaction (F < 1).

50

B. Hemforth et al. / Lingua 166 (2015) 43--64 Table 4 Percentage of high attachment choices for items with normed plausibility. Length

Mean plausibility ratings in parentheses Subject position

Object position

Short Long

35 (3.85) 43 (3.75)

37 (3.99) 48 (3.92)

4. Experiment 2 Recall that there was a stronger high attachment preference in German than in Spanish or English (only marginally significant across subjects). This difference could in principle be due to the fact that German RCs following the object could be taken as being extraposed. Note that while sentences like those we used are common (e.g., from (5), Der Arzt traf den Sohn des Majors, der starb.) it is perfectly acceptable to have an auxiliary verb following the first NP and a clause-final main verb, followed by the relative clause (Der Arzt hat den Sohn des Majors getroffen, der starb.). The relative clause is unambiguously extraposed in this case. While extraposition is not overtly signaled in the earlier example (from (5)), readers may have taken the RC to be extraposed, based on the explicitly extraposed model. Within the Construal theory presented by Frazier and Clifton (1996), extraposed RCs are predicted to preferentially attach high. The current thematic domain for an extraposed RC is the whole sentence and a RC is preferentially construed with a host within that domain that is relevant to the main assertion of the sentence, i.e. N1 in two-site constructions. Similar predictions are made by the attachment-binding account of Hemforth et al. (2000b). If German RCs in object position are sometimes taken to be extraposed, this could lead to their frequent high attachment interpretations.5 Experiment 2 was designed to evaluate this explanation. 4.1. Methods Thirty-two German item sets were derived from those in Experiment 1. Each item contained two versions of sentences with unambiguous extraposition over the main verb and two versions of object RC sentences taken directly from Experiment 1. The four versions of one item can be seen in (8), together with their literal English translations. Forty native German subjects participated in the experiment. The procedure was the same as in Experiment 1. (8)

a. b.

c. d.

Der Arzt hat den Sohn des Majors getroffen, der starb. (Extraposed RC, short) The doctor has the son of the colonel met, who died. Der Arzt hat den Sohn des Majors getroffen, der tragischerweise an einem Schlaganfall starb. (Extraposed RC, long) The doctor has the son of the colonel met, who tragically died of a stroke. Der Arzt traf den Sohn des Majors, der starb. (RC follows object NP, short RC) The doctor met the son of the colonel who died. Der Arzt traf den Sohn des Majors, der tragischerweise an einem Schlaganfall starb. (RC follows object NP, long RC) The doctor met the son of the colonel who tragically died of a stroke.

4.2. Results Table 5 presents the mean percentages of high attachment (N1) interpretations. Participants never considered the subject as the attachment site in this experiment, corroborating the unacceptability of this option. Analyses of variance

5 Extraposed RC sentences can also be attached to non-topicalized subjects in German (Kiss, 2003, 2005). Attaching the extraposed RC to the postverbal subject in (i) is acceptable for most German speakers.

i. Den Major hat die Ärztinfem getroffen, diefem tragischerweise an einem Schlaganfall starb. The colonelobj has the doctorsub/fem met whofem tragically died of a stroke. The doctorfem met colonel whofem tragically died of a stroke. Attaching the extraposed RC sentence to a topicalized subject without contrastive marking as in (8a,b) is, however, inacceptable.

B. Hemforth et al. / Lingua 166 (2015) 43--64

51

Table 5 Percentages of high attachment choice, Experiment 2 (German) (with by subject standard errors). Length

Non-extraposed

Extraposed

Short Long

53 (4.05) 63 (3.91)

75 (2.26) 74 (1.99)

showed a clear effect of extraposition. The number of N1-attachments was significantly increased when the RC was extraposed (F1(1,39) = 19.82, p < .001; F2(1,31) = 39.60; p < 0.001). Length was marginally significant across subjects (F1(1,39) = 3.86, p < 0,06) though not across items (F2(1,31) = 2.21; ns). The interaction between length and position was not significant (F1(1,39 = 2.73, p < .11; F2(1,23) = 0.92). However, there was a numerical length effect when the RC was not extraposed, as observed in Experiment 1, which was, however, only significant across subjects (F1(1,39) = 4.87, p < 0.04; but F2(1,31) = 2.50; ns), and in fact the size of the length effect (63 vs. 53% high attachment for long vs. short) was closely comparable to that observed in Experiment 1 (62 vs. 55%). Preferences for extraposed RCs did not differ reliably across subjects or items (F1, F2 < 0.5) for short and long RCs. 4.3. Discussion Experiment 2 showed that extraposed RCs exhibit a different pattern of attachment preferences than non-extraposed clause final RCs, namely a much stronger preference for N1-attachments. However, extraposed RCs showed no effect of length. The object RCs that were not unambiguously extraposed (8c, d) behaved much like the ones from Experiment 1, with a tendency for a length effect. The absence of any suggestion of a length effect in the unambiguously extraposed items (8a, b) leads us to conclude that the preference for high attachment of German RCs in object position is not due to their being taken as extraposed. We therefore dismiss the possibility that the German RCs but not the RCs in the other languages were treated as being extraposed in Experiment 1. 5. Experiment 3 In Experiment 1, we reported language specific differences of the effect of position on attachment preferences. For Spanish and German, there were more high attachments when N1 appeared in object position than when it appeared in subject position, whereas no such difference was found for English. We hypothesize that the nature of the subject/topic position is what’s important. English subject position differs from Spanish pre-verbal subject position because in English any type of subject must occur in subject position, whether it is given/topical or new, and it differs from German where constituents in initial position tend to be topical. If this property of expressing subjects in the same position whether they are given/topical or new is what’s important, then French should behave like English with respect to syntactic position.6 To test this hypothesis, we conducted Experiment 3, a parallel study to Experiment 1, but testing French materials. French is thought to pattern with German in that a general preference for high attachments is mostly reported in the literature (e.g., Zagar et al., 1997). However, syntactically, it patterns with English in that there is not much freedom in word order, and there is a specified position for the subject. 5.1. Methods Materials. The French materials were close translations of the English, German, and Spanish questionnaires. Seven of the French relative clauses were in the passé simple, 19 in the imperfect past, and six in present tense. Of the matrix predicates, 14 were in the passé simple, one in present perfect (passé compose), one in the present tense and the rest in the imperfect past. The short version averaged three and a half syllables; the long version averaged eleven (assuming standard French). One full set of sentences appears in (9). All sentences are given in Appendix B.

6 Post-verbal subjects exist in Modern French although their use is highly restricted (as opposed to Medieval French, see Kaiser and Zimmermann, 2011). They are particularly rare in declarative root clauses where postverbal subjects are restricted to intransitives in highly particular syntactic, semantic, and pragmatic contexts (Lahousse, 2006; Marandin, 2003). They are impossible in the constructions under investigation in this paper, where given and new subjects will have to appear pre-verbally.

52

(9)

B. Hemforth et al. / Lingua 166 (2015) 43--64

Subject RC/Short: Le fils du commandant qui mourut écrivit cinq livres sur des maladies tropicales. Subject RC/Long: Le fils du commandant qui mourut tragiquement d’une attaque d’apoplexie écrivit cinq livres sur des maladies tropicales. Object RC/Short: Le médecin rencontra le fils du commandant qui mourut. Object RC/Long: Le médecin rencontra le fils du commandant qui mourut tragiquement d’une attaque d’apoplexie.

Participants and procedure. Forty undergraduate students at the Université de Provence completed the French questionnaire. 17 of the participants were tested individually, 23 participants were tested in a single group. Otherwise, the procedure was identical to Experiment 1. 5.2. Results and discussion Table 1 (bottom row) shows the mean percentages of high (N1) attachments for the four experimental conditions. The only effect that showed up reliably was the length effect: longer relative clauses were attached high reliably more often than short relative clauses (F1(1,39) = 18.12, p < 0.001; F2(1,31) = 9.03, p < 0.01), replicating the length effect established for French by Pynte and Colonna (2000). There was no hint of a position effect (F1, F2 < 1) and no interaction (F1, F2 < 1.5). These results show that, although French patterns with German with respect to the number of high attachments, it patterns with English with respect to the lack of a position effect. 6. General discussion We return to the three questions we raised in the Introduction: (1) do languages have fundamental differences in their general tendency to favor high vs. low relative clause attachment, or do any apparent differences result from other, more basic, properties of their grammar; (2) do longer relative clauses tend to attach high, and if so, is this true across different languages; and (3) do relative clauses modifying subject (or topic) NPs behave like relative clauses modifying object NPs (where they have been studied in nearly all examinations of relative clause attachment), and are there differences across languages in any effect of syntactic position. The answers to these questions can help us decide whether we can maintain a strongly universal view of the parsing mechanisms in which the parsing preferences can be derived from those mechanisms together with the grammar of the language being parsed. 6.1. ‘Baseline’ attachment preference differences across languages A substantial literature dating to Cuetos and Mitchell (1988) suggests that languages differ in how their users prefer to interpret RCs in NP of NP RC constructions. We do not think it is accurate or revealing to classify languages as N1 (‘high’ attachment) or N2 (‘low’ attachment) languages. In the present study, as well as earlier studies (e.g., Gilboy et al., 1995; De Vincenzi and Job, 1993; Wijnen, 2001) so-called high attachment languages may show a majority of low attachment responses, depending on the particular content and structure of the materials being tested.7 Nevertheless, individual languages do differ in their grammatical properties in ways that can influence attachment preferences under certain circumstances. These include case assignment properties, flexibility of word order, prosodic properties, and probably more. In the present study, German and French showed more high attachment than Spanish or English. Why? It is possible that the German tendency for high attachment is due to the overt case features found on the relative pronoun in German, but not in English or Spanish. Hemforth et al. (2000a) suggest that case-marked relative pronouns are interpreted following binding principles, while non-case-marked relative pronouns are interpreted following attachment principles (e.g., Late Closure). However, claiming that basically different interpretive processes operate for German vs. Spanish or English RCs fails to explain why length of the RC and the position of the modified NP have such similar effects across languages (at least, across Spanish and German). Further, the basic idea that case- and number-marked relative pronouns are treated like other (e.g., personal) pronouns suffers from an underlying tension: personal pronouns particularly prefer topical antecedents but topics resist modification such as modification by a RC.

7 We reiterate that we do not claim that any conclusions turn on whether a particular language shows over, or under, 50% high attachment choices The preference is affected by a wide variety of factors (Gilboy et al., 1995), including plausibility -- which, in the present data, may have affected the overall baseline attachment preference. Our conclusions rest on the similarities and differences among languages in how length and syntactic position affect interpretation.

B. Hemforth et al. / Lingua 166 (2015) 43--64

53

An alternative explanation for the baseline differences is suggested by Gibson et al. (1996). They propose that languages with flexible word order (e.g., Spanish and German) show more high attachment than languages with rigid word order (English) because arguments must be kept available longer before they may be integrated in flexible order languages. Languages are presumably parameterized, with a strong high attachment tendency (perhaps due to greater weight given to a Predicate Proximity principle) in languages that normally require longer distance integration. However, this proposal does not appear to explain the difference between German and Spanish (except in an ad hoc fashion) and does not address the effect that RC length has on all languages.8 One might ask why the present experiment did not obtain the often-reported difference between Spanish and English, with Spanish showing an N1 attachment preference. We submit that this preference may actually be very limited. Gilboy et al. (1995) found a substantial language difference only for their ‘‘kinship’’ items (e.g., the daughter of the colonel. . .), and not for NP of NP items with different semantic relationships between the two NPs (including occupational relationships, possessives, and picture-phrases, among others). In the present study, the only appearance of a substantially greater N1 preference for Spanish than for English is for long RCs in object position. As mentioned before, previous experiments have typically studied modified NPs in object position, not in subject position, and they have not controlled or manipulated the length of the RC (see also Mitchell and Cuetos, 1991; Gibson et al., 1999). Following Gilboy et al., we suggest that the presumed N1 attachment preference for Spanish may hold for only a limited class of NPs and may not be a general property of the language. We conclude this section by noting briefly that the full pattern of data presented here is a challenge for an appeal to frequency of occurrence of N1 vs. N2 interpretations (e.g., Mitchell et al., 1995). To account for the effects of length and position, a frequency theorist would have to separately count occurrences of long vs. short RCs, in subject vs. in object position. At the very least, generality is lost. It is, to be sure, possible that language users could represent fine-grained differences in frequency. Desmet et al. (2002b) showed that fine-grained differences in corpus frequencies are correlated with attachment preferences due to the animacy of N1 and N2. However, even if it were found that (e.g.) short RCs in subject position are less frequently used to modify N1 than are long RCs in object position, we do not think that this constitutes a satisfying explanation of the preferences we have observed. Rather, we submit that the observed frequencies are data themselves in need of explanation and we propose that principles like the ones we advance here are good candidates for such an explanation. Grillo and Costa (2014) propose, as we do, that apparent differences among languages in relative clause attachment do not reflect language-specific parameterization for relative clauses, but instead reflect other, deep-seated, grammatical differences among the languages. They specifically proposed that genuine relative clauses always attach low, but that the appearance of high attachment is possible due to the availability of Pseudo Relative structures in some languages, like Spanish, Italian and French, but not English or German. On their analysis of Pseudo Relatives in Spanish, the Pseudo Relative structure is a small clause complement possible with a delimited class of verbs such as perception verbs. The small clause attaches to VP, and may only take N1 as subject. Unlike true restrictive RCs, a Pseudo Relative may take a proper name as subject, as illustrated below in (10), the verb in the small clause must be imperfective, and it must agree in tense with the main clause verb. (10)

He visto a Juan que corría. ‘I saw John that ran.’

While we accept Grillo and Costa’s general approach, and find their specific claims very interesting and attractive, we have to note that the Pseudo Relative hypothesis fails to account for the present data. First, German does not contain Pseudo Relatives and therefore it should behave like English, rather than showing slightly more high attachments and an effect of position comparable in size to the Spanish effect. Second, French like Spanish, permits the Pseudo Relative structure, but in terms of the absence of a position effect (if not baseline attachment preferences), French behaves more like English than like Spanish. Finally, the effects of length of a relative clause modifying an object were very similar for

8

A third, plausible if superficial, account of the greater preference for N1 modification in German than in Spanish or English appeals to the convention in German to introduce a comma between any RC and its host. Hemforth et al. (2000a) raise this possibility and the associated possibility of a prosodic difference in discussing the difference between RCs and prepositional phrases as postnominal modifiers, and Schimke et al. (2002) showed that the presence of a comma in French sentences increases the frequency of N1 interpretations. This prosodic explanation can also account for the increased number of high attachments in the French questionnaire: Since French has a highly regular prosody, French speakers may introduce a break before a relative clause more automatically than English speakers, resulting in an increased tendency to attach the RC high even without an explicit comma. However, recent investigation of German prosody suggests that native German speakers insert a pause before the RC for both low and high attachment, and that the pause is even longer in the forced low attachment sentences than in the forced high attachment sentences (Gryllia and Kügler, 2010). If pauses and prosodic boundaries do not correlate with high attachment in German, then it becomes unclear how the comma explanation would work.

54

B. Hemforth et al. / Lingua 166 (2015) 43--64

Spanish and for English, which is unexplained under the Grillo and Costa hypothesis: the availability of the Pseudo Relative structure, which can only take N1 as its subject, should not depend on its length. One very interesting recent study that convincingly showed high attachment for certain RCs in English tested main clauses with implicit causality verbs. Rohde et al. (2011) showed that RCs containing potential explanations for the state of affairs described in the main clause favored high attachment, where such an explanation would be afforded (e.g., contrast John detests/babysits the children of the musician who . . .; the main verb detests favors attachment to N1 when the content of the relative clause provides a reason for detesting the children). Although we find this work fascinating, it would not be expected to explain the present results, on the assumption that causal coherence relations are favored universally, and thus if there is a role for causal coherence in our materials at all, it should hold equally across the languages tested. 6.2. Length An effect of length was found in all languages tested: longer RCs led to more high attachments than shorter RCs, 7% on average in subject position and 12% in object position. This length effect is clearly predicted by the Balanced Sister hypothesis. It must however be noted that the length variation in our experiments always meant adding extra content. This increase in informational load may have played a role in addition to the well-known prosodic consequences of RC length (Fernández, 2005). Hemforth et al. (2013) disentangled effects of phonological length per se from effects due to extra content for PP-attachment ambiguities in French and German (e.g.;der Assistant des Notars aus Au/Au-Winterach/dem schönen Au; the assistant of the lawyer from Au/Au-Winterach/beautiful Au). Hemforth et al. created longer PPs by lengthening the city name (Au-Winterach) or by adding information about the city using a qualifying adjective (dem schönen Au/beautiful Au), keeping numbers of syllables constant across the longer conditions. In questionnaire experiments, similar to those presented in this paper, what they found was that it was the extra content that seemed to matter most, presumably because high information load should be included only when it serves a function (Almor, 1999) and justifying the presence of extra material is easier if it relates to the main assertion than if it modifies subordinated or less important material. We do not suggest that information based explanations of length will replace phonological accounts that have been reported repeatedly in the literature (e.g., Fernández, 2005). Adding content may however be part of the length effect. The experiments in this paper were not designed to disentangle phonological and informationbased hypotheses. 6.3. Language and position Overall, object position RCs showed more high attachment than subject/topic position RCs. This may be due to the fact that focus generally appears in object position, not subject position (except perhaps for contrastive focus, which was not warranted in the null contexts tested here). Focus on an NP can be interpreted as focus on the head of the NP, N1.9 It has been shown that focus attracts RCs (Schafer et al., 1996), and thus the greater tendency of RCs to modify N1 in object than in subject position may be attributed to an implicit focus effect. Hemforth and Konieczny (2002) report results from eyetracking and magnitude estimation experiments on German sentences, confirming a stronger N1 attachment preference for phrases in object than in topic position. What is perhaps most interesting is the significant interaction of position and language: the subject (topic) versus object difference was larger in German and Spanish than in English and French. We suggest that this difference reflects language differences in the properties of ‘subject’ position. German and Spanish provide distinct syntactic positions where subjects can appear, depending on whether they are topics; English and French do not. (By ‘topic’ we mean a constituent that is under discussion, and presumed to be given in the discourse). We will argue that in German and Spanish, a preverbal subject is more clearly a topic than it is in English or French. We assume that topics are generally treated as ‘old’ information and not focused (except in a context that supports a contrastive focus interpretation), that focused phrases attract modification while nonfocused phrases resist it, and that modification of an entire NP PP phrase appears as modification of its head, N1, and not simply of the head of the adjunct PP (see Appendix A for a confirmation of this latter assumption). Given these assumptions, we have an account of why the tendency to interpret a German or Spanish RC as modifying N1 of a subject NP is small compared to the tendency to interpret it as modifying N1 of an object NP, while syntactic

9 Consider a phrase like ‘‘The colonel on the porch.’’ If ‘‘on the porch’’ is said with a pitch accent, and thus is narrowly focused (since focus does not project from an adjunct to the head of the phrase containing it; Selkirk, 1984), the contrast set is colonels in various locations, not individuals other than colonels (i.e. ‘‘Mary hated only the colonel on the PORCH, not the lady with the poodle.’’ is awkward.) However, if the whole phrase is focused, as it would by default be when in object position, the contrast set is other individuals -- that is, alternatives to the referent of N1 -- not colonels in other locations. Our study presented in the introduction of Experiment 1 and in Appendix A, confirms these intuitions.

B. Hemforth et al. / Lingua 166 (2015) 43--64

55

position has little or no effect in English and French. German has a syntactic basis for the preverbal NP being taken as a topic. Consider (11), which sketches our assumed analysis of the fundamental syntax of a sentence in the three languages we studied: (11)

a. b. c. d.

English: [ Subject [ Verb [ Object ]]] (for given and new subject referents) Spanish: [ Subject [ Verb [ Object ]]] (primarily for given subject referents) French: [ Subject [ Verb [ Object ]]] (for given and new subject referents) German: [ Subjecti [ Verbj [ ti [ Object tj ]]]](mostly for topical/given subject referents)

Unlike Spanish, English and French, the German root clause structure must be analyzed as a CP (assuming one attributes a uniform syntactic structure for all German root clauses), not an IP, and the verb must raise to fill the head position of the CP. Some single constituent, often the subject and generally the topic, obligatorily raises to the position in front of the highest verb (see Besten, 1983, and for psycholinguistic evidence, Frazier and Flores d’Arcais, 1989).10 In our experimental sentences, the subject raises to first position, but in other sentences the raised constituent might be an adverb, the object, or whatever is topical. The clear conclusion is that the initial phrase in German is taken by a reader to be the topic. In Spanish, a subject may occur in either of two positions: a preverbal (Spec, IP) position, or a postverbal position, adjoined to VP. The former position is topical; the latter position is used for nontopical, focused phrases (see AlonsoOvalle et al., 2002; Pérez-Leroux and Glass, 1997, for evidence; see also Carminati, 2001, for extensive evidence in a closely related language, Italian). In English and French too, the highest subject tends to be topical. But unlike Spanish and German, all subjects in English and most subjects in French occur in Spec, IP and therefore the language does not allow a contrast between one syntactic position where topical subjects occur and a distinct position for nontopical ones. Thus we attribute the position by language interaction to the effects of topichood on RC attachment, together with the independently justified assumption that Spanish and German distinguish the position of topical and nontopical subjects. We assume that since the topical NP is given, neither it nor anything in it (head or modifier) will be focused, thus eliminating the tendency for the head of a focused NP to attract modification. 7. Conclusions Researchers have frequently claimed that there are differences among languages in the preferred interpretation of RCs in NP of NP RC constructions (e.g., Carreiras and Clifton, 1993, 1999; Cuetos and Mitchell, 1988; Mitchell and Brysbaert, 1998). However, the present results, together with our earlier studies (Gilboy et al., 1995; Hemforth et al., 2000a, 2000b) suggest that the between-language differences are limited in scope and that the similarities across languages are at least as worthy of attention. Longer RCs favor higher attachment, object position and focused antecedents are preferred hosts (in the present study and Schafer et al., 1996), and thematic-role assigning prepositions favor low attachment to their objects (Gilboy et al., 1995; Hemforth et al., 2000a, 2000b). We think the minor ‘baseline’ preference differences across languages can largely be explained by independent properties of specific languages and the way they affect the particular sentences tested, without any claims that languages follow different parsing principles or that they are differently parameterized. RC attachment is useful for uncovering language-specific properties that are important in processing precisely because RC attachment is so labile, and thus open to relatively minor factors not easily observed in simpler sentences, e.g., those with a single potential host for the RC. However, if our analysis is correct, there is nothing special about RC attachment. It is simply a window for observing general properties of language processing through the lens of language-specific properties that are independently warranted. No special theory of RC attachment or language parameterization is needed.

10

We recognize that our assumption that German root clauses are always CPs, and therefore involve movement of the subject in subject-initial root clauses, has been challenged. In influential work, Travis (1984, 1991) argued that German SVO structures should be analyzed as IP, not CP. She pointed out that SVO structures may have broad focus, unlike object-first structures, suggesting that SVO is basic. The other central argument concerned the distribution of the pronoun es. Both arguments have been countered. Subject before object structures may have broad focus in German independent of whether the subject precedes the verb or not, e.g., in Adverb--Verb--Subject--Object structures. As for the distribution of es, Schwartz and Vikner (1996) show that the distribution of es follows from independent properties of es (the fact that it is stressless and may not be an operator) and thus its distribution does not argue for a special (IP) structure for SVO sentences. Indeed, if SVO structures were IPs, one would expect Adv-SVO sentences and S-Adv-VO sentences to be grammatical, as in English. They are clearly ungrammatical in German, as expected if SVO sentences are verb-second CPs, as we assume here. See Schwartz and Vikner (1996) for additional arguments for the analysis we assume.

56

B. Hemforth et al. / Lingua 166 (2015) 43--64

Acknowledgments This research was supported in part by grant HD-18708 from the National Institutes of Health to the University of Massachusetts and by a public grant overseen by the French National Research Agency (ANR) as part of the program ‘‘Investissements d’Avenir’’ (reference: ANR-10-LABX-0083). We are grateful to Markus Bader and Ellen Brandner for helpful discussion of German syntax. We also want to thank Sarah Schimke and Saveria Colonna who helped us setting up the French materials. Correspondence can be addressed to Hemforth at barbara.hemforth@linguist. univ-paris-diderot.fr or Clifton at [email protected]. Appendix A. Methods and results of focus study Sixteen respondents were recruited using Amazon’s Mechanical Turk (https://requester.mturk.com/). All reported being native speakers of English and being over 18 years of age, and received $1.00 for their participation. They were instructed to read each of 28 sentence fragments, each followed by two continuations, and to choose the more natural continuation by clicking it with a computer mouse. Sixteen of the sentence fragments they saw were the beginnings of 16 of the English materials used in Experiment 1. One example appears in the text as (3). Another appears below: (12)

a. b.

The tenant greeted even the grandfather of the caretaker (1) not only the caretaker’s grandmother. (2) not only the housekeeper’s grandfather

All the NP of NP phrases appeared in object position of the fragment. To ensure that they received focus, each was preceded by the focus particle only or even. Experimental subjects also saw 12 fillers that contained a focus particle but had some other focus ambiguity besides the NP PP, e.g., The janitor only cleaned the bathroom yesterday. Presentation of the materials was controlled by IbexFarm software (http://spellout.net/ibexfarm/). Presentation of the 28 items was individually randomized, and the order of the two continuations was randomized on each trial. Choices and reaction times were recorded by the software (one additional subject was rejected because most response times were under 500 ms). Considering the 16 critical sentences, 82.8% of the choices were of the option illustrated as (1) in (3b) and (12b), that is, the option that contrasted with the head of the focused noun phrase (grandmother vs. grandfather) (SE = 0.024), providing very clear support for our assumption that focusing a NP of NP phrase imparted focus to its head. Appendix B. Sentences used in questionnaire, with short and long versions divided by /. Subject position relative clause version appears first for each item English 1. The son of the colonel who died/who tragically died of a stroke wrote five books on tropical disease. The doctor met the son of the colonel who died/who tragically died of a stroke. 2. The servant of the actress who lied/who shamelessly lied at every chance left town last Saturday. Mr. Johnson visited the servant of the actress who lied/who shamelessly lied at every chance. 3. The attorney of the defendant who mumbled/who always unconsciously mumbled was questioned about personal matters. The reporters interviewed the attorney of the defendant who mumbled/who always unconsciously mumbled. 4. The chauffeur of the millionaire who complained/who complained vociferously about the schedule was taunted by the hecklers in the crowd. John despised the chauffeur of the millionaire who complained/who complained vociferously about the schedule. 5. The gardener of the executive who whistled/who always whistled in the shower was nearly blind. Maria loved the gardener of the executive who whistled/who always whistled in the shower. 6. The mechanic of the officer who disappeared/who frequently disappeared from work was from Alaska. John liked the mechanic of the officer who disappeared/who frequently disappeared from work. 7. The client of the realtor who vanished/who suddenly and unexpectedly vanished was married to the teacher’s daughter. The attorney defended the client of the realtor who vanished/who suddenly and unexpectedly vanished. 8. The agent of the author who quit/who quit abruptly and without warning had red hair and a mustache. Everyone knows the agent of the author who quit/who quit abruptly and without warning. 9. The brother of the visitor who left/who suddenly and unexpectedly left was sick and had a high fever. Emily angered the brother of the visitor who left/who suddenly and unexpectedly left.

B. Hemforth et al. / Lingua 166 (2015) 43--64

57

10. The student of the chemist who fainted/who often unexpectedly fainted was understandably upset. Mr. Miller ignored the student of the chemist who fainted/who often unexpectedly fainted. 11. The assistant of the politician who stutters/who reportedly always stutters went on a cruise to the Bahamas. Maria knows the assistant of the politician who stutters/who reportedly always stutters. 12. The surgeon of the movie star who gossips/who almost always gossips was recently mentioned in the press. Louise recognized the surgeon of the movie star who gossips/who almost always gossips. 13. The companion of the basketball player who stumbles/who allegedly often stumbles didn’t say anything to reporters. Ben interviewed the companion of the basketball player who stumbles/who allegedly often stumbles. 14. The uncle of the brain surgeon who vacationed/who only very rarely vacationed was well known as an author. Friends liked the uncle of the brain surgeon who vacationed/who only very rarely vacationed. 15. The relative of the actor who drank/who too frequently drank hated the cameraman. The cameraman hated the relative of the actor who drank/who too frequently drank. 16. The cousin of the tourist who ate/who quickly and hastily ate watched the waiter. The waiter watched the cousin of the tourist who ate/who quickly and hastily ate. 17. The grandfather of the tenant who stank/who very frequently stank greeted the tenant. The tenant greeted the grandfather of the caretaker who stank/who very frequently stank. 18. The son of the teacher who coughed/who almost continuously coughed teased the pupil. The pupil teased the son of the teacher who coughed/who almost continuously coughed. 19. The advisor of the mayor who blinked/who often unwittingly blinked was welcomed by the chancellor. The chancellor welcomed the advisor of the mayor who blinked/who often unwittingly blinked. 20. The coach of the player who waited/who silently and patiently waited met the sponsor. The sponsor met the coach of the player who waited/who silently and patiently waited. 21. The father-confessor of the monk who prayed/who regularly and intensely prayed met the abbot. The abbot met the father-confessor of the monk who prayed/who regularly and intensely prayed. 22. The son of the journalist who sneezed/who repeatedly and noisily sneezed paid the investor. The interviewer paid the son of the journalist who sneezed/who repeatedly and noisily sneezed. 23. The co-worker of the manager who smokes/who almost always smokes greeted the employee. The employee greeted the co-worker of the manager who smokes/who almost always smokes. 24. The friend of the artist who juggles/who still occasionally juggles yelled at Robert. Robert yelled at the friend of the artist who juggles/who still occasionally juggles. 25. The colleague of the professor who resigned/who foolishly and impetuously resigned met the student. The student met the colleague of the professor who resigned/who foolishly and impetuously resigned. 26. The uncle of the fruit merchant who chatters/who permanently and persistently chatters visited the boss. The boss visited the uncle of the fruit merchant who chatters/who permanently and persistently chatters. 27. The companion of the tramp who fled/who suddenly and unexpectedly fled warned the busker. The busker warned the companion of the tramp who fled/who suddenly and unexpectedly fled. 28. The friend of the pharmacist who answered/who slowly and hesitantly answered insulted the host. The host insulted the friend of the pharmacist who answered/who slowly and hesitantly answered. 29. The brother of the bridegroom who snores/who often unknowingly snores impressed the guest. The guest impressed the brother of the bridegroom who snores/who often unknowingly snores. 30. The employee of the gardener who mowed/who carefully and dutifully mowed liked the supplier. The supplier liked the employee of the gardener who mowed/who carefully and dutifully mowed. 31. The father of the boy who slept/who often soundly slept bothered the grandfather. The grandfather bothered the father of the boy who slept/who often soundly slept. 32. The patron of the artist who swore/who inconsiderately and insultingly swore missed the client. The client missed the patron of the artist who swore/who inconsiderately and insultingly swore. German 1. Der Sohn des Majors, der starb/der tragischerweise an einem Schlaganfall starb, schrieb fünf Bücher über tropische Krankheiten. Der Arzt traf den Sohn des Majors, der starb/der tragischerweise an einem Schlaganfall starb. 2. Die Dienerin der Schauspielerin, die log/die bei jeder Gelegenheit schamlos log, verließ die Stadt letzten Samstag. Herr Müller besuchte die Dienerin der Schauspielerin, die log/die bei jeder Gelegenheit schamlos log.

58

B. Hemforth et al. / Lingua 166 (2015) 43--64

3. Der Anwalt des Angeklagten, der nuschelte/der immer unbewusst nuschelte, wurde nach persönlichen Dingen gefragt. Die Reporter interviewten den Anwalt des Angeklagten, der nuschelte/der immer unbewusst nuschelte. 4. Der Chauffeur des Millionärs, der meckerte/der heftig über den Terminplan meckerte, wurde von den Zwischenrufern in der Menge verspottet. Hans verachtete den Chauffeur des Millionärs, der meckerte/der heftig über den Terminplan meckerte. 5. Der Gärtner des Direktors, der pfiff/der immer unter der Dusche pfiff, war fast blind. Maria liebte den Gärtner des Direktors, der pfiff/der immer unter der Dusche pfiff. 6. Der Mechaniker des Beamten, der verschwand/der oft von der Arbeit verschwand, kam aus Alaska. Hans mochte den Mechaniker des Beamten, der verschwand/der oft von der Arbeit verschwand. 7. Der Kunde des Immobilienmaklers, der verschwand/der plötzlich und unerwartet verschwand, war mit der Tochter des Lehrers verheiratet. Der Anwalt verteidigte den Kunden des Immobilienmaklers, der verschwand/der plötzlich und unerwartet verschwand. 8. Der Agent des Autors, der kündigte/der plötzlich und ohne Vorwarnung kündigte, hatte rote Haare und einen Schnurrbart. Jeder kennt den Agenten des Autors, der kündigte/der plötzlich und ohne Vorwarnung kündigte. 9. Der Bruder des Besuchers, der abreiste/der plötzlich und unerwartet abreiste, war krank und hatte sehr hohes Fieber. Sarah verärgerte den Bruder des Besuchers, der abreiste/der plötzlich und unerwartet abreiste. 10. Der Schüler des Chemikers, der zusammenbrach/der oft unerwartet zusammenbrach, war verständlicherweise verärgert. Herr Müller ignorierte den Schüler des Chemikers, der zusammenbrach/der oft unerwartet zusammenbrach. 11. Der Assistent des Politikers, der stottert/der bekanntlich immer stottert, ging auf eine Kreuzfahrt auf die Bahamas. Karina kennt den Assistenten des Politikers, der stottert/der bekanntlich immer stottert. 12. Der Chirurg des Kinostars, der tratscht/der fast immer tratscht, wurde vor kurzem in der Presse erwähnt. Lydia erkannte den Chirurgen des Kinostars, der tratscht/der fast immer tratscht. 13. Der Begleiter des Basketballspielers, der stolpert/der angeblich oft stolpert, sagte nichts zu den Reportern. Harald interviewte den Begleiter des Basketballspielers, der stolpert/der angeblich oft stolpert. 14. Der Onkel des Neurochirurgen, der verreiste/der nur sehr selten verreiste, war als Autor sehr bekannt. Freunde mochten den Onkel des Neurochirurgen, der verreiste/der nur sehr selten verreiste. 15. Der Verwandte des Schauspielers, der trank/der zu oft trank, hasste den Kameramann. Der Kameramann hasste den Verwandten des Schauspielers, der trank/der zu oft trank. 16. Der Cousin des Touristen, der aß/der schnell und hastig aß, beobachtete den Kellner. Der Kellner beobachtete den Cousin des Touristen, der aß/der schnell und hastig aß. 17. Der Großvater des Haueisters, der stank/der sehr häufig stank, grüßte den Mieter. Der Mieter grüßte den Großvater des Haueisters, der stank/der sehr häufig stank. 18. Der Sohn des Lehrers, der hustete/der fast ununterbrochen hustete, hänselte den Schüler. Der Schüler hänselte den Sohn des Lehrers, der hustete/der fast ununterbrochen hustete. 19. Der Berater des Bürgermeisters, der blinzelte/der oft unwissentlich blinzelte, wurde vom Kanzler begrüßt. Der Kanzler begrüßte den Berater des Bürgermeisters, der blinzelte/der oft unwissentlich blinzelte. 20. Der Trainer des Spielers, der wartete/der ruhig und geduldig wartete, traf den Sponsor. Der Sponsor traf den Trainer des Spielers, der wartete/der ruhig und geduldig wartete. 21. Der Beichtvater des Mönchs, der betete/der regelmäßig und intensiv betete, traf den Abt. Der Abt traf den Beichtvater des Mönchs, der betete/der regelmäßig und intensiv betete. 22. Der Sohn des Journalisten, der nieste/der wiederholt und laut nieste, bezahlte den Interviewer. Der Interviewer bezahlte den Sohn des Journalisten, der nieste/der wiederholt und laut nieste. 23. Der Mitarbeiter des Managers, der raucht, grüßte den Angestellten. Der Mitarbeiter des Managers, der raucht/der fast immer raucht, grüßte den Angestellten. Der Angestellte grüßte den Mitarbeiter des Managers, der raucht/der fast immer raucht. 24. Der Freund des Künstlers, der jongliert/der noch gelegentlich jongliert, beschimpfte Robert. Robert beschimpfte den Freund des Künstlers, der jongliert/der noch gelegentlich jongliert. 25. Der Kollege des Professors, der kündigte/der vorschnell und unüberlegt kündigte, traf den Studenten. Der Student traf den Kollegen des Professors, der kündigte/der vorschnell und unüberlegt kündigte. 26. Der Onkel des Obsthändlers, der schwatzt/der permanent und ausdauernd schwatzt, besuchte den Chef. Der Chef besuchte den Onkel des Obsthändlers, der schwatzt/der permanent und ausdauernd schwatzt.

B. Hemforth et al. / Lingua 166 (2015) 43--64

59

27. Der Gefährte des Landstreichers, der floh/der plötzlich und unerwartet floh, warnte den Straßenmusikanten. Der Straßenmusikant warnte den Gefährten des Landstreichers, der floh/der plötzlich und unerwartet floh. 28. Der Freund des Apothekers, der antwortete/der langsam und zögerlich antwortete, beleidigte den Gastgeber. Der Gastgeber beleidigte den Freund des Apothekers, der antwortete/der langsam und zögerlich antwortete. 29. Der Bruder des Bräutigams, der schnarcht/der oft unbewusst schnarcht, beeindruckte den Gast. Der Gast beeindruckte den Bruder des Bräutigams, der schnarcht/der oft unbewusst schnarcht. 30. Der Angestellte des Gärtners, der mähte/der sorgfältig und pflichtbewusst mähte, mochte den Lieferanten. Der Lieferant mochte den Angestellten des Gärtners, der mähte/der sorgfältig und pflichtbewusst mähte. 31. Der Vater des Jungen, der schlief/der oft tief schlief, verärgerte den Großvater. Der Großvater verärgerte den Vater des Jungen, der schlief/der oft tief schlief. 32. Der Mäzen des Künstlers, der fluchte/der rücksichtslos und beleidigend fluchte, verpasste den Kunden. Der Kunde verpasste den Mäzen des Künstlers, der fluchte/der rücksichtslos und beleidigend fluchte. Spanish 1. El hijo del coronel que murio´/que trágicamente murio´ de apoplejía escribio´ cinco libros sobre enfermedades tropicales. El doctor conocio´ al hijo del coronel que murio´/que trágicamente murio´ de apoplejía. 2. El criado de la actriz que mentía/que desvergonzadamente mentía en toda ocasio´n dejo´ la ciudad el pasado sábado. El señor Soler visito´ al criado de la actriz que mentía/que desvergonzadamente mentía en toda ocasio´n. 3. El abogado del acusado que farfullaba/que siempre inconscientemente farfullaba fue interrogado acerca de cuestiones personales. Los reporteros entrevistaron al abogado del acusado que farfullaba/que siempre inconscientemente farfullaba. 4. El cho´fer del millonario que se quejaba/que se quejaba a gritos sobre el programa fue increpado por los hostigadores de la muchedumbre. Juan desdeño´ al cho´fer del millonario que se quejaba/que se quejaba a gritos sobre el programa. 5. El jardinero del ejecutivo que silvaba/que siempre silva en la ducha estaba casi ciego. María amaba al jardinero del ejecutivo que silvaba/que siempre silva en la ducha. 6. El mecánico del oficial que desaparecía/que frecuentemente desaparecía del trabajo era de Alaska. A Juan le gustaba el mecánico del oficial que desaparecía/que frecuentemente desaparecía del trabajo. 7. El cliente del abogado que desaparecio´/que repentina e inesperadamente desaparecio´ estaba casado con la hija del profesor. El abogado defendio´ al cliente del corredor que desaparecio´/que rápida e Inesperadamente desaparecio´. 8. El agente del autor que abandono´/que abandono´ inopinadamente y sin avisar tenía el pelo rojo y un bigote Todo el mundo conoce al agente del autor que abandono´/que abandono´ inopinadamente y sin avisar. 9. El hermano del visitante que se fue/que inesperada y repentinamente se fue estaba enfermo y tenía fiebre muy alta. Emilia se enfado´ con el hermano del visitante que se fue/que inesperada y repentinamente se fue. 10. El estudiante del químico que se desmayaba/que frecuente e inesperadamente se desmayaba estaba inexplicablemente enfadado. El señor Vazquéz ignoro´ al estudiante del químico que se desmayaba/que frecuente e espontáneamente se desmayaba. 11. El asistente del político que tartamudea/que siempre públicamente tartamudea se fue a un crucero por las Bahamas. María conoce al asistente del político que tartamudea/que públicamente siempre tartamudea. 12. El cirujano de la estrella de cine que chismorreaba/que casi siempre chismorreaba fue recientemente mencionado en la prensa. Luisa reconocio´ al cirujano de la estrella de cine que chismorreaba/que casi siempre chismorreaba. 13. El compañero del jugador de baloncesto que tropieza/que supuestamente tropieza a menudo, no dijo nada a los reporteros. Miguel entrevisto´ al compañero del jugador de baloncesto que tropieza/que supuestamente tropieza a menudo. 14. El tío del neurocirujano que cogía vacaciones/que cogía muy raramente vacaciones era bien conocido como autor. A algunos amigos les gustaba el tío del neurocirujano que cogía vacaciones/que en raras ocasiones cogía vacaciones. 15. El familiar del actor que bebía/que demasiado frecuentemente bebía odiaba al cámara. El cámara odiaba al familiar del actor que bebía/que demasiado frecuentemente bebía.

60

B. Hemforth et al. / Lingua 166 (2015) 43--64

16. El primo de el turista que comio´/que rápida y ansiosamente comio´ vigilaba al camarero. El camarero vigilaba al primo de el turista que comio´/que rápida y ansiosamente comio´. 17. El abuelo del celador que apestaba/que muy amenudo apestaba saludo´ al inquilino. El inquilino saludo´ al abuelo del celador que apestaba/que muy amenudo apestaba. 18. El hijo del profesor que tosía/que casi continuamente tosía molestaba al alumno. El alumno molestaba al hijo del profesor que tosía/que casi continuamente tosía. 19. El consejero del alcalde que parpadeaba/que frecuente e inconscientemente parpadeaba fue bienvenido por el Ministro. El Ministro dio la bienvenida al consejero del alcalde que parpadeaba/que frecuente e inconscientemente parpadeaba. 20. El entrenador del jugador que esperaba/que silenciosa y pacientemente esperaba conocio´ al patrocinador. El patrocinador conocio´ al entrenador del jugador que esperaba/que silencios y pacientemente esperaba. 21. El padre-confesor del monje que rezaba/que regularmente e intensamente rezaba conocio´ al abad. El abad conocio´ al padre-confesor del monje que rezaba/que regularmente e intensamente rezaba. 22. El hijo del periodista que estornudaba/que repetida y ruidosamente estornudaba pago´ al entrevistador. El entrevistador pago´ al hijo del periodista que estornudaba/que reiterada y estrepitosamente estornudaba. 23. El colaborador del manager que fumaba/que casi siempre fumaba felicito´ al empleado. El empleado felicito´ al colaborador del manager que fumaba/que casi siempre fumaba. 24. El amigo del artista que hacía malabares/que todavía ocasionalmente hacía malabares grito´ a Roberto. Roberto grito´ al amigo del artista que hacía malabares/que hacía malabares. 25. El colega del catedrático que dimitio´/que impetuosa y estúpidamente dimitio´ se encontro´ con el estudiante. El estudiante se encontro´ al colega del catedrático que dimitio´/que impetuosa y estúpidamente dimitio´. 26. El tío del frutero que cotorreaba/que contínua y persistentemente cotorreaba. visito´ al jefe. El jefe visito´ al tío del frutero que cotorreaba/que continua y persistentemente cotorreaba. 27. El colega del titiritero que huyo´/que repentina y inesperadamente huyo´ aviso´ al músico ambulante. El músico ambulante aviso´ al colega del titiritero que huyo´/que repentina e inesperadamente huyo´. 28. El amigo del farmacéutico que contesto´/que lenta y dubitativamente contesto´ insulto´ al anfitrio´n. El anfintrio´n insulto´ al amigo del farmacéutico que contesto´/que lenta y dubitativamente contesto´. 29. El hermano del novio que roncaba/que a menudo incoscientemente roncaba impresiono´ al invitado. El invitado impresiono´ al hermano del novio que roncaba/que a menudo incoscientemente roncaba. 30. Al empleado del jardinero que segaba/que cuidadosa y alegremente segaba le gustaba el proveedor. Al proveedor le gustaba el empleado del jardinero que segaba/que cuidadosa y obedientemente segaba. 31. El padre del niño que dormía/que muy ruidosament dormía molesto´ al abuelo. El abuelo molesto´ al padre del niño que dormía/que frecuentemente y ruidosamente dormía. 32. El representante del artista que blasfemaba/que desconsiderada e insultantemente blasfemaba perdio´ el cliente. El cliente perdio´a al representante del artista que blasfemaba/que desconsiderada e insultantemente blasfemaba. French 1. Le fils du commandant qui mourut/qui mourut tragiquement d’une attaque d’apoplexie écrivit cinq livres sur des maladies tropicales. Le médecin rencontra le fils du commandant qui mourut/qui mourut tragiquement d’une attaque d’apoplexie. 2. La servante de l’actrice qui mentait/qui mentait impunément à toute occasion a quitté la ville samedi dernier. M. Leroux a rendu visite à la servante de l’actrice qui mentait/qui mentait impunément à toute occasion. 3. L’avocat de l’accusé qui chuchotait/qui chuchotait tout le temps sans s’en apercevoir fut interrogé sur des questions personnelles. Les reporters interviewèrent l’avocat de l’accusé qui chuchotait/qui chuchotait tout le temps sans s’en apercevoir. 4. Le chauffeur du millionnaire qui râlait/qui râlait violemment contre le planning fut insulté par des interpellateurs dans la foule. Jean méprisait le chauffeur du millionnaire qui râlait/qui râlait violemment contre le planning. 5. Le jardinier du directeur qui sifflait/qui sifflait toujours sous la douche était presque aveugle. Marie aimait le jardinier du directeur qui sifflait/qui sifflait toujours sous la douche. 6. Le mécanicien du fonctionnaire qui disparaissait/qui disparaissait souvent de son travail venait de l’Alaska. Jean aimait bien le mécanicien du fonctionnaire qui disparaissait/qui disparaissait souvent de son travail.

B. Hemforth et al. / Lingua 166 (2015) 43--64

61

7. Le client de l’agent immobilier qui disparuˆ t/qui disparuˆ t soudainement et à l’improviste était marié avec la fille du professeur. L’avocat défendait le client de l’agent immobilier qui disparuˆ t/qui disparuˆ t soudainement et à l’improviste. 8. L’agent de l’auteur qui démissionna/qui démissionna brusquement et sans préavis avait les cheveux roux et une mustache. Tout le monde connaît l’agent de l’auteur qui démissionna/qui démissionna brusquement et sans préavis. 9. Le frère du visiteur qui partit/qui partit soudainement et sans préavis était malade et avait une fièvre très élevée. Sarah irritait le frère du visiteur qui partit/qui partit soudainement et sans préavis. 10. L’élève du chimiste qui s’évanouissait/qui s’évanouissait souvent et sans préavis était énormément irrité. M. Leroux ignora l’élève du chimiste qui s’évanouissait/qui s’évanouissait souvent et sans préavis. 11. L’ assistant de l’homme politique qui bégayait/qui bégayait beaucoup et tout le temps faisait une croisière aux Bahamas. Carine connaît l’ assistant de l’homme politique qui bégayait/qui bégayait beaucoup et tout le temps. 12. Le chirurgien de la chanteuse qui cancane/qui cancane presque tout le temps était mentionné dans la presse il y a peu de temps. Sandrine reconnut le chirurgien de la chanteuse qui cancane/qui cancane presque tout le temps. 13. L’accompagnateur du footballeur qui trébuche/qui soi-disant trébuche souvent ne disait rien aux reporters. Paul interviewait l’accompagnateur du footballeur qui trébuche/qui soi-disant trébuche souvent. 14. L’oncle du neurochirurgien qui voyageait/qui ne voyageait que très rarement était très connu comme auteur. Des amis aimaient bien l’oncle du neurochirurgien qui voyageait/qui ne voyageait que très rarement. 15. Le parent de l’acteur qui buvait/qui buvait trop souvent haïssait le cadreur. Le cadreur haïssait le parent de l’acteur qui buvait/qui buvait trop souvent. 16. Le cousin du touriste qui mangeait/qui mangeait vite et hâtivement observait le serveur. Le serveur observait le cousin du touriste qui mangeait/qui mangeait vite et hâtivement. 17. Le grand-père du concierge qui puait/qui très fréquemment puait saluait le locataire. Le locataire saluait le grand-père du concierge qui puait/qui très fréquemment puait. 18. Le fils du professeur qui toussait/qui toussait presque sans cesse se moquait de l’élève. L’élève se moquait du fils du professeur qui toussait/qui toussait presque sans cesse. 19. Le conseiller du ministre qui tremblait/qui tremblait souvent et sans le savoir était salué par le chancelier. Le chancelier saluait le conseiller du ministre qui tremblait/qui tremblait souvent et sans le savoir. 20. L’entraîneur du joueur qui attendait/qui attendait tranquillement et patiemment rencontra le sponsor. Le sponsor rencontra l’entraîneur du joueur qui attendait/qui attendait tranquillement et patiemment. 21. Le confesseur du moine qui priait/qui priait régulièrement et avec intensité rencontra l’abbé. L’abbé rencontra le confesseur du moine qui priait/qui priait régulièrement et avec intensité. 22. Le fils du journaliste qui s’éternua/qui s’éternua bruyamment et à plusieurs reprises payait l’intervieweur. L’intervieweur payait le fils du journaliste qui éternua/qui éternua bruyamment et à plusieurs reprises. 23. Le collaborateur du manageur qui fume/qui fume presque toujours salua l’employé. L’employé salua le collaborateur du manageur qui fume/qui fume presque toujours. 24. Le partenaire de l’artiste qui jongle/qui jongle encore de temps en temps insulta Robert. Robert insulta le partenaire de l’artiste qui jongle/qui jongle encore de temps en temps. 25. Le collègue du professeur qui démissionna/qui démissionna de façon précipitée et sans réfléchir rencontra l’étudiant. L’étudiant rencontra le collègue du professeur qui démissionna/qui démissionna de façon précipitée et sans réfléchir. 26. L’oncle du marchand qui bavarde/qui bavarde constamment et avec insistance rendit visite au chef. Le chef rendit visite à l’oncle du marchand qui bavarde/qui bavarde constamment et avec insistance. 27. Le compagnon du vagabond qui s’enfuit/qui s’enfuit tout d’un coup et à l’improviste prévint le musicien ambulant. Le musicien ambulant prévint le compagnon du vagabond qui s’enfuit/qui s’enfuit tout d’un coup et à l’improviste. 28. L’ami du pharmacien qui répondit/qui répondit d’une façon lente et hésitante offensa l’hôte. L’hôte offensa l’ami du pharmacien qui répondit/qui répondit d’une façon lente et hésitante. 29. Le frère du marié qui ronfle/qui ronfle souvent sans le savoir impressionnait l’invité. L’invité impressionnait le frère du marié qui ronfle/qui ronfle souvent sans le savoir. 30. L’employé du jardinier qui moissonnait/qui moissonnait soigneusement et bien conscient de ses devoirs aimait bien le fournisseur.

62

B. Hemforth et al. / Lingua 166 (2015) 43--64

Le fournisseur aimait bien l’employé du jardinier qui moissonnait/qui moissonnait soigneusement et bien conscient de ses devoirs. 31. Le père du fils qui dormait/qui dormait souvent profondément énervait le grand-père. Le grand-père énervait le père du fils qui dormait/qui dormait souvent profondément. 32. Le mécène de l’artiste qui jurait/qui jurait d’une manière offensante et sans scrupules manqua le client. Le client manqua le mécène de l’artiste qui jurait/qui jurait d’une manière offensante et sans scrupules. References Almor, A., 1999. Noun-phrase anaphora and focus: the informational load hypothesis. Psychol. Rev. 106, 748--765. http://dx.doi.org/10.1037/ 0033-295x.106.4.748 Alonso-Ovalle, L., Fernández-Solera, S., Frazier, L., Clifton Jr., C., 2002. Null versus overt pronouns and the topic-focus articulation in Spanish. Riv. Linguist. 14, 151--169. Besten, H. den., 1983. On the interactions of root transformations and lexical deletion rules. In: Abraham, W. (Ed.), On the Formal Syntax of the West Germania. John Benjamin, Amsterdam, Netherlands, pp. 47--131. Brysbaert, M., Mitchell, D.C., 1996. Modifier attachment in sentence parsing: evidence from Dutch. Q. J. Exp. Psychol. 49A, 664--695. http://dx. doi.org/10.1080/027249896392540 Carlson, K., Clifton Jr., C., Frazier, L., 2001. Prosodic boundaries in adjunct attachment. J. Mem. Lang. 45, 58--81. http://dx.doi.org/10.1006/ jmla.2000.2762 Carlson, K., Dickey, M.W., Frazier, L., Clifton Jr., C., 2009. Information structure expectations in sentence comprehension. Q. J. Exp. Psychol. 62 (1), 114--139. http://dx.doi.org/10.1080/17470210701880171 Carminati, M., 2001. The Processing of Italian Subject Pronouns (Doctoral dissertation). University of Massachusetts Amherst, Amherst, Massachusetts. Carreiras, M., Clifton Jr., C., 1993. Relative clause interpretation preferences in Spanish and English. Lang. Speech 36, 353--372. Carreiras, M., Clifton Jr., C., 1999. Another word on parsing relative clauses: eyetracking evidence from Spanish and English. Mem. Cognit. 27, 826--833. http://dx.doi.org/10.3758/Bf03198535 Carreiras, M., Salillas, E., Barber, H., 2004. Event-related potentials elicited during parsing of ambiguous relative clauses in Spanish. Cogn. Brain Res. 20, 98--105. http://dx.doi.org/10.1016/j.cogbrainres.2004.01.009 Clifton Jr., C., Carlson, K., Frazier, L., 2002. Informative prosodic boundaries. Lang. Speech 45, 87--114. Cuetos, F., Mitchell, D.C., 1988. Cross-linguistic differences in parsing: restrictions on the use of the Late Closure strategy in Spanish. Cognition 30, 73--105. http://dx.doi.org/10.1016/0010-0277(88)90004-2 De Vincenzi, M., Job, R., 1993. Some observations on the universality of the late-closure strategy. J. Psycholinguist. Res. 22, 189--206. http://dx. doi.org/10.1016/S0093-934X(03)00055-5 De Vincenzi, M., Job, R., 1995. A cross-linguistic investigation of late closure: the role of syntax, thematic structure, and pragmatics in initial and final interpretation. J. Exp. Psychol. Learn. Mem. Cogn. 21, 1303--1321. http://dx.doi.org/10.1037/0278-7393.21.5.1303 Desmet, T., Declerc, Q.M., 2006. Cross-linguistic priming of syntactic hierarchical configuration information. J. Mem. Lang. 54, 610--632. http://dx. doi.org/10.1016/j.jml.2005.12.007 Desmet, T., de Baecke, C., Brysbaert, M., 2002. The influence of referential discourse context on modifier attachment in Dutch. Mem. Cogn. 30, 150--157. Desmet, T., Brysbaert, M., de Baecke, C., 2002. The correspondence between sentence production and corpus frequencies in modifier attachment. Q. J. Exp. Psychol. 55A, 879--896. http://dx.doi.org/10.1080/02724980143000604 Fernández, E.M., 2005. The prosody produced by Spanish--English bilinguals: a preliminary investigation and implications for sentence processing. Rev. Abralin 4 (1), 109--141. http://www.abralin.org/revista/RV4N1_2/RV4N1_2_art4.pdf Ferreira, F., 1993. Creation of prosody during sentence production. Psychol. Rev. 100, 233--253. Fodor, J.D., 1998. Learning to parse? J. Psycholinguist. Res. 27, 285--319. http://dx.doi.org/10.1023/A:1023258301588 Fodor, J.D., 2002. Psycholinguistics Cannot Escape Prosody. Paper Presented at Speech Prosody 2002, Aix-en-Provence, France, pp. 83--90 http://sprosig.isle.illinois.edu/sp2002/pdf/fodor.pdf Frazier, L., 1978. On Comprehending Sentences: Syntactic Parsing Strategies (Unpublished doctoral dissertation). University of Connecticut. Frazier, L., 1990. Parsing modifiers. Special purpose routines in the human sentence processing mechanism? In: Balota, D., d’Arcais, G.F., (Eds.), Comprehension Processes in Reading. Lawrence Erlbaum, Hillsdale, NJ, pp. 303--330. Frazier, L., Clifton Jr., C., 1996. Construal. MIT Press, Cambridge, MA. Frazier, L., Flores d’Arcais, G.B., 1989. Filler-driven parsing: a study of gap filling in Dutch. J. Mem. Lang. 28, 331--344. http://dx.doi.org/10.1016/ 0749-596x(89)90037-5 Gee, J., Grosjean, F., 1983. Performance structure: a psycholinguistic and linguistic appraisal. Cogn. Psychol. 15, 411--458. http://dx.doi.org/ 10.1016/0010-0285(83)90014-2 Gibson, E., 1991. A Computational Theory of Human Linguistic Processing: Memory Limitations and Processing Breakdown (Doctoral dissertation). Carnegie Mellon University, Pittsburgh, Pennsylvania. Gibson, E., Pearlmutter, N., Canseco-Gonzalez, E., Hickok, G., 1996. Cross-linguistic attachment preferences: evidence from English and Spanish. Cognition 59, 23--59. http://dx.doi.org/10.1016/0010-0277(95)00687-7 Gibson, E., Pearlmutter, N., Torrens, V., 1999. Recency and lexical preferences in Spanish. Mem. Cogn. 27, 603--611. Gilboy, E., Sopena, J., Clifton Jr., C., Frazier, L., 1995. Argument structure and association preferences in Spanish and English compound NPs. Cognition 54, 131--167. http://dx.doi.org/10.1016/0010-0277(94)00636-Y

B. Hemforth et al. / Lingua 166 (2015) 43--64

63

Grillo, N., Costa, J., 2014. A novel argument for the Universality of Parsing principles. Cognition 133, 156--187. http://dx.doi.org/10.1016/j. cognition.2014.05.019 Gryllia, S., Kügler, F., 2010. What does prosody tell us about relative clause attachments in German. Speech Prosody 1--4 Retrieved from http:// speechprosody2010.illinois.edu/papers/100927.pdf Hemforth, B., Konieczny, L., 2002. Where Pronouns and Relative Clauses Differ: Information Structure and Binding Preferences. Paper Presented at the 15th Annual CUNY Conference on Human Sentence Processing, New York, March 2002. Hemforth, B., Konieczny, L., Seelig, H., Walter, M., 2000. Case matching and relative clause attachment. J. Psycholinguist. Res. 29, 81--88. http:// dx.doi.org/10.1023/A:1005176507878 Hemforth, B., Konieczny, L., Scheepers, C., 2000a. Syntactic attachment and anaphor resolution: the two sides of relative clause attachment. In: Crocker, M., Pickering, M., Clifton, Jr., C. (Eds.), Architecture and Mechanisms of Language Processing. Cambridge University Press, Cambridge, pp. 259--281. Hemforth, B., Konieczny, L., Scheepers, C., 2000b. Modifier attachment: relative clauses and coordinations. In: Hemforth, B., Konieczny, L (Eds.), German Sentence Processing. Kluwer, Dordrecht, pp. 161--186. Hemforth, B., Colonna, S., Petrone, C., D’Imperio, M., 2013. Length matters: informational load in ambiguity resolution. Discours 12. http://dx.doi. org/10.4000/discours.8780 http://discours.revues.org/8780 Jun, S.-A., 2010. The implicit prosody hypothesis and overt prosody in English. Lang. Cogn. Process. 25 (7--9), 1201--1233. http://dx.doi.org/ 10.1080/01690965.2010.503658 Kaiser, G.A., Zimmermann, M., 2011. On the decrease in subject--verb inversion in French declaratives. In: Rinke, E., Kupisch, T. (Eds.), The Development of Grammar: Language Acquisition and Diachronic Change; in Honour of Jürgen M. Meisel. Benjamins, Amsterdam, pp. 355--382. Kiss, T., 2003. Phrasal typology and the interaction of topicalization, wh-movement, and extraposition. In: Kim, J, Wechsler, S. (Eds.), The Proceedings of the 9th International Conference on Head-Driven Phrase Structure Grammar. CSLI Publications, Stanford, pp. 109--128. Kiss, T., 2005. Semantic constraints on relative clause extraposition. Nat. Lang. Linguist. Theory (23), 281--334. Konieczny, L., Hemforth, B., 2000. Modifier attachment in German: relative clauses and prepositional phrases. In: Kennedy, A., Radach, R., Heller, D., Pynte, J. (Eds.), Reading as a Perceptual Process. Elsevier, Amsterdam, pp. 517--528. Lahousse, K., 2006. NP subject inversion in French: two types, two configurations. Lingua 116 (4), 424--461. http://dx.doi.org/10.1016/j. lingua.2004.08.020 Marandin, J.-M., 2003. Inversion du sujet et structure de l’information dans les langues romanes. In: Godard, D. (Ed.), Langues romanes. Problèmes de la phrase simple. Éditions du CNRS, Paris. Maynell, L.A., 1999. Effect of Pitch Accent Placement on Resolving Relative Clause Ambiguity in English. Poster Presented at 12th Annual CUNY Conference on Human Sentence Processing, New York. Mitchell, D.C., Brysbaert, M., 1998. Challenges to recent theories of cross-linguistic variation in parsing: evidence from Dutch. In: Hillert, D. (Ed.), Sentence Processing: A Cross-Linguistic Perspective. Academic Press, San Diego, pp. 313--335. Mitchell, D.C., Cuetos, F., 1991. The Origins of Parsing Strategies. Conference Proceedings, Current Issues in Natural Language Processing. University of Texas at Austin, Austin, TX, pp. 1--12. Mitchell, D.C., Cuetos, F., Corley, M.M.B., Brysbaert, M., 1995. Exposure-based models of human parsing: evidence for the use of coarse-grained (non-lexical) statistical records. J. Psycholinguist. Res. 24, 469--488. http://dx.doi.org/10.1007/BF02143162 Pérez-Leroux, A., Glass, W., 1997. OPC effects on the L2 acquisition of Spanish. In: Perez-Leroux, A., Glass, W. (Eds.), Contemporary Perspectives on the Acquisition of Spanish vol. 1: Developing Grammar. Cascadilla Press, Somerville, MA, pp. 149--165. Pynte, J., Colonna, S., 2000. Decoupling syntactic parsing from visual inspection. In: Kennedy, A., Radach, R., Heller, D., Pynte, J. (Eds.), Reading as a Perceptual Process. Elsevier, Oxford, pp. 529--547. Rohde, H., Levy, R., Kehler, A., 2011. Anticipating explanations in relative clause processing. Cognition 118 (3), 339--358. http://dx.doi.org/ 10.1016/j.cognition.2010.10.016 Schafer, A., Carter, J., Clifton Jr., C., Frazier, L., 1996. Focus in relative clause construal. Lang. Cogn. Process. 11, 135--163. http://dx.doi.org/ 10.1080/016909696387240 Scheepers, C., 2003. Syntactic priming of relative clause attachments: persistence of structural configuration in sentence production. Cognition 89 (3), 179--205. http://dx.doi.org/10.1016/S0010-0277(03)00119-7 Scheepers, C., Sturt, P., Martin, C.J., Myachykov, A., Teevan, K., Viskupova, I., 2011. Structural priming across cognitive domains: from simple arithmetic to relative-clause attachment. Psychol. Sci. 22 (10), 1319--1326. http://dx.doi.org/10.1177/0956797611416997 Schimke, S., Konieczny, L., Hemforth, B., 2002. Relative Clause Attachment in French is Influenced by Commas, Not Case Matching. Poster Presented at Architectures and Mechanisms of Language Processing, Tenerife, Spain, September, 2002. Schwartz, B.D., Vikner, S., 1996. The verb always leaves IP in V2 clauses. In: Belletti, A., Rizzi, L. (Eds.), Parameters and Functional Heads. Oxford University Press, New York, pp. 11--61. Selkirk, E., 1984. Phonology and Syntax: The Relation between Sound and Structure. MIT Press, Cambridge, Mass.. Swets, B., DeSmet, T., Hambrick, D., Ferreira, F., 2007. The role of working memory in syntactic ambiguity resolution: a working memory approach. J. Exp. Psychol. Gen. 136, 64--81. http://dx.doi.org/10.1037/0096-3445.136.1.64 Thornton, R., MacDonald, M.E., Gil, M., 1999. Pragmatic constraint on the interpretation of complex noun phrases in Spanish and English. J. Exp. Psychol. Learn. Mem. Cogn. 25, 1347--1365. http://dx.doi.org/10.1037/0278-7393.25.6.1347 Travis, L., 1984. Parameters and Effects of Word Order Variation (Ph.D. diss.). Massachusetts Institute of Technology. Travis, L., 1991. Parameters of phrase structure and V2 phenomena. In: Freidin, R. (Ed.), Principles and Parameters in Comparative Grammar. MIT Press, Cambridge, MA, pp. 339--364. Traxler, M.J., 2007. Working memory contributions to relative clause attachment processing: a hierarchical linear modeling analysis. Mem. Cogn. 35, 1107--1121. http://dx.doi.org/10.3758/BF03193482 Traxler, M.J., 2009. A hierarchical linear modeling analysis of working memory and implicit prosody in the resolution of adjunct attachment ambiguity. J. Psycholinguist. Res. 38, 491--509. http://dx.doi.org/10.1007/s10936-009-9102-x

64

B. Hemforth et al. / Lingua 166 (2015) 43--64

Traxler, M., Pickering, M., Clifton Jr., C., 1998. Adjunct attachment is not a form of lexical ambiguity resolution. J. Mem. Lang. 39, 558--592. http:// dx.doi.org/10.1006/jmla.1998.2600 Watson, D., Gibson, E., 2004. The relationship between intonational phrasing and syntactic structure in language production. Lang. Cogn. Process. 19, 713--755. http://dx.doi.org/10.1080/01690960444000070 Wijnen, F., 2001. Implicit Prosody in Sentence Processing. Paper Presented at the Workshop on Prosody in Processing, Utrecht, July 2001. Zagar, D., Pynte, J., Rativeau, S., 1997. Evidence for early-closure attachment on first-pass reading times in French. Q. J. Exp. Psychol. 50A, 421--438. http://dx.doi.org/10.1080/027249897392161