Available online at www.sciencedirect.com Acta Psychologica 128 (2008) 490–500 www.elsevier.com/locate/actpsy Activation of lexical and syntactic target language properties in translation C. Ruiz, N. Paredes, P. Macizo, M.T. Bajo * Department of Psychology, University of Granada, Campus de Cartuja, 18071 Granada, Spain Received 16 January 2007; received in revised form 31 July 2007; accepted 2 August 2007 Available online 19 September 2007 Abstract Is reading for translation equal to reading in monolingual contexts? Horizontal/parallel theories of translation propose that normal reading and reading for translation differ because the translator engages in partial reformulation while reading for translating the source text. In contrast, vertical/serial theories assume that the translators first extract the meaning of the message, and only then they proceed to reformulate it. In two experiments, we manipulated lexical and syntactic properties of the target language (TL) while translators read for repetition or for translation. On-line sentence comprehension was affected by the lexical frequency of words in the TL (Experiment 1) and the syntactic congruency between the source language (SL) and TL sentences (Experiment 2). However, the influence of lexical and syntactic TL properties was restricted to the reading for translation task. According to our results, the horizontal view of translation includes code-to-code links between the SL and TL involving at least the lexical and syntactic level of processing.  2007 Elsevier B.V. All rights reserved. PsycINFO classification: 2340 Keywords: Translation; Lexical frequency; Syntactic structure 1. Introduction Translation involves rephrasing a message expressed in one language, the source language (SL), into another language, the target language (TL). Although it is possible to distinguish among many translation tasks depending on particular variations of the modality of the input (e.g., visual, auditory), the modality of the output (e.g., written, oral) and the temporal parameters relating the input and output (e.g., simultaneous, consecutive, self-paced) (see Christoffels & de Groot, 2005; Macizo & Bajo, in press), most theories of translation agree that there are three major processes involved in all these tasks: analysis and understanding of SL, switching between two linguistic codes, and production of the message in the TL. Thus, there is a general agreement that the translator, as all normal readers * Corresponding author. Tel.: +34 958243769; fax: +34 958246239. E-mail address: mbajo@ugr.es (M.T. Bajo). 0001-6918/$ - see front matter  2007 Elsevier B.V. All rights reserved. doi:10.1016/j.actpsy.2007.08.004 or listeners, performs lexical, syntactic and discourse analyses in order to construct a mental representation of the source message. The theories also agree that planning and lexical selection are needed to produce the target message. However, there are different theoretical positions about the nature of the reformulation or code switching process. Specifically, the disagreement concerns the way in which comprehension and reformulation are articulated and coordinated. The horizontal/parallel view of translation defends that comprehension and reformulation occur in parallel, that is, during comprehension of the SL, lexical and syntactic properties of both languages are accessed (Gerver, 1976). In contrast, the vertical view proposes that comprehension and reformulation occurs in a sequential manner, so that reformulation into the TL takes place only after comprehension of the SL has ended. When the input message is processed, its superficial form is lost (Seleskovitch’s deverbalization theory, 1976) so that reformulation will only be C. Ruiz et al. / Acta Psychologica 128 (2008) 490–500 based on the meaning representation extracted from the SL message. According to this view, comprehension for translation does not differ from normal monolingual comprehension since comprehension and reformulation occur at different stages. In contrast, the horizontal view argues that comprehension in translation is different from normal comprehension since comprehension and reformulation cooccur in time and, therefore, they influence each other. In addition, while the horizontal perspective defends the existence of direct connections between the linguistic forms of both languages at different levels of analysis (lexical and syntactic), vertical/serial theories do not include code-tocode direct connections. Still, a third hybrid approach proposes that both serial and parallel models of translation only represent two possible routes to a translation response, so that any of these two processes or strategies can be used by any bilingual; experience and training determines which processes are used the most (Paradis, 1994; Paradis, Goldblum, & Abidi, 1982). However, within this hybrid approach there is some bias regarding the relative importance of both translation strategies. Thus, horizontal, code-to-code translation, is considered more probable in amateur translators (Seleskovitch, 1976, 1999) or in conditions where performance temporarily breaks down (Paradis, 1994), but it will rarely occur in experienced translators. In addition, recoding is also more probable when the SL linguistic form is continuously present (as in text-to-text translation), but it will rarely take place in conditions where the SL linguistic form is transient (as in consecutive interpretation) (Seleskovitch, 1976). Even though the parallel/serial issue has been a topic of discussion within the translation literature (Gile, 1991; Paradis, 1994; Paradis et al., 1982; Seleskovitch, 1976), very few empirical studies have attempted to disentangle the relative importance of the two types of translation processes and most translation and interpreting textbooks emphasise deverbalization as the only way of achieving quality translations (Lederer, 1994, 2003). However, there is some support for the influence of direct code-to-code connections in translation. For example, Isham (1994) and De Bot (2000) reported that recently presented SL words and syntactic structures influence lexical and syntactic selection in the target language. Similarly, many studies with bilingual participants have shown that during comprehension some properties of the non-presented language are activated in parallel. For example, properties of L2 such as the cognate status of a word or the orthographic similarity between languages influence recognition and lexical decision of L1 words (Dijkstra, Timmermans, & Schriefers, 2000; Garcı́a-Albea, Sánchez-Casas, & Valero, 1996; Van Hell & Dijkstra, 2002). Recently, Macizo and Bajo (2006) directly investigated whether lexical entries of the TL get activated while reading SL sentences for later translations. Professional translators read Spanish (L1) sentences word-by-word under instructions to either repeat the sentences in Spanish (reading in 491 a monolingual context) or to translate them into English (L2). The cognate status of words located at the beginning or the end of the sentences was manipulated. The results indicated that when the translators were asked to read the sentences for later translation, the reading times for cognate words (zebra in English, cebra in Spanish) were faster than the reading times for control words located at the end of the sentence (caterpillar in English, oruga in Spanish). However, when they were asked to read and repeat the sentences, no facilitation effects were found for cognate words. This different pattern of cognate effects when comparing comprehension in a monolingual context and comprehension in a translation context directly shows that experienced translators activate the TL lexical entries while processing the meaning of the SL sentences that had to be translated later. Thus, this experiment provides evidence supporting that, when reading for translation, units in the SL and the TL become activated in parallel. However, because they manipulated the cognate status of the words as the critical property of the TL and it has been suggested that cognate words have a special status in the bilingual’s mental lexicon (De Groot, 1992), it can be argued that these results only hold when the cognate status is the critical manipulation to observe TL activation. In addition, because the cognate effect might be located at the lexical or semantic level of processing (De Groot & Nas, 1991), the data do not clearly indicate the nature of the TL activation during the SL sentence processing. The purpose of Experiment 1 was to provide further evidence that at least, lexical properties of the TL become active during SL comprehension. In Experiment 1 we replicated the Macizo and Bajo (2006) procedure but we manipulated the frequency of usage of the translation equivalents (TL) of the presented SL words. Lexical frequency has been shown to affect performance in a broad variety of tasks, including picture or word naming (e.g., Forster & Chambers, 1973), lexical decision (e.g., Frederiksen & Kroll, 1976), and semantic and syntactic categorisation (e.g., Monsell, 1985); and its effect has been clearly located at the lexical level (e.g., Balota & Chumbley, 1984; Seidenberg, 1995). Hence, if the TL lexical entries are accessed while reading and understanding the SL sentences to later translate them, we should find facilitation effects for more frequent TL equivalents relative to less frequent ones, even though their frequency in the SL was held constant. Therefore, Experiment 1 will allow us to extend Macizo and Bajo (2006) findings to non-cognate words and to clearly specify the lexical nature of the effect. In addition, in Experiment 2 we aimed to explore if activation of the TL also extends to the syntactic level. In a study, Hartsuiker, Pickering, and Veltkamp (2004) reported a syntactic priming effect in Spanish-English bilinguals. Studies using syntactic priming in monolingual contexts have shown that the use of a specific syntactic form affects subsequent syntactic processing (Bock, 1986; Branigan, Pickering, & Cleland, 2000). In the Hartsuiker et al. study, bilingual participants listened to the descrip- 492 C. Ruiz et al. / Acta Psychologica 128 (2008) 490–500 tion of a picture in Spanish and they had to later describe a different picture in English. Their results showed that participants tended to use the same syntactic structure that they previously heard (e.g., passive form) when describing pictures in English. Therefore, syntactic processing in one language may influence syntactic processing in another. In Experiment 2, we manipulated the syntactic congruency between the SL and TL sentences. Again, professional translators read Spanish sentences word-by-word with the instructions to understand and repeat the sentences or to understand and translate them. If understanding for translation involves parallel searches for syntactic structures in the TL, congruent SL/TL structures should facilitate the reading process. Therefore, we expected that congruency would have an effect when the translators were asked to read for translation, but it would not when they were asked to read within a monolingual context. Note that in both experiments, in order to control for the speech planning processes necessarily involved in overt translations, we asked the participants to repeat the sentences in the language of presentation (Spanish) when they were in the control non-translation condition. This request for repetition may make the task more demanding than normal reading. And therefore, the reading for repetition conditions of our two experiments cannot be considered exactly as normal reading as this increment in demands may influence the comprehension processes. 2. Experiment 1 As indicated, the purpose of Experiment 1 was to explore whether there is lexical activation of the TL during reading for translation. In this experiment, professional translators performed self-paced reading of Spanish sentences either for repeating them into Spanish or for translating them into English. Apart from the type of task (reading for repetition/reading for translation), we manipulated lexical frequency of critical words in the TL (low/ high frequency words) while lexical frequency of words in the SL was held constant. If the lexical entries of the TL are being activated while comprehending the SL, the presence of high frequency words should facilitate the reading processes when participants are asked to translate the sentences. In contrast, if comprehension and reformulation proceed serially, the presence of TL high frequency words in the sentences should not facilitate comprehension in either of the two reading conditions. 2.1. Method 2.1.1. Participants and design The type of reading (reading for repetition or reading for translation) and the TL lexical frequency of the SL words (high or low lexical frequency) were manipulated within-participants. Sixteen professional translators (with more than 2 years of experience) participated in the exper- iment. Their mean age was 33.25 years (SD = 5.49). All of them were native speakers of Spanish (their L1), but were very fluent in English (their L2). They were paid for participating in the study. Before performing the actual experiment, all the participants were asked to perform a Spanish version of the Reading Span Test (Daneman & Carpenter, 1980). The mean span for the group was (4.19, SD = 0.92). Following criteria used by other authors (e.g., Miyake, Just, & Carpenter, 1994), our participants were considered as having a high span score (reading span scores higher than 3.5). In addition, the translators completed a language history questionnaire to assess their fluency in L1 and L2 and they also rated their professional experience in translation and interpreting tasks. The results of these questionnaires are shown in Table 1. 2.1.2. Materials A set of 184 Spanish words with similar frequency of usage in Spanish (source language, SL, frequency) were selected. These words were divided based on the frequency of usage of their English translation (target language, TL, frequency) in two groups of 92 words each. The two groups of words varied in their TL frequency, low TL frequency (24.18, SD = 13.73), high TL frequency (161.20, SD = 130.65), t(182) = 10.00, p < .05 (Kucera & Francis, 1967, based on a count of one million English words). The natural log transformation of the low TL frequency and the high TL frequency was 2.92 (SD = 0.90) and 4.85 (SD = 0.63), respectively. However, the low and high TL frequency groups were similar in their SL frequency, 97.52 (SD = 25.76) and 105.23 (SD = 33.51), respectively, t(182) = 1.75, p > .05 (Alameda & Cuetos, 1995, based on a count of two million Spanish words). The natural log transformation of the Spanish frequency was 4.56 (SD = 0.19) and 4.62 (SD = 0.26) for the low TL frequency group and the high TL frequency group, respectively. Table 1 Characteristics of participants in the experiments Language proficiency questionnaire Speech fluency Speech comprehension Writing proficiency Reading proficiency Professional experience in translation tasks Professional experience in interpreting tasks Mean L1 (SD) Mean L2 (SD) 9.12 9.56 9.31 9.56 6.37 7.68 7.43 8.12 (1.02) (0.72) (1.07) (0.72) (2.52) (1.40) (1.54) (1.25) 5.75 (1.43) 8.12 (0.71) Note: The self-report ratings in the language proficiency questionnaire for L1 (Spanish) and L2 (English) ranged from 1 to 10 where 1 was not fluent and 10 was very fluent. The self-report rating in the professional experience for translation and interpreting tasks ranged from 1 to 10 where 1 indicated low experience and 10 indicated high experience. Standard deviation is given in brackets. C. Ruiz et al. / Acta Psychologica 128 (2008) 490–500 Ninety-two Spanish sentences were created embedding a low TL frequency word and a high TL frequency word in each sentence. In addition, two versions of each sentence were created: (1) in the LH (Low–High) version, a low TL frequency word was at the initial position of the sentence (one of the first three words of the sentence) and a high TL frequency word was at the end of the sentence (one of the last three words of the sentence) and (2) in the HL (High–Low) version, the same two critical words appeared in the reverse order. The two types of sentences were identical in their structure and content and only differed in the position of the critical words (see Table 2 for examples). Thus, if a sentence contained a high frequency word in its initial portion and a low frequency word in its final portion, there was an identical sentence in which the low frequency word appeared at the initial portion and the high frequency word appeared at the final portion. It is important to remember that although the critical words varied in their English frequency, they were presented in Spanish. Therefore, the frequency manipulation refers to the English equivalent of the presented critical Spanish words. The presentation of the two versions of the sentences was balanced across participants. Each participant was presented with only one version (LH or HL) of each sentence, but across participants, the two versions were presented an equal number of times. To make sure that the two versions of the sentences (LH version and HL version) were equated in difficulty and they would be equally well understood, we asked a group of 15 psychology students to rate the sentences from 1 (easy to understand) to 5 (difficult to understand). The ratings for the LH and HL versions were 2.59 (SD = 0.86) and 2.71 (SD = 0.89), respectively. There were no differences in rated comprehensibility among the sentences, t(91) = 1.62, p > .05. Also, to force participants to engage in comprehension processes, all of the trials included a verification task regarding the meaning of the sentences. Thus, 184 verification sentences were constructed to be included in the sentence verification task (see Table 2 for examples). Table 2 Example of the type of experimental materials used in Experiment 1 Version Sentence Low–High Desde la torre, construida entre esos dos recintos, se observaba perfectamente aquel puente. (From the tower, that was built between these two sites, the bridge can be observed perfectly.) Desde el puente construido entre esos dos recintos se observaba perfectamente aquella torre. (From the bridge, that was built between these two sites, the tower can be observed perfectly.) Desde el puente se observaba la torre. (si/no) (From the bridge, the tower can be observed.) (yes/no) High–Low Verification sentence Note: Sentences were presented in Spanish. The critical words in each sentence are in italics. Approximate English translations are given in brackets. 493 2.1.3. Procedure The sentences appeared word-by-word in the middle of a computer screen (Just, Carpenter, & Woolley, 1982). Participants read at their own pace by pressing the space bar to see new words. The time between consecutive key presses was taken as an index of the processing time for the displayed words. Each participant received two blocks of 46 sentences. Each block contained 23 LH sentences and 23 HL sentences. In one block participants were instructed to read for understanding each Spanish sentence and to repeat it after finishing reading, in the other they were instructed to read and then to translate (from Spanish to English). Instructions to repeat or to translate were displayed at the beginning of the block. The order of the blocks was balanced across participants. The order of the sentences within each block was randomised for each participant. Each trial started with a fixation point followed by the word-by-word sentence. Right after the last word of the sentence, the message ‘‘repeat’’ or ‘‘translate’’ appeared in the middle of the screen. Once participants had finished the repetition/translation task, the message ‘‘verification’’ appeared followed by a sentence on the centre of the screen. This sentence remained on the screen until the true/false response was made. Participants indicated whether the meaning of the sentence was congruent with the previously repeated or translated sentence by pressing the ‘‘true’’ or ‘‘false’’ key. 2.2. Results and discussion On-line comprehension was evaluated by considering the reading times of the critical words for the correctly verified and produced sentences. Global comprehension of the sentences was assessed by looking at the accuracy of the responses in the verification task. In addition, oral productions in the repetition and the translation tasks were recorded and their general quality was evaluated. Quality of task performance was measured by scoring each sentence for how well the participant performed the task. The scoring system for the translation output emphasized both how well the meaning of the input was preserved and how well the lexical and syntactic constructions matched those of the English language. For the repetition task, the scoring system emphasized the extent to which the output matched the lexical and syntactic forms of the input. Each production was rated from 1 to 5 where 1 indicated very poor production and 5 indicated very good production. Sentences with a rating of 3 or higher were considered correct and included in the analysis. Comprehension accuracy for this experiment was (79%).1 The 1 Given that the error rate for this experiment was high (close to 21%), we decided to perform ANOVA with frequency and task as within participants variable on the accuracy data. The results of these analyses indicated that none of the main effects or the interaction reached significance by participants or by items with all p > .10. 494 C. Ruiz et al. / Acta Psychologica 128 (2008) 490–500 mean reading times for correctly understood sentences within each condition was estimated for each participant or item and treated as a single observation. Reading times that were 2.5 standard deviations above the participant’s mean for the critical word were replaced with the participant’s mean for that word (1.7% of the data; see for Miyake et al., 1994, for a similar approach). Data from one participant were excluded from the analysis because he failed to correctly produce more than 60% of the sentences. Fig. 1 presents the mean reading times for the critical words in Experiment 1 as a function of type of task and type of word (High/Low frequency). The effect of type of task (reading for repetition/reading for translation) was significant by participants, F1(1, 14) = 6.12, MSE = 268,414, p < .03, and by items, F2(1, 91) = 47.54, MSE = 87,488, p < .0001. Thus, reading for translation was significantly slower (1560 ms) than reading for repetition (1229 ms). The effect of lexical frequency (Low/High) was significant by participants, F1(1, 14) = 5.36, MSE = 15,871, p < .04, and by items, F2(1, 91) = 23.91, MSE = 119,399, p < .0001. Response times for the low frequency words were longer (1432 ms) than for the high frequency words (1357 ms). Interestingly, the type of task · lexical frequency interaction was significant by participants, F1(1, 14) = 4.79, MSE = 16,837, p < .05, and by items, F2(1, 91) = 35.16, MSE = 73,217, p < .001. This interaction indicated that the effect of lexical frequency of the English translation of the presented Spanish words depended on the goal of reading (see Fig. 1). In this way, analyses of simple effects showed that lexical frequency was significant when the goal was to translate the sentences, F1(1, 14) = 8.37, MSE = 19,796, p < .01, F2(1, 91) = 83.75, MSE = 64,785, p < .0001, but it was not when the goal was to repeat them, F1(1, 14) < 1, F2(1, 91) < 1. This pattern of results suggests that reading for translation is different from reading in a monolingual context since they indicate that when participants are reading for translation, they start activating lexical candidates in the TL. In order to explore the possible time course of the TL lexical activation in translation, we examined frequency effects at the initial and final portion of the sentences in the translation condition (see Fig. 2). Note that this type of analysis is possible because for each sentence with a low frequency word at the beginning (LH sentences) there was an identical sentence with a high frequency word at this same position (HL sentences). Similarly, for each sentence with a low frequency word at the final portion of the sentence (HL sentences) there was an identical sentence with a high frequency word at this same position (LH sentences). The results of the ANOVA performed on data of the translation condition indicated that the interaction between frequency and position was significant by participants, F1(1, 14) = 5.99, MSE = 42,846, p < .03, and marginally significant by items, F1(1, 91) = 3.6, MSE = 415,855, p = .06. Planned comparison of the reading times for high and low frequency words at the beginning of the sentence showed that English frequency did not have an effect at this position, F1(1, 14) = 0.20, MSE = 12,100, p > .66, F2(1, 91) = 0.00, MSE = 149,111, p > .99. In contrast, comparison of the reading times for high and low frequency words placed at the final portion of the sentences 2400 2400 Low 2200 Low 2200 High High 2000 Reading Times (ms) Reading Times (ms). 2000 1800 1600 1400 1200 1800 1600 1400 1200 1000 1000 800 800 600 600 Repetition Translation Type of Reading Fig. 1. Reading times as a function of type of reading (reading for repetition and reading for translation) and target language lexical frequency (low and high) in Experiment 1. The bars indicate standard errors. Initial Final Sentence Position Fig. 2. Reading times in the reading for translation condition of Experiment 1 as a function of target language lexical frequency (low and high) and position of the critical word in the sentence (initial or final). The bars indicate standard errors. C. Ruiz et al. / Acta Psychologica 128 (2008) 490–500 indicated that the frequency effect was significant, F1(1, 14) = 8.33, MSE = 70,339, p < .01, F2(1, 91) = 4.28, MSE = 703,105, p < .04. Therefore, the overall pattern of results indicated that the lexical frequency of the TL affects reading times depending on the type of reading and the position of the critical word. Only in the translation task, high frequency words placed at the end of sentence were read faster than low frequency words. This frequency effect was not found for words placed at the beginning of sentence. This pattern of results replicates that of Macizo and Bajo (2006) with cognate words. Thus, Macizo and Bajo found that cognate effects were present only when participants had to translate the sentences and when the critical word was in the final portion of the sentence. The pattern of results found in this experiment clearly supports horizontal/parallel approaches to translation. First, comparisons of the overall reading times for the repeat and translate conditions of the experiment indicated that when participants were asked to translate, reading times were slower than when they were asked to understand and repeat the sentences. The slower reading times in the translation condition may indicate that when instructed to translate, participants engaged in additional processes needed for translation. In translation, beside the cognitive resources required for comprehension of the input, additional resources are needed for activating and switching between the two languages (Gile, 1997; Grosjean, 1997); the need for this additional resources slowed down the reading process. Second, the TL frequency effect in the translation condition provides direct evidence of lexical activation of the TL when participants read for translating. The frequency effect has been clearly located at the lexical level (e.g., Balota & Chumbley, 1984; Seidenberg, 1995) and, therefore, results of this experiment extend the findings of Macizo and Bajo (2006) to non-cognate words and clearly show the lexical nature of the effect. It is important to note that the frequency of the presented Spanish words (SL) was held constant and that, therefore, the observed frequency effect can only be due to the parallel activation of the target language. In addition, since the frequency effect was not found when participants were asked to read and repeat the sentences, it cannot be argued that the obtained effects were due to particular uncontrolled features of the materials. In both reading for repetition and reading for translation conditions, participants were exposed to the same reading materials and to the same reading situations, since repeating or translating was only performed after the reading and understanding processes had ended. These results rule out theories of translation that assume that comprehension and code-switching proceed in a serial manner. The presence of facilitation at the end of the sentences and its absence at the beginning of them cannot be interpreted as indicating that comprehension and codeswitching proceed in a serial manner since the effect appeared on-line, that is, when the critical word was being 495 read. Rather, this pattern might suggest that TL activation has a specific time course, but this interpretation will be discussed further later. 3. Experiment 2 As mentioned, the purpose of Experiment 2 was to explore whether there is syntactic activation of the TL during a reading for translation task. With this purpose in mind, our professional translators were asked again to read sentences and either repeat or translate them. We manipulated the congruency of the syntactic structure between the presented Spanish sentence and the equivalent translation in English. This manipulation was achieved by varying the order in which adjective–noun forms were presented and by including (or not) the pronoun that constituted the subject of a verbal form of a relative clause embedded in the sentence. Thus, in the congruent version of the sentences the adjective was presented before the noun and the pronoun/subject accompanied the verb (see Table 3 for an example). If understanding for later translation involves syntactic activation of the TL, we would expect to find syntactic congruency effects when participants are asked to translate the sentences, otherwise the effects would be absent in both reading conditions. 3.1. Method 3.1.1. Participants and design The type of reading (reading for repetition vs. reading for translation) and the type of syntactic structure (congruent versus incongruent Spanish/English structure) were manipulated within-participants. The same sixteen professional translators that participated in Experiment 1 were called again for participation in Experiment 2 (see Table 1 for details about participants’ characteristics). 3.1.2. Materials A set of 100 sentences was constructed. All the sentences were composed of 10–12 Spanish words arranged in three critical regions: initial (noun accompanied by an adjective), medium (verbal form embedded in a relative clause), and Table 3 Example of the type of experimental materials used in Experiment 2 Version Sentence Congruent La bonita casa que yo alquilé este verano tenı´a un verde jardı´ n. (The nice house that I rented this summer had a green garden.) La casa bonita que alquilé este verano tenı´a un jardı´ n verde. (The nice house that I rented this summer had a green garden.) Este verano la casa era verde. (This summer the house was green.) (yes/no) Incongruent Verification sentence Note: Sentences were presented in Spanish. The critical fragments in each sentence are in italics. Approximate English translations are given in brackets. C. Ruiz et al. / Acta Psychologica 128 (2008) 490–500 final (noun accompanied by an adjective). For each sentence, two versions were constructed. In the English congruent version the adjective (e.g., bonita, -nice-) was always presented before the noun form (e.g., casa, -house-) (e.g., bonita casa) and the personal pronoun (the subject) was presented next to the verbal form. In the English incongruent version, the adjective was always presented after the noun (e.g., casa bonita) and the personal pronoun (the subject) was omitted (see Table 3 for examples). Note that in Spanish the adjective can appear both before and after the noun and the syntactic subject can be absent or present, while in English the adjective should appear before the noun and the subject is always present. In order to make sure that both versions of the sentences were equally difficult to understand, thirty psychology students were asked to judge the comprehensibility of the sentences in a scale from 1 (easy to understand) to 5 (difficult to understand). The mean ratings for the congruent and incongruent sentences was 2.40 (SD = 0.51) and 2.36 (SD = 0.60) respectively. A t-test performed on the ratings indicated that both type of sentences were equated in comprehensibility, t(99) = .68, p > .49. As in Experiment 1, one hundred additional sentences were constructed to be used in the sentence verification task (see Table 3). 3.1.3. Procedure Each participant received two blocks of 50 sentences. Each block contained 25 congruent sentences and 25 incongruent sentences. In one block participants were instructed to read for understanding each Spanish sentence and to repeat it after finishing reading; in the other, they were instructed to read and understand and then to translate (from Spanish to English). Instructions to repeat or to translate were displayed at the beginning of the block. The order of the blocks was balanced across participants. The order of the sentences within each block was randomised for each participant. All other details of the procedure were identical to those in Experiment 1. 3.2. Results and discussion For the analyses performed on the reading times, the mean reading times for the critical words in the correctly comprehended and produced sentences were averaged and introduced in two ANOVAs (by participants and by items) with type of reading (reading for repetition vs. reading for translation) and type of sentence (congruent vs. incongruent) as independent variables. Note that the critical words in this experiment were the adjective and the noun at the beginning and end of the sentence and the verb of the relative clause (see italic words in Table 3). Thus the critical words in the congruent and incongruent condition were identical. Global accuracy for this experiment (accuracy on the verification task and production accuracy) was 90%. As in Experiment 1, reading times that were 2.5 standard deviations above the participant’s mean for the critical word were replaced with the participant’s mean for that word (1.8% of the data). Fig. 3 shows the mean reading times for sentences in Experiment 2 as a function of type of task and type of sentence. The results of the ANOVAs performed on the reading times indicated that the effect of type of reading was significant by participants and by items, F1(1, 15) = 9.47, MSE = 75,509, p < .007, and F2(1, 98) = 239.33, MSE = 22,500, p < .0001. Participants were slower when they were reading for translation (757 ms) than when they were reading for repetition (546 ms). The type of sentence was also significant by participants and by items, F1(1, 15) = 6.49, MSE = 6997, p < .02, and F2(1, 98) = 10.23, MSE = 25,089, p < .001. Congruent sentences were read faster than incongruent sentences (625 ms and 678 ms, respectively). The type of task · type of structure interaction was also significant by items, F1(1, 15) = 2.73, MSE = 8235, p > .11, and F2(1, 98) = 4.78, MSE = 27,829, p < .03, indicating that the effect of the type of sentence depended on the task to be performed after reading. Thus, when participants were instructed to read and translate the sentences, congruent sentences were read faster (734 ms) than incongruent sentences (822 ms), F1(1, 15) = 4.48, MSE = 14,680, p < .05, and F2(1, 98) = 8.34, MSE = 45,503, p < .004. However, when participants were instructed to read and repeat the sentences, the type of sentence structure did not show an effect, F1(1, 15) = 3.62, 1200 Incongruent 1100 Congruent 1000 Reading Times (ms) 496 900 800 700 600 500 400 Repetition Translation Type of Reading Fig. 3. Reading times as a function of type of reading (reading for repetition and reading for translation) and syntactic congruency between the source language and the target language structures (congruent and incongruent) in Experiment 2. The bars indicate standard errors. C. Ruiz et al. / Acta Psychologica 128 (2008) 490–500 MSE = 552, p > .07, and F2(1, 98) = 1.35, MSE = 7414, p > .24. As in Experiment 1, in order to explore the time course in which activation of the TL syntactic structures is initiated in translation, we performed analyses by regions in the sentences for this task. Additional ANOVAs on the reading times data in the translation task were made by averaging the reading times for each critical word in each of the critical regions of the sentence (initial, medium, and the final regions) for the congruent and incongruent conditions (see Fig. 4). Planned comparison of the reading times for the congruent and incongruent conditions at each region of the sentence indicated that when participants read the initial region of the sentence, the effect of the structure was close to significant by participants and items, F1(1, 15) = 3.97, MSE = 10,935, p < .06, F2(1, 98) = 2.99, MSE = 68,824, p < .08. When participants read the medium region of the sentence, the effect of the type of structure was significant by participants and items, F1(1, 15) = 7.69, MSE = 21,433, p < .01, F2(1, 98) = 11.17, MSE = 80,367, p < .001. Finally, when the participants read the final region of the sentences, the effect of the structure did not reach significance, F1(1, 15) = 0.003, MSE = 26,640, p > .85, F2(1, 98) = 0.29, MSE = 127,413, p > .58. In summary, reading times were affected by the congruency between the Spanish and English syntactic structures of the presented sentences when participants were asked to 1200 Incongruent 1100 Congruent Reading Times (ms) 1000 900 497 read the sentences and then translate them. The presentation of sentences, which were syntactically congruent with their English translations facilitates reading relative to sentences, which are incongruent, but this facilitation effect was only produced when participants were asked to translate the sentences and not when participants had to read in a monolingual context. In support of horizontal/parallel approaches to translation, these results suggest that within-language reading and reading for translation differ. Thus, the facilitation produced by presenting congruent syntactic structures seem to indicate that reading for translation involves activation and search for TL syntactic structures and these processes do not seem to be involved in within-language reading. It could be argued that this facilitation may have been due to the congruent sentences being more marked and distinctive than the incongruent ones (although the adjective in Spanish can go either before or after the noun, it appears more frequently after the noun). However there are two reasons against this argument: (1) comprehensibility ratings were equivalent for congruent and incongruent sentences and (2) the effect was not present in the reading for repetition condition as it should have been if the effect was due to the marked character of the sentence. Hence, our results suggest that in contrast to the predictions of serial models of translation, the reformulation processes involved in translations are initiated before the end of SL comprehension. Differently to Experiment 1, results of Experiment 2 indicate that when participants are reading for translation, the TL is accessed from the very beginning of the reading processes. Thus, congruency effects appeared at the initial and medium parts of the sentences. This may indicate that the time course of syntactic activation is different than the time course of lexical activation. Summarising, the results of Experiment 2 provide additional support to the horizontal/parallel approach by finding parallel processing of the languages involved. More importantly, they show that activation of the TL occurs not only at the lexical level but also at the syntactic level. 800 4. General discussion 700 The purpose of the present study was to determine whether lexical and syntactic activation of the TL take place during comprehension in translation. There are two theoretical contrasting approaches to this question. The vertical/serial view proposes that during understanding, the linguistic form of the input is lost and an abstract representation of the message is constructed from which the translator will give lexical expression to the message in the TL. According to this view, there are no direct codeto-code links between the two languages involved in the task. On the contrary, the horizontal/parallel view defends that comprehension in translation differs from comprehension in a monolingual context, since the translator activates and searches for possible linguistics matches in the TL while perceiving the SL. Hence, this view assumes that 600 500 400 Initial Medium Final Sentence Position Fig. 4. Reading times in the reading for translation condition of Experiment 2 as a function of syntactic congruency between the source language and the target language (congruent and incongruent) and position of the critical fragments in the sentence (initial, medium and final). The bars indicate standard errors. 498 C. Ruiz et al. / Acta Psychologica 128 (2008) 490–500 there are direct code-to-code links that are activated during SL comprehension. As we indicated, between these two more extreme positions, a third view proposes that both serial and parallel reformulation processes are possible in translation, but that code-to-code activation is more probable for less experienced translators and for conditions in which the message is continuously present. In the experiments reported here we contrasted these approaches by having experienced translators read sentences word-by-word with instructions to read and repeat them or to read and translate them. Activation of the TL during sentence comprehension was studied by manipulating the lexical frequency of the TL translation of some critical words (Experiment 1) and by varying the syntactic congruency between the Spanish presented sentence and its English translation (Experiment 2). The results of the two experiments consistently showed that the two comprehension conditions (reading for repetition and reading for translation) differ since the manipulation of TL lexical frequency and SL/TL syntactic congruency had only effects when the participants were instructed to read and translate. This pattern suggests that, in accordance with the horizontal/parallel view, comprehension in translation differ from comprehension in monolingual contexts, since SL understanding co-occurs with activation of the TL lexical and syntactic entries. Thus, both languages seem to be concurrently active during SL comprehension, and they may influence each other. This pattern of results and interpretation is consistent with the results obtained by Macizo and Bajo (2006, Experiment 2). In their experiment they showed that the cognate status of critical words in sentences influence reading times when participants were instructed to read and translate the sentences, but not when they had to read and repeat them. Our results extend their findings by showing a similar pattern of facilitation when non-cognate words varying in TL frequency were used. This finding is important because, as we mentioned, cognate words are sometimes assumed to have a special status in the bilingual mental lexicon (shared lexical store) and to share semantic as well as lexical features with the TL (De Groot, 1992; De Groot & Nas, 1991). However, the effect of lexical frequency has clearly been associated to the lexical level of linguistic processing. In addition, the results of Experiment 2 also extend the finding of Macizo and Bajo (2006) by showing that activation of the TL is restricted not only to the lexical level, but also to the syntactic one. Thus, reading times for the presented Spanish sentences were facilitated when there was congruency between the syntactic forms in the source and target languages, and this facilitation was only present when participants were asked to read and translate the sentences, indicating that comprehension for translation also involves looking for syntactic matches in the TL. When the source and target language share these structures, comprehension is facilitated. These results are congruent with those obtained by Hartsuiker et al. (2004) indicating that syntactic processing in one language influences syntactic processing in another. Their results showed that untrained bilingual participants tended to use the same syntactic structure that they previously heard in Spanish when describing pictures in English. Hence, results of Experiments 1 and 2 support parallel theories of translation. It is important to note that the conditions in our experiments are very different from those proposed by authors holding a mixed parallel/serial view to favour horizontal/parallel strategies. According to this view, horizontal, code-to-code, translation will be more probable in amateur translators (Seleskovitch, 1976) or in conditions where performance temporarily breaks down (Paradis, 1994). It is also more probable in text-to-text translation than in interpretation because the presented SL linguistic forms are continuously present in translation whereas those forms are transient in interpretation (Seleskovitch, 1976). In contrast, our experiment showed evidence of code-to-code lexical and syntactic searches with trained experienced translators and in a task very similar to consecutive translation. Since the words were presented one-by-one on the computer screen and participants were asked to translate at the end of the sentence (consecutive interpretation), the linguistic form of the SL was also transient and not present. Hence, our data suggest that parallel processing occurs also in conditions where some have thought it unlikely. Although lexical frequency and syntactic congruency were similar in being present in reading for translation and absent in reading for repetition, they did not behave exactly the same as their time courses seem to differ. Thus, TL lexical frequency seems to have its effects toward the end of the sentence (Experiment 1), whereas syntactic congruency has its effect in the initial part of the sentence (Experiment 2). The results from Experiment 1 are similar to those obtained by Macizo and Bajo (2006) with cognate words, which might indicate that although comprehension and code-switching processes co-occur, a minimum unit of information has to be processed before planning the translation output (Goldman-Eisler, 1972). Note, however, that the absence of lexical effects at the beginning of the sentence does not mean that comprehension and code-switching proceed in a serial manner since the effect appeared online, that is, when the critical word was being read. More importantly, results from Experiment 2 clearly shows that code-switching starts from the very beginning of the sentence, but that activation of the TL is restricted to the syntactic level. Although speculative at the moment, this might suggest that lexical and syntactic activation of the TL follow different time courses with TL syntactic searches starting before than searches for TL lexical entries. This will be consistent with some current theories of language production (e.g., Dell & O’Seaghdha, 1992) that propose different level of processing acting with different temporal parameters. When producing a unit of information, semantic, syntactic and lexical entries are activated. Although the activation of these entries overlaps in time, a temporal sequence of C. Ruiz et al. / Acta Psychologica 128 (2008) 490–500 semantic, syntactic and lexical peaks of activation can be locally observed. However, this conclusion is at the moment speculative since it can also be argued that different regions of the sentence differ in the demand that they impose on WM (e.g.,the final part of the sentence is specially demanding since integration processes occur at this time), so that the obtained differences may be due to the influence of WM load on the activation of the TL lexical and syntactic properties. Further research should address this issue. The absence of TL lexical frequency (Experiment1) and syntactic congruency (Experiment 2) when participants were asked to read and repeat the sentences and translation was not involved indicate that these effects may depend on the context, the goal, and/or the resources available for the reading task (Macizo & Bajo, 2006). This result contrasts with experiments in bilingual word recognition showing that lexical properties of the non-presented language (e.g., cognate status) affect recognition times. This result has been interpreted as indicating that lexical access in bilinguals is not language selective (e.g., Dijkstra & Van Heuven, 1998; see Dijkstra, 2005, for a review). In contrast, results of the read and repeat condition of Experiments 1 and 2 indicate that properties of the unintended language do not always have an effect on reading times. Recently, Kroll, Bobb, and Wodniecka (2006) have suggested that language selection in production may depend on a set of factors such as the demands of the task or the experience of the bilinguals. Similarly, lexical and syntactic access in comprehension by bilingual participants may also depend on similar factors. For example, some experiments have shown that the semantic context may restrict language activation (Schwartz, 2003; Schwartz & Kroll, 2006; Van Hell, 1998). For example, in Schwartz’s study, bilingual participants were presented sentences with the instruction to name the word printed in red. These red words consisted of cognates, interlingual homographs and controls and they were embedded in high constraint or low constraint sentences (the critical word was more or less contextually probable within the sentence). The results indicated that cognate effects disappeared when semantically constrained sentences were presented and when participants had high comprehension skills. This may indicate that when the context provides sufficient semantic information and the reader has enough processing resources, the meaning of the sentences can be integrated and the presence of cognates is no longer important. Hence, the richness of the semantic context and the readers’ comprehension skills may be two factors constraining selectivity. Similarly, it is possible that the rich semantic context and the high WM demands of our reading for repetition condition (the participants had to maintain the sentence in memory to be able to reproduce it) induced language selectivity. It is possible that if our participants were not asked to repeat the sentences but only to read them for comprehension, frequency effect would have also appeared in the only reading conditions. Further research on bilingual comprehension should investigate this issue. 499 In conclusion, Experiments 1 and 2 favoured horizontal/ parallel theories of translation. Reading in monolingual contexts seems to differ from reading for later translation. Reading for translation involves the parallel activation of TL lexical and syntactic entries. Thus, lexical and syntactic TL properties had an effect when participants were asked to read and then translate, but they did not when they were not asked to translate. This parallel activation seems, however, to take place at different points in the sentence depending on whether lexical or syntactic properties are manipulated. This may indicate that activation of lexical and syntactic properties may follow different time courses, but very importantly they also suggest that TL activation can occur at early stages of sentence processing. Acknowledgements This research was supported by grants (BS02002-00159 and SEJ2005-00842) to M.T. Bajo, and by the research program ‘‘Ramón y Cajal’’ and grant SEJ2006-05728/ PSIC, to P. Macizo. References Alameda, J. R., & Cuetos, F. (Eds.). (1995). Diccionario de frecuencias de las unidades lingüı´sticas del castellano. [Frequency dictionary of Spanish linguistic units]. Oviedo: Servicio de Publicaciones de la Universidad de Oviedo. Balota, D. A., & Chumbley, J. I. (1984). Are lexical decisions a good measure of lexical access? The role of word frequency in the neglected decision stage. Journal of Experimental Psychology: Human Perception and Performance, 10, 340–357. Bock, J. K. (1986). Syntactic persistence in language production. Cognitive Psychology, 18, 355–387. Branigan, H. P., Pickering, M. J., & Cleland, A. A. (2000). Syntactic coordination in dialogue. Cognition, 75, B13–B25. Christoffels, I. K., & de Groot, A. M. B. (2005). Simultaneous interpreting: A cognitive perspective. In J. F. Kroll & A. M. B. de Groot (Eds.), Handbook of bilingualism: Psycholinguistic approaches (pp. 454–479). New York: Oxford University Press. Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behavior, 19, 450–466. De Bot, K. (2000). Simultaneous interpreting as language production. In B. E. Dimitrova & K. Hyltenstam (Eds.), Language processing and simultaneous interpreting. Interdisciplinary perspectives (pp. 65–88). Amsterdam: John Benjamins. De Groot, A. M. B. (1992). Determinants of word translation. Journal of Experimental Psychology: Learning, Memory & Cognition, 8, 1001–1018. De Groot, A. M. B., & Nas, G. L. J. (1991). Lexical representation of cognates and non-cognates in compound bilinguals. Journal of Memory and Language, 30, 90–123. Dell, G. S., & O’Seaghdha, P. A. (1992). Stages of lexical access in language production. Cognition, 42, 287–314. Dijkstra, A. (2005). Bilingual visual word recognition and lexical access. In J. F. Kroll & A. M. B. de Groot (Eds.), Handbook of bilingualism: Psycholinguistic approaches (pp. 179–201). New York: Oxford University Press. Dijkstra, A., Timmermans, M., & Schriefers, H. (2000). Cross-language effects on bilingual homograph recognition. Journal of Memory and Language, 42, 445–464. 500 C. Ruiz et al. / Acta Psychologica 128 (2008) 490–500 Dijkstra, A., & Van Heuven, W. J. B. (1998). The BIA model and bilingual word recognition. In J. Grainger & A. Jacobs (Eds.), Localist connectionists approaches to human cognition (pp. 189–225). Mahwah, NJ: Lawrence Erlbaum Associates. Forster, K. I., & Chambers, S. M. (1973). Lexical access and naming time. Journal of Verbal Learning and Verbal Behavior, 12, 627–635. Frederiksen, J. R., & Kroll, J. F. (1976). Spelling and sound: Approaches to the internal lexicon. Journal of Experimental Psychology: Human Perception and Performance, 2, 361–379. Garcı´a-Albea, J. E., Sánchez-Casas, R., & Valero, T. (1996, September). Form and meaning contribution to word recognition in CatalanSpanish bilinguals. Paper presented at the meeting of the Ninth Conference of the European Society for Cognitive Psychology, University of Würzburg, Germany. Gerver, D. (1976). Empirical studies of simultaneous interpretation: A review and a model. In R. W. Brislin (Ed.), Translation: Applications and research (pp. 165–207). New York: Gardiner. Gile, D. (1991). Methodological aspects of interpretation (and translation) research. Target, 3, 153–174. Gile, D. (1997). Conference interpreting as a cognitive management problem. In H. J. Danks, G. M. Shreve, S. B. Fountain, & M. K. McBeath (Eds.), Cognitive processes in translation and interpreting (pp. 196–214). Thousand Oak, CA: Sage. Goldman-Eisler, F. (1972). Segmentation of input in simultaneous translation. Journal of Psycholinguistic Research, 1, 127–140. Grosjean, F. (1997). The bilingual individual. Interpreting, 2, 163–187. Hartsuiker, R. J., Pickering, M. J., & Veltkamp, E. (2004). Is syntax separate or shared between languages? Cross-linguistic syntactic priming in Spanish/English bilinguals. Psychological Science, 15, 409–414. Isham, W. P. (1994). Memory for sentence form after simultaneous interpretation: Evidence both for and against deverbalization. In S. Lambert & B. Moser-Mercer (Eds.), Bridging the gap: Empirical research in simultaneous interpretation (pp. 191–211). Amsterdam: John Benjamins. Just, M. A., Carpenter, P. A., & Woolley, J. D. (1982). Paradigms and processes in reading comprehension. Journal of Experimental Psychology: General, 111, 228–238. Kroll, J. F., Bobb, S. C., & Wodniecka, Z. (2006). Language selectivity is the exception, not the rule: Arguments against a fixed locus of language selection in bilingual speech. Bilingualism: Language and Cognition, 9, 119–135. Kucera, H., & Francis, E. (1967). Computational analysis of present day American English. Providence, RI: Brown University Press. Lederer, M. (Ed.). (1994). La traduction aujourd’hui. Le modèle interprétatif. Paris: Hachette. Lederer, M. (Ed.). (2003). Translation: The interpretive model. Manchester: St. Jerome Publishing. Macizo, P., & Bajo, M. T. (in press). Comprehension processes in translation. In: Alamargot, D., Terrier, P. & Collier, J. M. (Eds.), Improving the production and understanding of written documents in the work place. Amsterdam: Elsevier. Macizo, P., & Bajo, M. T. (2006). Reading for understanding and reading for translation: Do they involve the same processes? Cognition, 99, 1–34. Miyake, A., Just, M. A., & Carpenter, P. A. (1994). Working memory constraints on the resolution of lexical ambiguity: Maintaining multiple interpretations in neutral contexts. Journal of Memory and Language, 33, 175–202. Monsell, S. (1985). Repetition and the lexicon. In A. W. Ellis (Ed.). Progress in the psychology of language (Vol. 2, pp. 147–195). London: Erlbaum. Paradis, M. (1994). Toward a neurolinguistic theory of simultaneous translation: The framework. International Journal of Psycholinguistics, 10, 319–335. Paradis, M., Goldblum, M. C., & Abidi, R. (1982). Alternate antagonism with paradoxical translation behavior in two bilingual aphasic patients. Brain and Language, 15, 55–69. Schwartz, A. (2003). Word and sentence-based processes in second language reading. Unpublished doctoral dissertation, Pennsylvania State University, University Park, PA. Schwartz, A. I., & Kroll, J. F. (2006). Bilingual lexical activation in sentence context. Journal of Memory and Language, 55, 197–212. Seidenberg, M. S. (1995). Visual word recognition: An overview. In P. Eimas & J. L. Miller (Eds.). Handbook of perception and cognition: Speech, language and communication (pp. 137–179). New York: Academic Press. Seleskovitch, D. (1976). Interpretation: A psychological approach to translating. In R. W. Brislin (Ed.), Translation: Applications and research (pp. 92–116). New York: Gardner. Seleskovitch, D. (1999). The teaching of conference interpretation in the course of the last 50 years. Interpreting, 4, 55–66. Van Hell (1998). Cross-language processing and bilingual memory organization. Unpublished dissertation, University of Amsterdam, Amsterdam, The Netherlands. Van Hell, J. G., & Dijkstra, T. (2002). Foreign language knowledge can influence native language performance in exclusively native contexts. Psychonomic Bulletin and Review, 9, 271–278.