This is the accepted version of the journal article: Stadthagen-González, Hans; Parafita Couto, M. Carmen; Parraga, Carlos Ale- jandro; [et al.]. «Testing alternative theoretical accounts of code-switching : Insights from comparative judgments of adjective-noun order». International Journal of Bilingualism, Vol. 23, Issue 1 (February 2019), p. 200-220. DOI 10.1177/1367006917728390 This version is available at https://ddd.uab.cat/record/275083 under the terms of the license Testing alternative theoretical accounts of code-switching: Insights from comparative judgments of adjective–noun order Hans Stadthagen-González, M Carmen Parafita Couto, C Alejandro Párraga, and Markus F Damian Abstract Objectives: Spanish and English contrast in adjective–noun word order: for example, brown dress (English) vs. vestido marrón (‘dress brown’, Spanish). According to the Matrix Language model ( MLF) word order in code-switched sentences must be compatible with the word order of the matrix language, but working within the minimalist program (MP), Cantone and MacSwan arrived at the descriptive generalization that the position of the noun phrase relative to the adjective is determined by the adjective’s language. Our aim is to evaluate the predictions derived from these two models regarding adjective–noun order in Spanish–English code-switched sentences. Methodology: We contrasted the predictions from both models regarding the acceptability of code-switched sentences with different adjective–noun orders that were compatible with the MP, the MLF, both, or none. Acceptability was assessed in Experiment 1 with a 5-point Likert and in Experiment 2 with a 2-Alternative Forced Choice (2AFC) task. Data and analysis: Data from both experiments were subjected to linear mixed model analyses. Results from the 2AFC task were also analyzed using Thurstone’s law of comparative judgment. Conclusions: We found an additive effect in which both the language of the verb and the language of the adjective determine word order. Originality: Both experiments examine adjective–noun word order in English–Spanish code-switched sentences. Experiment 2 represents a novel application of Thurstone’s law of comparative judgements to the study of linguistic acceptability which yielded clearer results than Likert scales. We found convincing evidence that neither the MLF nor the MP can fully account for the acceptability of adjective–noun switches. Implications: We suggest that advances in our understanding of grammaticality in code-switching will be achieved by combining the insights of the two frameworks instead of considering them in isolation, or by espousing a probabilistic model of code-switching. Testing alternative theoretical accounts of code-switching: Insights from comparative 4 5 judgments 67 8 Code-switching is the phenomenon by which bilinguals go back and forth between their two 9 10 11 languages in the same conversation (see Deuchar, 2012). It is generally accepted that switches 12 13 between languages do not occur at random, but follow specific patterns and rules (Bullock & 14 15 Toribio, 2009; Guzzardo Tamargo, Mazak & Parafita Couto, 2016, i.a.), and the typifying of 16 17 18 those rules represents a rich and active field of research in the bilingualism literature. However, 19 20 scholars do not agree on the best theoretical account of these regularities. Recently, an interest 21 22 has emerged in evaluating the predictions of theoretical models to try to disentangle between 23 24 25 theoretical predictions (see Gullberg & Parafita Couto, 2016; Herring, Deuchar, Parafita Couto 26 27 & Moro Quintanilla, 2010; Parafita Couto, Deuchar & Fusser, 2015; Fairchild & Van Hell, 2015; 28 29 vanden Wyngaerd, 2016, or Eppler, Luescher & Deuchar, 2016). These researchers have shown 30 31 32 that different theories account for some aspects of the observed data, but there is no overarching 33 34 theory that can explain all the code-switching patterns. In what follows we will experimentally 35 36 37 examine the accuracy of predictions regarding adjective-noun order derived from an approach 38 39 within the Minimalist Program (Cantone & MacSwan 2009) and the Matrix Language 40 41 Framework (Myers-Scotton, 2002). Of particular interest are the so-called conflict sites (Poplack 42 43 44 & Meechan, 1998), instances where the grammars of the two languages differ. For example, the 45 46 default order for noun-adjective constructions in English is adjective-noun as in ‘red book’, 47 48 whereas the default order in Spanish is the opposite (libro rojo –literally ‘book red’). In Spanish, 49 50 51 some adjectives may appear before or after the noun with a change in meaning. Post-nominal 52 53 adjectives have a basic, attributive meaning (1a), while prenominal adjectives have a more 54 55 restricted, intrinsic meaning (1b). 3 (1) a. un hombre grande b. un gran hombre 4 5 6 ‘a tall man’ ‘a great man’ 7 8 However, in this paper, we will focus on post-nominal adjectives. 9 10 11 In the context of Spanish-English code-switching, we may consider whether libro red, 12 13 red libro, rojo book, and book rojo are equally probable. It is unclear, how a conflict is resolved 14 15 (i.e., what constrains word order) when a Spanish noun is combined with an English adjective, or 16 17 18 vice versa. Some comparative analyses addressing the question of adjective-noun order in code- 19 20 switched noun phrases have been done couched within the two theoretical models (MP and 21 22 MLF) but with different outcomes (cf.; Parafita Couto, Deuchar & Fusser, 2015; vanden 23 24 25 Wyngaerd, 2016). 26 27 Of the two models in question, Myers-Scotton’s (2002) MLF differentiates the languages 28 29 involved: one language is known as the matrix language (ML), the other as the embedded 30 31 32 language (EL). In a code-switched clause, the ML is assumed to provide the morphosyntactic 33 34 frame and allows predictions regarding acceptable constructions. On the other hand, generativist 35 36 37 MacSwan (2005a, 2005b) argues that all instances of code-switching may be accounted for based 38 39 on general mechanisms of grammar (for further details see MacSwan, 2005a, 2005b; Jake, 40 41 Myers-Scotton & Gross, 2005). 42 43 44 The MLF predicts that (i) ‘late outsider morphemes’ such as finite verb morphology marking 45 46 subject-verb agreement and (ii) word order within a clause that contains code-switching will be 47 48 sourced from the ML. Previous studies of Welsh-English code-switching show that Welsh- 49 50 51 English bilinguals tend to produce clauses with code-switching in which the language of the 52 53 finite verb morphology matches clause word order (Davies, 2010; Davies & Deuchar, 2010; 54 55 Deuchar & Davies, 2009; Deuchar, 2006; Parafita Couto, Deuchar & Fusser, 2015). This leads 3 us to predict that if the bound morphology of the finite verb is from language A, adjective and 4 5 6 noun word order in a code-switched noun phrase should also be dictated by language A. 7 8 Cantone and MacSwan (2009) in their analysis of noun-adjective order in Italian-German 9 10 11 code-switching follow Cinque’s (1994) proposal that a universal base underlies adjective 12 13 position, with adjectives assumed to universally precede the noun. On this view, differences in 14 15 word order between English and Spanish follow from overt movement of the noun in Spanish to 16 17 18 a position to the left of the adjective, resulting in the contrasting surface word order outlined 19 20 above, with adjectives preceding the noun in English and following it in Spanish. Cantone and 21 22 MacSwan (2009) find that for Italian-German code-switching, “the language of the adjective 23 24 25 determines the position of the NP relative to the adjective” (p. 268). We can test this descriptive 26 27 generalization, which in the context of Spanish-English code-switching, would mean that 28 29 English adjectives should precede Spanish nouns whereas Spanish adjectives should follow 30 31 32 English nouns. 33 34 There have been attempts to evaluate these two approaches at different grammatical 35 36 37 switching points, but they have been conducted in different language pairs and using different 38 39 methodologies. Hence, it is perhaps not surprising that they have also yielded conflicting results. 40 41 For example, Herring, et al. (2010) examined the predictions of both theories regarding 42 43 44 determiner-noun switches from naturalistic Spanish-English and Welsh-English code-switching 45 46 data. They do not find statistically significant differences between the accuracy of the predictions 47 48 of the two theories. Fairchild and Van Hell (2015) experimentally examined (through a series of 49 50 51 picture naming tasks) the ability of the MP and the MLF to explain determiner-noun switches in 52 53 Spanish-English bilinguals, finding no support for either theory. Focusing specifically on 54 55 adjective-noun order in code-switching, Parafita Couto et al. (2015) designed a study to evaluate 3 the predictions of these two models within Welsh-English mixed nominal constructions by using 4 5 6 a multi-task approach comprising (1) naturalistic corpus data, (2) an elicitation task, and (3) an 7 8 auditory judgment task. They found that Likert-style judgment tasks are not very useful in this 9 10 11 community of code-switchers, due to the stigmatized nature of the phenomenon. However, the 12 13 data from the naturalistic corpus and the elicitation task supported the relative superiority of the 14 15 MLF model. Nevertheless, it was only a small proportion of their data that could distinguish 16 17 18 between the two models, so no definite conclusion could be reached. On the other hand, vanden 19 20 Wyngaerd (2016), basing her analysis on elicited data and judgment tasks, examined word order 21 22 in French-Dutch mixed nominal constructions, finding support for Cantone and MacSwan’s 23 24 25 generalization. 26 27 As described above, the formulation of these theoretical models, particularly the MLF, 28 29 has mainly relied on naturalistic corpora analysis. Multiple studies (Herring et al. 2010, Parafita 30 31 32 Couto, Deuchar & Fusser, 2015; Parafita Couto & Gullberg, 2015; Eppler et al., 2016) have cited 33 34 corpus data with no definitive conclusion with regards to word order within code-switching. This 35 36 37 situation highlights a limitation of corpus data: despite its descriptive richness and ecological 38 39 validity, it is not probative in nature. Other studies have used acceptability judgments (see 40 41 Schütze, 1996), where informants provide a yes or no answer as to whether they accept or reject 42 43 44 a given sentence as “correct” or “acceptable”, or rate how well the sentence “sounds” on a given 45 46 scale (e.g.: “On an ascending scale of 1 to 7, how acceptable do you find the following 47 48 sentence?”). As pointed out by Cowart (1996), this technique is very similar to introspective 49 50 51 judgments that are routinely studied in the field of psychophysics and signal detection theory to 52 53 determine the limits of our sensory system (e.g.: how dim a light our eyes can sense). While 54 55 acceptability judgments afford more control, and, potentially more probative value than corpus 3 data, they are vulnerable to different types of response biases that may obscure their results (see 4 5 6 Parafita Couto et al. 2015 for a discussion of how the stigma associated with code-switching may 7 8 affect grammaticality judgments). 9 10 11 From the signal detection theory literature (e.g.: Gescheider, 1997; Green & Swets, 1988) 12 13 we can learn about some of the vulnerabilities of yes/no and Likert-style judgments. Yes/no 14 15 judgments can suffer from criterion effects, where factors external to the variable under study 16 17 18 may affect the willingness of the informant to report a “yes” or a “no” answer (for a more 19 20 detailed discussion of detection thresholds, see Gescheider, 1997). Scaled responses present a 21 22 different set of challenges, particularly regarding consistency of use of the scale. In order to 23 24 25 position all items along the same scale, informants must not only “calibrate” their ratings along 26 27 the entire range, but also keep a memory record of the ratings on earlier stimuli in order to avoid 28 29 possible shifts in the internal rating scale when new items are presented (Parraga, 2015). Both 30 31 32 yes/no and Likert-style acceptability judgments have attracted criticism because of lack inter- 33 34 rater reliability (Labov, 1972, 1975; Ross, 1979; Stokes, 1974; Bader & Häussler, 2010) and 35 36 37 stability within the same informants at different times or under different testing conditions (e.g: 38 39 Carol, Bever & Pollack, 1981; Nagata, 1988; Snow & Meijer, 1977). 40 41 From very early on, the psychophysics literature (e.g.: Fechner, 1876) identified and 42 43 44 addressed these problems concerning introspective judgments by using paired comparison 45 46 scales, also known as the 2-Alternative Forced Choice (2AFC) method. This is still now 47 48 preferred for quantifying detecting fine-grained differences between value judgments that can 49 50 51 only be made based on subjective criteria (David, 1988; Parraga, 2015). 52 53 54 55 The Two-Alternative Forced Choice Task and the Law of Comparative Judgment. 3 4 5 6 In a 2AFC task, participants are presented with pairs of stimuli and must choose which item is 7 8 “better” according to a specified criterion (without the possibility of a tie); in our case, this 9 10 11 means deciding which of the sentences in a pair is more acceptable or “sounds better”. Besides 12 13 greatly reducing the response bias effects mentioned above (Green & Swets, 1966), comparative 14 15 judgments are a more natural task for participants than rating scales. Nunnally (1976, p. 40) 16 17 18 states that “People simply are not accustomed to making absolute judgments in daily life, since 19 20 most judgments are inherently comparative [...] people are notoriously inaccurate when judging 21 22 the absolute magnitude of stimuli.” Paired comparisons also greatly reduce the demands on 23 24 25 memory required for avoiding shifts in the internal rating scale (Parraga, 2015). 26 27 Gustav Fechner (1876), the pioneer of psychophysics, proposed that the proportion of times 28 29 an items is chosen over another in a series of pairwise comparisons provides a measure of the 30 31 32 distance between the two items in some pleasantness continuum. For example, if item A is 33 34 chosen over item B half the time, both objects are equally pleasant. Correspondingly, if object C 35 36 37 is consistently chosen over item A, then by the same measure, object C is likely to be the most 38 39 pleasant item of the three. Thurstone (1927), with his law of comparative judgment, generalized 40 41 this concept into a measurement model which converts simple pairwise comparisons between 42 43 44 stimuli into one-dimensional quality scores (for further explanation see Bock and Jones, 1968 45 46 and Torgerson, 1958). Thurstone proposed that the proportion of times a stimulus is judged 47 48 greater than another is related to the number of psychological scale units separating the two 49 50 51 sensations. In this method, which is the standard scaling method in many disciplines today, 52 53 pairwise judgments are performed many times by a comparatively large number of subjects, 54 55 resulting in statistically robust results. The outcome of the analysis is a ranking of preference for 3 stimuli along an interval scale that reflects the relative distance between conditions. Moreover, 4 5 6 the results from this analysis can be tested for significance using standard statistical methods 7 8 such as standard errors and ANOVA. Despite the multiple benefits associated with the 2AFC 9 10 11 method attested by its widespread use in psychophysics and other areas of research, its use for 12 13 linguistic acceptability judgments has been conspicuously absent so far. 14 15 In the present study we used both a traditional Likert scale task and a novel application of 16 17 18 the 2AFC task and Thurstone’s (1927) Law of Comparative Judgments to test the predictions on 19 20 adjective-noun order derived from the the MLF and the MP. We directly contrasted sentences 21 22 that reflected (or not) the structures derived from such predictions. This contrast generated four 23 24 25 conditions: sentences that followed the CS pattern predicted by either the MP (but not the MLF), 26 27 or the MLF (but not the MP), both, or neither. Table 1 presents all possible combinations of 28 29 noun- and adjective order and language (for each direction of switch), as well as the predictions 30 31 32 from each model as to whether each combination should be acceptable to bilinguals or not. We 33 34 asked participants to engage in pairwise comparisons of acceptability in all possible 35 36 37 combinations of those conditions and the results were analyzed by applying Thurstone’s law of 38 39 comparative judgment case V to generate an interval scale based on comparisons of pairs of 40 41 code-switched sentences that indicates a ranking of acceptability but also the relative distance 42 43 44 between conditions. 45 46 [Table 1 about here] 47 48 49 50 51 If code-switching is indeed constrained by the grammatical properties proposed by one 52 53 model or the other, the presence of such property should predict the acceptability of a given 54 55 sentence. For example, if the matrix language determines word order within the noun phrase (as 3 proposed by the MLF model), sentences 1, 3, 5, and 7 should show higher acceptability than the 4 5 6 others; if, on the other hand, word order is defined by the language of the adjective itself (as 7 8 according to Cantone and MacSwan´s generalization), sentences 2, 4, 6, and 8 would have higher 9 10 11 scores. Furthermore, if code-switching acceptability is entirely predicted by one of the models, 12 13 there should be no difference between the preferred construction (whichever one may be) and 14 15 those sentences “acceptable” according to both models (exemplars 1 and 5); conversely, the least 16 17 18 favored construction according to that model should not show an advantage over sentences 19 20 rejected by both models (exemplars 4 and 8). 21 22 23 24 25 Experiment 1 26 27 28 29 Participants 30 31 32 A total of 40 early English/Spanish bilinguals took part in this experiment. Most were born in the 33 34 United States (N=34) or moved there before the age of 5 (N=6; mean age when they migrated 35 36 37 =3:6 years). There were 24 men and 16 women and their mean age was 30:3 years (S.D.: 9.6; 38 39 Range: 18-56). All stated that they spoke the Mexican variety of Spanish. The highest level of 40 41 education for 22 participants was high school, 17 had attended at least some college, and 1 42 43 44 attended graduate school. 45 46 Participants were recruited through Amazon Mechanical Turk, an online crowdsourcing 47 48 website that has been shown to be a good source of participants for collection of acceptability 49 50 51 judgments (Gibson, Piantadosi & Fedorenko, 2011). Participants were paid a small fee for 52 53 completing the study and only workers with an acceptance rate of 90% or above and at least 100 54 55 3 tasks completed were allowed to take part in the study (following the guidelines proposed by 4 5 6 Peer, Vosgerau & Acquisti, 2014). 7 8 9 10 11 Language Proficiency and Dominance. In order to take part in the acceptability judgment task, 12 13 each participant completed an English and a Spanish test to confirm their proficiency in both 14 15 languages. These tests were an adaptation from the Online Placement Tests used by Oxford 16 17 18 University’s Language Center (Oxford University Language Center, n.d.). The tests were 19 20 modified to reflect Latin American (rather than Iberian) verb conjugations and vocabulary (e.g.: 21 22 “ustedes” instead of “vosotros” for the second person plural pronoun), and geographical 23 24 25 landmarks to reflect U.S. or Latin American locations (e.g.: New York instead of London). Only 26 27 participants that attained a score of 34 (out of 50) or more were allowed to continue with the 28 29 study. This range of scores is classified as “Higher proficiency” by the Oxford website. 30 31 32 Participant’s average scores were about the same in both languages (English: M =44.9; S.D. 33 34 =2.8; Spanish: M =43.9; S.D. =3.9; t(39) =2.02, p = .2). All participants stated that they were 35 36 37 able to speak Spanish before age four and English by the time they entered elementary school. 38 39 They declared to be slightly more proficient in English than Spanish, with ratings of 4.00 (S.D.: 40 41 0.00) vs. 3.80 (S.D.: 0.46) respectively, in a scale of 1 to 4, where 4 indicates that they are 42 43 44 “Confident in extended conversations”. 45 46 47 48 Materials 49 50 51 Critical trials: We compiled 5 translation-equivalent “base sentences” in Spanish and English 52 53 (please see the Appendix for a list of base sentences). We created two sets of code-switched 54 55 sentences, one in which the first switch went from English to Spanish (E.g.: I like the MARRÓN 3 dress; I like the dress MARRÓN) and another one in which it went from Spanish to English (E.g.: 4 5 6 ME GUSTA EL VESTIDO brown; ME GUSTA EL brown VESTIDO). Each base sentence was 7 8 modified into code-switched forms following four switching patterns going from Spanish to 9 10 11 English and four going from English to Spanish as follows: 12 13 14 15 MP-/MLF+: Sentences that follow the predictions of the Matrix Language Frame but not the 16 17 18 Minimalist Program (E.g.: I like the dress MARRÓN; ME GUSTA EL brown VESTIDO) 19 20 21 22 MP+/MLF-: Sentences that follow the predictions of the Minimalist Program but not the Matrix 23 24 25 Language Frame (E.g.: I like the MARRÓN dress; ME GUSTA EL VESTIDO brown) 26 27 28 29 MP-/MLF-: Sentences that do not follow the predicted pattern from either theory (e.g.: I like the 30 31 32 VESTIDO brown; ME GUSTA EL MARRÓN dress) 33 34 35 36 37 MP+/MLF+: Sentences that follow the predictions of both models (E.g.: I like the brown 38 39 VESTIDO; ME GUSTA EL dress MARRÓN) 40 41 42 43 44 In Spanish, all nouns were masculine or invariable (e.g.: vestido – ‘dress’, baño – ‘bath’, 45 46 cuaderno – ‘notebook’, cantante – ‘singer’, abogado – ‘attorney’) and all adjectives were 47 48 invariable with respect to gender (marron – ‘brown’, caliente – ‘hot’, verde – ‘green’, 49 50 51 nicaragüense – ‘Nicaraguan’, independiente – ‘independent’). Proper names in the sentences 52 53 were chosen so that they were commonplace in both Spanish and English (e.g.: Max, Claudia). 54 55 We avoided using nouns and adjectives whose onset would elicit changes in the preceding 3 indefinite article depending on their position in some conditions (that is, avoid changes such as 4 5 6 “Hugo is a singer estadounidense” vs. “Hugo is an estadounidense singer”). 7 8 9 10 11 Filler trials: In each session we included 72 non-critical trials where the focus of the CS was not 12 13 the adjective but the determiner or the adverb. Some of the results for those trials will be reported 14 15 elsewhere. By including these filler trials plus the quality control trials described below, critical 16 17 18 trials made up only about a third of all pairs seen by participants. This was done to make it 19 20 harder for raters to engage in strategic choices for their response (Cowart, 1996). 21 22 23 24 25 Quality control trials: There were 8 quality-control trials that consisted of sentences with inter- 26 27 sentential code-switches. Half of the sentences had an uncontroversial error that could be easily 28 29 detected if the sentences were read carefully (e.g.: LA PASÉ MUY BIEN, the music *were 30 31 32 excellent). These errors were equally distributed among the following factors: first vs. second 33 34 half of the sentence, English vs. Spanish portion, and type of error (verb tense, number 35 36 37 agreement, gender agreement, and word order). If a participant failed more than 2 of these trials, 38 39 they were removed from the sample and substituted with a new participant. 40 41 42 43 44 Procedure 45 46 The survey was administered online using Qualtrics and testing occurred across 3 separate 47 48 sessions: one in which participants completed the language tests and the background 49 50 51 questionnaire, and two counterbalanced sessions in which participants were presented with NPs 52 53 in which the switch went from English to Spanish and from Spanish to English, respectively. 54 55 Test sessions were about a week apart from each other. Participants were given the choice of 3 reading the instructions in English or Spanish; all but 4 participants chose to complete the 4 5 6 questionnaire in English. Participants were informed that they would see a series of sentences 7 8 and that they were to indicate on a 5-point scale how “permitted” a sentence was according to the 9 10 11 way they would speak to- or hear from another bilingual person. In the scale, a score of 1 stood 12 13 for “never permitted” while 5 stood for “always permitted”. Participants were then presented 14 15 with the 76 code-switched sentences as described above. Each sentence was presented one at a 16 17 18 time and the order of presentation was individually randomized for each participant. Participants 19 20 had to rate each item before progressing to the next one and could not go back to previous 21 22 sentences. 23 24 25 26 27 Results 28 29 Table 2 shows the mean acceptability ratings for each condition (on a scale from 1 to 5). For 30 31 32 sentences where the switch went from English to Spanish, Mauchly’s test indicated that the 33 34 assumption of sphericity had been violated, χ2 (5) = 17.91, p= .003, therefore degrees of freedom 35 36 37 were corrected using the Huynh-Feldt estimates of sphericity (ε = .880). The results show that 38 39 there was a significant effect of CS pattern, F(2.40, 93.57) = 23.52; p < .001. Post-hoc pairwise 40 41 comparisons revealed that there was no significant difference between CS patterns MP+/MLF- 42 43 44 and MP-/MLF+ (p = 1.0), while all other differences between CS patterns were significant (all p 45 46 values ≤ .02). For sentences where the switch went from Spanish to English, Mauchly’s test also 47 48 indicated that the assumption of sphericity had been violated, χ2 (5) = 17.72, p= .003, therefore 49 50 51 degrees of freedom were corrected using the Huynh-Feldt estimates of sphericity (ε = .877). The 52 53 results show that there was a significant effect of CS pattern, F(2.63, 102.57) = 11.05; p < .001. 54 55 Post-hoc pairwise comparisons revealed that the only significant differences were between CS 3 pattern MP+/MLF+ and each of the other CS patterns (all p values ≤ .01), while all other 4 5 6 contrasts were not significant (all p values ≥ .11). 7 8 9 10 11 [Table 2 about here] 12 13 14 In summary, MP+/MLF+ switches are the most preferred in both directions, while MP- 15 16 /MLF- switches are the least preferred in the Spanish to English direction (with a non-significant 17 18 19 trend in the same direction for English to Spanish). There were no significant differences 20 21 between MP+/MLF- and MP-/MLF+ sentences. The results from Experiment 1 show evidence of 22 23 24 a hierarchy of preference for certain CS patterns, particularly in the English to Spanish direction, 25 26 but do not differentiate between the two models of CS under evaluation. The average 27 28 acceptability scores were all close to each other, slightly on the positive side of the middle of the 29 30 31 scale. 32 33 In Experiment 2 we repeated the acceptability judgments for the same code-switched 34 35 sentences with a different group of participants, but this time using a two-alternative forced 36 37 38 choice presentation of the stimulus and using Thurstone’s Law of Comparative Judgment for the 39 40 analysis of the results to find out if this alternative method provides a clearer picture of this 41 42 43 phenomenon. Given the suitability of this method for measuring subtle differences in 44 45 introspective judgments, we expected to obtain a more fine-grained picture of the contrasts under 46 47 study. 48 49 50 51 52 Experiment 2 53 3 A total of 40 early English/Spanish bilinguals living in the United States took part in this 4 5 6 experiment, none of which participated in Experiment 1. Thirty-five participants were born in the 7 8 United States, while the rest immigrated before age 5 (mean age when they migrated =3:7 years). 9 10 11 There were 22 men and 18 women and their mean age was 29:5 years (S.D.: 8.8; Range: 18-57). 12 13 All stated that they spoke the Mexican variety of Spanish. The highest level of education for 21 14 15 participants was high school, 17 had attended at least some college, and 2 attended graduate 16 17 18 school. Participants were also recruited through Amazon Mechanical Turk and completed the 19 20 same questionnaire and language tests as those in Experiment 1. 21 22 24 25 Language Proficiency and Dominance. Participant’s average scores in the language proficiency 26 27 tests were slightly but significantly better for English (M =45.2; S.D. =2.8) than for Spanish (M 28 29 =43.7; S.D. =3.8); t(78) =2.01, p < .05. All participants stated that they were able to speak 30 31 32 Spanish before age four and English by the time they entered elementary school. They declared 33 34 to be slightly more proficient in English than Spanish, with ratings of 3.98 (S.D.: 0.16) vs. 3.68 35 36 37 (S.D.: 0.47) respectively. 38 40 41 Materials 42 43 44 Critical trials: Participants were presented with the same critical sentences as Experiment 1, but 45 46 they were compared pairwise among themselves to yield 30 judgments for each direction of the 47 48 switch (English to Spanish and Spanish to English). Sentences were contrasted only within their 49 50 51 own directionality set in two separate sessions, that is, code-switched sentences going from 52 53 English to Spanish were only contrasted with other sentences also going from English to Spanish 54 55 and so on. This was done in order to control for possible preferences in the directionality of the 3 switch (Parafita Couto et al. 2015) and insulate the effect of noun-adjective order, which is the 4 5 6 focus of the present study. This was also a pragmatic decision because the combinatorial 7 8 outcome of comparing all possible pairs of the resulting 8 variations for each sentence made the 9 10 11 experiment unfeasible (it would have generated 28 combinations per sentence instead of 6). 12 13 14 15 Monolingual trials: We presented a separate group of 40 early Spanish/English bilinguals with 16 17 18 monolingual versions (either English or Spanish) of the base sentences used in this experiment. 19 20 English and Spanish sentences were presented separately and the order of presentation of 21 22 language blocks was counterbalanced across participants. The monolingual sentences were 23 24 25 presented in pairs in which the adjective was pre- or post-nominal and asked them to choose 26 27 which one was more acceptable. In Spanish, 99% of responses indicated a preference for a post- 28 29 nominal adjective, while in English 98.5% of responses indicated a preference for pre-nominal 30 31 32 adjectives. 33 34 35 36 37 Filler trials: In each session we included 72 non-critical trials where the focus of contrast 38 39 between choices in a pair was not the adjective but the determiner or the adverb so that the 40 41 critical trials made up only about a third of all pairs seen by participants. Some of the results for 42 43 44 the filler trials will be reported elsewhere. 45 46 47 48 Quality control trials: There were 12 quality-control trials that consisted of pairs of sentences 49 50 51 with inter-sentential code-switches. One of the members of each pair had an uncontroversial 52 53 error that could be easily detected if the sentences were read carefully. If a participant failed 3 more than 2 of these trials, they were removed from the sample and substituted with a new 4 6 participant. 7 8 9 10 11 Procedure 12 13 The survey was administered online using Qualtrics and testing occurred across 3 separate 14 15 sessions: one in which participants completed the language tests and the background 16 17 18 questionnaire, and two in which they completed half of the 2AFC trials. Participants were 19 20 presented with NPs in which the switch went from English to Spanish in one session and from 21 22 Spanish to English in the other, counterbalancing across participants. At the beginning of each 23 24 25 session participants were given the choice of reading the instructions in English or Spanish; all 26 27 but 5 participants chose to complete the questionnaire in English. The instructions informed 28 29 participants that they would see a series of sentence pairs, and asked them to pick the one closer 30 31 32 to the way they would speak to another bilingual person. They were asked to make a choice even 33 34 if both sentences sounded “right” or both sounded “wrong”. Then participants were presented 35 36 37 with the 114 pairs of code-switched sentences described above. The pairs of sentences were 38 39 presented one at a time and the order of presentation of the pairs (critical trials, fillers, and 40 41 quality control trials), as well as the order of each sentence within each pair, was individually 42 43 44 randomized for each participant. Participants had to make a choice for each trial before 45 46 progressing to the next one and could not go back to previous sentences. 47 48 50 51 Results 52 53 Participants’ responses were analyzed using Thurstone’s (1927) Law of Comparative Judgment, 54 55 Case V, which analyzes participants’ pairwise comparison of the stimuli to generate a ranking of 3 preference among conditions as well as a measure for relative comparison between them. The 4 5 6 results of Thurstone’s analysis must be interpreted within the context of signal detection theory 7 8 (c.f. Cowart, 1996): every measurement includes a noise component that usually follows a 9 10 11 normal distribution. If we repeatedly measure the same construct, we will end up with a normal 12 13 distribution centered on the likely “ideal” value of the magnitude, with a lesser probability of 14 15 obtaining large errors of judgment than small errors in either direction. Thurstone’s “measure” 16 17 18 provides the center or mean of such a normal distribution, which Thurstone called “discriminal 19 20 dispersions”, for each condition. These measures can be interpreted values on an interval scale 21 22 that represents a psychological continuum (in our case, the acceptability of the sentences). The 23 24 25 unit of measurement along that scale is defined as the standard deviation of the distribution. 26 27 (Brown & Peterson, 2009). 28 29 30 31 32 Table 3 shows the rank order and Thurstone’s measure for each condition. The measure 33 34 values are relative to the pattern with the lowest acceptability (which is by convention set to 0). 35 36 37 The 95% confidence interval for this set of data was 0.071. A within-participants ANOVA 38 39 revealed a significant effect of sentence type both for sentences in which the switch went from 40 41 English to Spanish (F(3,796)=526.22, p < .001) as well as when it went from Spanish to English 42 43 44 (F(3,796) =184.48, p < .001). Post-Hoc Tukey HSD tests indicated that all contrasts were 45 46 significant (all p values < .01). 47 48 49 50 51 [Table 3 about here] 52 53 54 The results of Experiment 2 show a clear hierarchy of preference for the CS patterns tested. 3 the predictions of the MP over MLF. It is important to point out, however, that MP-/MLF+ 4 5 6 condition is preferred over the MP-/MLF- condition, and that, similar to the results of 7 8 Experiment 1, the preferred condition was that in which both constraints were satisfied. 9 10 11 12 13 Discussion 14 15 In this study, we investigated the contrasting predictions derived from the MLF and the MP 16 17 18 (specifically, the generalization by Cantone & MacSwan, 2009) regarding the mechanisms 19 20 underpinning code-switching between the noun and the adjective. In two critical conditions, the 21 22 predictions for adjective positioning made by the MLF and the MP differed. The MLF predicts a 23 24 25 violation when the adjective position is incompatible with the word order of the sentence’s 26 27 Matrix Language, while the MP predicts a violation when the adjective position is disallowed by 28 29 the language of the adjective. While Experiment 1 (with traditional Likert scales) didn’t yield a 30 31 32 clear differentiation between the predictive power of each model, Experiment 2 indicated an 33 34 advantage for the MP. However, the difference between the MP+/MLF- and MP+/MLF+ 35 36 37 conditions points towards an additive effect in which both the language of the verb and the 38 39 language of the adjective are used to determine word order in the NP. If decisions were based 40 41 only on the language of the adjective (as predicted by the MP), there would be no difference 42 43 44 between MP+/MLF- and MP+/MLF+ because the language of the verb would have nothing to 45 46 add above and beyond what can be explained by the language of the adjective in regards to the 47 48 acceptability of a given sentence. On the other hand, the fact that there is a difference between 49 50 51 MP-/MLF+ and MP-/MLF- confirms that the language of the verb contributes to the 52 53 acceptability decisions of informants. 3 Our results constitute strong support for Eppler et al’s (2016, p. 1) suggestion that “advances 4 5 6 in our understanding of grammaticality in code-switching will be achieved by combining the 7 8 insights of different theoretical models instead of considering them in isolation”. Our findings 9 10 11 could be interpreted as evidence for a potential parallel between the MLF and the MP, as already 12 13 pointed out by Radford, Kupisch, Köppe, and Azzaro (2007), who argued that it is possible to 14 15 equate the MLF’s notion of morphosyntactic frame with the MP’s notion of Phase (a derivational 16 17 18 unit in minimalist syntax). The rationale behind this possibility is that “the head of a phase is 19 20 responsible (via a form of selection) for “handing over” functional features to subordinate items 21 22 within the phase” (Radford et al, 2007, p. 245). Hence, our data could be interpreted as support 23 24 25 for the suggestion that the MLF may dominate the whole Complementizer Phrase (CP) phase, 26 27 thus making both theoretical models compatible and complementary. 28 29 Additionally, our results seem to provide evidence against the more general proposal within 30 31 32 Minimalism that the DP (Determiner Phrase) is a separate phase (Svenonius, 2004, Hiraiwa, 33 34 2005), since information outside the DP needs to be taken into account when building a nominal 35 36 37 construction. 38 39 In addition to understanding the mechanisms involved in solving conflict-sites in the grammar of 40 41 bilinguals, the use of the 2AFC task provided robust and unprecedented insights into the validity 42 43 44 of theories of code-switching. As discussed above (and more extensively in the psychophysics 45 46 literature), the 2AFC task represents an improvement over other methods of measuring 47 48 introspective judgments and provides clear, interpretable data that can be used for directly 49 50 51 contrasting the predictions of linguistic theoretical models. This approach opens the way for 52 53 systematic testing of predictions from linguistic theory in general, especially in cases where 54 55 corpus data appear insufficient to test all possible combinations of theoretical predictions. 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 3 4 5 REFERENCES 6 7 8 Bader, M., & Häussler, J. (2010). Toward a model of grammaticality judgments. Journal of 9 10 Linguistics, 46, 273-330. 11 12 13 Bock, R. D., & Jones, L.V. (1968). The measurement and prediction of judgment and choice. 14 15 San Francisco, CA: Holden-Day. 16 17 Brown, T.C., Peterson, G.L. (2009). An enquiry into the method of paired comparison: 18 19 20 reliability, scaling, and Thurstone’s Law of Comparative Judgment. Fort Collins, CO: 21 22 U.S. Department of Agriculture, Forest Service. 23 24 Bullock, B.E. & Toribio, A. J. (Eds.) (2009). The Cambridge Handbook of Linguistic Code- 25 26 27 switching. Cambridge: Cambridge University Press. 28 29 Cantone, K. F., & MacSwan, J. (2009). The syntax of DP-internal codeswitching. In L. Isurin, D. 30 31 Winford & K. de Bot (eds.) Multidisciplinary Approaches to Codeswitching (pp. 243-78). 32 33 34 Amsterdam: John Benjamins Publishing. 35 36 Carol, J.M., Bever, T.G., & Pollack, C.R. (1981). The non-uniqueness of linguistic intuitions. 37 38 39 Language, 57, 368-383. 40 41 Cinque, G. (1994). On the evidence for partial N-movement in the Romance DP. In G. Cinque, J. 42 43 Koster, J-Y Pollock, L. Rizzi & R. Zanuttini (Eds.), Paths towards Universal Grammar. 44 45 46 Studies in honor of Richard S. Kayne (pp. 85-110). Washington, DC: Georgetown 47 48 University Press. 49 50 Cowart, W. (1996). Experimental Syntax: Applying Objective Methods to Sentence Judgments. 51 52 53 California: Sage Publications Inc. 54 55 David, H.A. (1988). The method of paired comparisons (2nd ed.). New York, NY: Oxford 3 Davies, P. (2010). Identifying word-order convergence in the speech of Welsh-English 4 5 6 bilinguals. PhD dissertation, Bangor University. 7 8 Davies, P., & Deuchar, M. (2010). Using the Matrix Language Frame model to measure the 9 10 11 extent of word order convergence in Welsh-English bilingual speech. In A. Breitbarth, C. 12 13 Lucas, S. Watts & D. Willis (eds.) Continuity and change in grammar (pp. 77–96). 14 15 Philadelphia: John Benjamins. 16 17 18 Deuchar, M. (2012). Code Switching. In Chapelle, C.A. (ed.) The Encyclopedia of Applied 19 20 Linguistics (pp. 664-675). Chichester, UK: Wiley-Blackwell. 21 22 Deuchar, M. (2006). Welsh-English code-switching and the Matrix Language frame model. 23 24 25 Lingua, 116, 1986-2011. 26 27 Deuchar, M. and Davies, P. (2009). Code-switching and the future of Welsh. International 28 29 Journal of the Sociology of Language, 195, 15–38. 30 31 32 Eppler, E.D., Luescher, A. & Deuchar, M. (2016) Evaluating the predictions of three syntactic 33 34 frameworks for mixed determiner–noun constructions. Corpus Linguistics and Linguistic 35 36 37 Theory, Advance online publication. DOI: 10.1515/cllt-2015-0006 38 39 Fairchild, S., & Van Hell, J. (2015). Determiner-noun code-switching in Spanish heritage 40 41 speakers. Bilingualism: Language and Cognition. Advance online publication. DOI: 42 43 44 10.1017/S1366728915000619. 45 46 Fechner, G. T. 1876. Vorschule der Aesthetik, Vol. 2. Leipzig: Breitkopf & Härtel. 47 48 Gescheider, G. (1997). Psychophysics: the fundamentals (3rd ed.). New Jersey: Lawrence 49 50 51 Erlbaum Associates. 3 Gibson, E., Piantadosi, S., & Fedorenko, K. (2011). Using Mechanical Turk to Obtain and 4 5 6 Analyze English Acceptability Judgments. Language and Linguistics Compass, 5, 509- 7 8 524. 9 10 11 Green, D. M., & Swets, J. A. (1988). Signal Detection Theory and Psychophysics. Los Altos, 12 13 CA: Penninsula. 14 15 Guzzardo Tamargo, R.E., Mazak, C., & Parafita Couto, M.C. (Eds.) (2016) Spanish-English 16 17 18 CodeSwitching in the Caribbean and the U.S. Amsterdam: John Benjamins. 19 20 Gullberg, M. & Parafita Couto, M.C. (2016) An integrated perspective on code-mixing patterns 21 22 beyond doubling? Bilingualism: Language and Cognition. Advance online publication. 23 24 25 DOI: 10.1017/S1366728916000080 26 27 Herring , J., Deuchar, M., Parafita Couto, M.C., & Moro Quintanilla, M. (2010). "When I went 28 29 to Canada, I saw the madre": evaluating two theories' predictions about codeswitching 30 31 32 between determiners and nouns using Spanish-English and Welsh-English bilingual 33 34 corpora, International journal of bilingual education and bilingualism, 13, 553-573. 35 36 37 Hiraiwa, K. (2005) Dimensions of Symmetry in Syntax: Agreement and Clausal Architecture. 38 39 PhD dissertation, Massachusetts Institute of Technology. 40 41 Jake, J. L., Myers-Scotton, C. M., & Gross, S. (2005). A response to MacSwan (2005): Keeping 42 43 44 the Matrix Language. Bilingualism: Language and Cognition, 8, 271–276. 45 46 Labov, W. (1972). Some principles of linguistic methodology. Language in Society, 1, 97-120. 47 48 Labov, W. (1975). What is a linguistic fact? Lisse: Peter de Ridder. 49 50 51 MacSwan, J. (2009). Generative approaches to codeswitching. In A. J. Toribio & B. E. Bullock 52 53 (eds.), Cambridge Handbook of Linguistic Codeswitching (pp. 309-335). Cambridge, 54 55 UK: Cambridge University Press. 3 MacSwan, J. (2005a) Codeswitching and generative grammar: A critique of the MLF model and 4 5 6 some remarks on “modified minimalism.” Bilingualism: Language and Congition, 8,1- 7 8 22. 9 10 11 MacSwan, J. (2005b) Comments on Jake, Myers-Scotton and Gross’s response: There is no 12 13 “matrix language.” Bilingualism: Language and Cognition, 8, 277-284. 14 15 MacSwan, J. (2000) The architecture of the bilingual language faculty: Evidence from 16 17 18 codeswitching. Bilingualism: Language and Cognition, 3, 37–54. 19 20 Myers-Scotton, C. (2002). Contact Linguistics: Bilingual Encounters and Grammatical 21 22 Outcomes. Oxford, UK: Oxford University Press. 23 24 25 Nagata, H. (1988). The relativity of linguistic intuition: The effect of repetition on 26 27 grammaticality judgments. Journal of Psycholinguistic Research, 171, 1-17. 28 29 Nunnally, J.C. (1976). Psychometric theory. New York: McGraw Hill. 30 31 32 Oxford University Language Centre (n.d.). “Placement Tests.” lang.ox.ac.uk. 33 34 http://www.lang.ox.ac.uk/tests/index.html (accessed July 1st, 2015) 35 36 37 Parafita Couto, M. C., Deuchar, M., & Fusser, M. (2015). How do Welsh-English bilinguals deal 38 39 with conflict? Adjective-noun order resolution. In G. Stell, & K. Yakpo (eds.) Code- 40 41 Switching Between Structural and Sociolinguistic Perspectives (pp. 65-84). Berlin: De 42 43 44 Gruyter. 45 46 Parafita Couto, M.C., & Gullberg, M. (2015). Code-switching within the noun phrase. Evidence 47 48 from three corpora. Paper presented at the 10th International Symposium on 49 50 51 Bilingualism. Rutgers University, New Jersey 20-24 May, 2015. 3 Parraga, C.A. (2015). Perceptual Psychophysics. In G. Cristobal, M. Keil & L. Perrinet (eds.) 4 5 6 Biologically-Inspired Computer Vision: Fundamentals and Applications (pp. 81-108). 7 8 New York, NY: Wiley. 9 10 11 Peer, E., Vosgerau, J., & Acquisti, A. (2014). Reputation as a sufficient condition for data quality 12 13 on Amazon Mechanical Turk. Behavior Research Methods, 46, 1023-1031. 14 15 Poplack, S., & Meechan, M. (1998). How languages fit together in codemixing. International 16 17 18 Journal of Bilingualism, 2, 127-38. 19 20 Radford, A., Kupisch, T., Köppe, R., & G. Azzaro, G. (2007). Concord, Convergence and Crash- 21 22 Avoidance in bilingual children. Bilingualism: Language and Cognition, 10, 239-256. 23 24 25 Ross, J.R. (1979). Where’s English? In C.J. Fillmore, D. Kemper & W.S. Wang (eds.) Individual 26 27 differences in language ability and language behavior (pp. 127-163). New York: 28 29 Academic Press. 30 31 32 Schütze, C.T. (1996). The empirical base of linguistics: Grammaticality judgments and linguistic 33 34 methodology. Chicago, Il: University of Chicago Press. 35 36 37 Snow, C., & Meijer, G. (1977). On the secondary nature of syntactic intuitions. In S. Greenbaum 38 39 (ed.) Acceptability in language (pp. 163-177). The Hague, the Netherlands: Mouton. 40 41 Stokes, W. (1974). All of the work on quantifier-negation isn’t convincing. In M.W. La Galy, 42 43 44 R.A. Fox & A. Bruck (eds.) Papers from the tenth regional meeting, Chicago Linguistic 45 46 Society (pp. 692-700). Chicago: Chicago Linguistic Society. 47 48 Svenonius, P. (2004). On the edge. In D. Adger, C. de Cat & G. Tsoulas (eds.) Peripheries: 49 50 51 Syntactic edges and their effects (pp. 261-287). Dordrecht: Kluwer. 52 53 Thurstone, L. (1927). A Law of Comparative Judgment. Psychological Review, 34, 273-286. 54 55 Torgerson, W.S. (1958). Theory and methods of scaling. New York, NY: John Wiley & Sons. 3 vanden Wyngaerd, E.V. (2016) The adjective in Dutch-French codeswitching: Word order and 4 5 6 agreement. International Journal of Bilingualism. Advance online publication. DOI: 7 8 10.1177/1367006916632302 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 3 Table 1. Prediction of acceptability for code-switched constructions according to the MP and MLF 4 models. 5 6 English→Spanish Spanish→English 7 Predicted Predicted Exemplar Exemplar 8 Acceptability Acceptability 9 (1) I like the marrón dress MP-/MLF+ (5) Me gusta el vestido brown MP-/MLF+ 10 11 (2) I like the dress marrón MP+/MLF- (6) Me gusta el brown vestido MP+/MLF- 12 13 (3) I like the vestido brown MP-/MLF- (7) Me gusta el marrón dress MP-/MLF- 14 (4) I like the brown vestido MP+/MLF+ (8) Me gusta el dress marrón MP+/MLF+ 15 16 17 MP+ =Acceptable according to the Minimalist Program; MP- =Not acceptable according to the Minimalist Program; 18 MLF+ =Acceptable according to the Matrix Language Frame model; MLF- =Not acceptable according to the Matrix 19 Language Frame model. 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 3 Table 2. Mean acceptability rating for each condition 4 5 6 Condition English→Spanish Spanish→English 7 MP+/MLF+ 3.71 (S.D. 0.93) 3.71 (S.D. 0.86) 8 9 MP+/MLF- 3.08 (S.D. 0.90) 3.25 (S.D. 0.96) 10 MP-/MLF+ 3.24 (S.D. 0.94) 3.12 (S.D. 1.04) 11 12 MP-/MLF- 2.60 (S.D. 0.88) 2.89 (S.D. 1.04) 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 3 Table 3. Ranking and measure for each condition 4 5 6 Thurstone’s Measure 7 Rank Condition English→Spanish Spanish→English 8 9 1 MP+/MLF+ 2.26 1.32 10 2 MP+/MLF- 1.29 0.82 11 12 3 MP-/MLF+ 0.90 0.50 13 4 MP-/MLF- 0 0 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 3 4 5 APPENDIX 6 7 8 9 10 Monolingual base sentences 11 12 13 English Spanish 14 15 I like the brown dress Me gusta el vestido marrón 16 17 I need a hot bath Necesito un baño caliente 18 19 Samuel wrote it all in his green notebook Samuel lo escribió todo en su cuaderno verde 20 21 Hugo is a Nicaraguan singer Hugo es un cantante Nicaragüense 22 23 24 They had to hire an independent attorney Tuvieron que contratar un abogado independiente 25 26 27 Code-Switched Critical Sentences 28 29 MP-/MLF+ 30 31 English → Spanish Spanish → English 32 33 34 I like the marrón dress Me gusta el vestido brown 35 36 I need a caliente bath Necesito un baño hot 37 Samuel wrote it all in his verde notebook Samuel lo escribió todo en su cuaderno green 38 39 Hugo is a Nicaragüense singer Hugo es un cantante Nicaraguan 40 41 They had to hire an independiente attorney Tuvieron que contratar un abogado independent 42 43 44 45 MP+/MLF- 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 3 4 5 MP-/MLF- 6 7 Spanish → English 8 English → Spanish 9 10 I like the vestido brown Me gusta el marrón dress 11 I need a baño hot Necesito un caliente bath 12 13 Samuel wrote it all in his cuaderno green Samuel lo escribió todo en su verde notebook 14 15 Hugo is a cantante Nicaraguan Hugo es un Nicaragüense singer 16 17 They had to hire an abogado independent Tuvieron que contratar un independiente attorney 18 19 20 21 MP+/MP+ 22 23 English → Spanish Spanish → English 24 25 I like the brown vestido Me gusta el dress marrón 26 27 I need a hot baño Necesito un bath caliente 28 29 Samuel wrote it all in his green cuaderno Samuel lo escribió todo en su notebook verde 30 31 Hugo is a Nicaraguan cantante Hugo es un singer Nicaragüense 32 33 They had to hire an independent abogado Tuvieron que contratar un attorney independiente 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60