Words in receptive and productive vocabulary are comprehended and produced more efficiently with language experience as the words are encountered and used in natural language contexts. However, the impact of a single additional contextualized exposure to a word on later comprehension and production is unknown. Repetition-priming studies do measure the impact of a single exposure to a word on comprehension and production processes, but these studies are typically done with isolated words as stimuli. Research on vocabulary learning and research on repetition priming have progressed separately as subareas within language and memory subfields, respectively, with very little intersection. The learning mechanisms involved in these phenomena may be one and the same, and the present study begins to link them by examining repetition priming for words that are comprehended and produced in the context of sentences, as they would be with natural language.

Processes and repetition priming in word translation

Evidence that translation equivalents for concrete nouns access the same conceptual representation is well established in the cognitive bilingual literature (for reviews, see Francis, 1999, 2005). Translation tasks that involve access to this common concept are said to be concept mediated. There is consensus in the literature that spoken word translation in both directions is concept mediated in fluent early bilinguals (e.g., De Groot, Dannenburg, & Van Hell, 1994; De Groot & Poot, 1997; Duyck & Brysbaert, 2004; Francis, Augustini, & Sáenz, 2003; La Heij, Hooglander, Kerling, & van der Velden, 1996; Miller & Kroll, 2002; Potter, So, von Eckhardt, & Feldman, 1984). Late bilinguals who are highly dominant in their first language may use direct word-to-word connections, or word mediation, for translation from the nondominant language (L2) to the dominant language (L1) (Kroll & Stewart, 1994; Sholl, Sankaranarayanan, & Kroll, 1995), but such a population was not included in the present study. When concept mediated, spoken word translation can be decomposed into two sets of processes. The first, word comprehension, includes the processes that occur between presentation of the stimulus and access to the concept. The second, word production, includes the processes that occur after the concept is accessed and lead to overt articulation of the word.

Word translation exhibits repetition priming, such that repeated words are translated more quickly than new words (Francis et al., 2011; Francis & Gallard, 2005; Francis & Sáenz, 2007; Francis, Tokowicz, & Kroll, 2014). Repetition priming in word translation is based on both word comprehension and word production processes (Francis et al., 2011; Francis & Gallard, 2005), and both components exhibited priming across a 1-week interval (Francis & Sáenz, 2007). The durability of these effects shows that long-term learning can be observed for both word comprehension and word production using repetition-priming methodology.

Three sources of evidence suggest that the processes of comprehension and production in word translation are independent. First, in a trilingual study that involved all six possible directions of translation, production time was independent of the comprehension language, and comprehension time was independent of the production language (Francis & Gallard, 2005). Second, using additive factors methodology, Francis et al. (2011) showed that the effects of repetition priming in these processes are independent, which suggests that the processes themselves are independent. Third, across a 1-week retention interval, priming in word comprehension remained stable, but priming in word production decreased substantially (Francis & Sáenz, 2007).

Processes and repetition priming in picture naming

Picture naming requires the execution of several distinct processes, including, minimally, identification of the pictured object, the retrieval of a word that names the object, and the overt articulation of the corresponding phonology. Existing models are in agreement about both the necessity of these three types of processes and the order of their initiation (Johnson, Paivio, & Clark, 1996). Picture naming is usually thought to include access to amodal concepts as the endpoint of the object identification process (e.g., Levelt et al., 1991; Potter et al., 1984; Theios & Amrhein, 1989; but see Johnson et al., 1996, for an argument against this position). Previous research has provided strong evidence that picture naming is concept mediated in monolinguals (Durso & Johnson, 1979; Potter & Faulconer, 1975; Smith & Magee, 1980) and in fluent bilinguals (Chen & Leung, 1989; Francis et al., 2003; Potter et al., 1984). The processes required for picture naming can be decomposed into two sets of processes. The first, object identification, includes perceptual processes and retrieval of the concept. The second, word production, includes the processes that occur after the concept is accessed and lead to an overt verbal response.

Repetition priming in picture naming is based on both object identification and word production processes (for a review, see Francis, 2014). The word production processes of picture naming are the same as the word production processes in word translation, as indicated by the strong and symmetric priming observed between these two tasks (Francis et al., 2003). Repetition priming in picture naming is durable over delays of several weeks (e.g., Brown, Jones, & Mitchell, 1996; Cave, 1997; Mitchell & Brown, 1988), which indicates that it represents long-term learning. Long-term learning occurs for both the object identification and word production components of picture naming (Francis & Sáenz, 2007).

Two sources of evidence suggest that the object identification and word production components of picture naming are independent. First, using additive factors methodology, Francis, Corral, Jones, and Sáenz (2008) showed that the effects of repetition priming on these processes were independent, which suggests that the processes themselves are independent. Second, priming in object identification remained stable across a 1-week interval, but priming in word production declined substantially (Francis & Sáenz, 2007).

Effects of word frequency and bilingual language dominance

Responses are slower for low-frequency words than for high-frequency words in both word translation (De Groot, 1992) and picture naming (e.g., Gollan, Montoya, Cera, & Sandoval, 2008; Wheeldon & Monsell, 1992). Responses are also slower when responses are made in L2 than when responses are made in L1 in word translation (e.g., Kroll & Stewart, 1994; Potter et al., 1984) and picture naming (e.g., Potter et al., 1984; Sholl et al., 1995). A common mechanism for all of these effects has been proposed: the weaker links hypothesis (Gollan et al., 2008). The idea is that because L2 words have weaker links to their concepts than do L1 words, they operate functionally as lower frequency words in a bilingual person’s vocabulary.

Several repetition-priming results are also consistent with the weaker links hypothesis. Repetition-priming effects in picture naming are larger for low-frequency words than for high-frequency words (Wheeldon & Monsell, 1992). Similarly, repetition-priming effects in picture naming are greater for L2 words than for L1 words (Francis et al., 2003; Francis et al., 2008; Francis & Sáenz, 2007). Word frequency effects on repetition priming have not been tested in word translation. However, repetition priming in word translation is greater for L1–L2 translation than for L2–L1 translation (Francis et al., 2011; Francis & Gallard, 2005; Francis et al., 2014). The repetition-priming advantages for low-frequency words and for L2 words are also consistent with a proceduralist view put forth by Kolers and Roediger (1984), in which less skilled encoding results in a higher degree of transfer to the test task.

Repetition priming for words comprehended or produced in context

Repetition priming in comprehension and production processes may be the fundamental learning process that leads to increased proficiency with experience. The impact of an individual exposure to a word has a lasting impact. For this explanation of real vocabulary acquisition to be plausible, it would have to be the case that the phenomenon is not restricted to the isolated word conditions of a typical laboratory experiment on repetition priming.

Several previous studies have examined priming from words in text passages to words presented in isolation at test. Lexical decision priming for words encountered in the context of text passages at encoding was reliable but weaker than priming for words encountered in isolation at encoding (Speelman, Simpson, & Kirsner, 2002). Lexical decision priming was stronger for isolated words than when they were presented as part of the experimental instructions or a questionnaire before the test phase (Oliphant, 1983). Perceptual identification priming was eliminated when words were presented as part of a text, rather than isolated (Levy & Kirsner, 1989). Word fragment completion priming was substantial for word previously read in meaningful text, but it was weaker than for isolated words (MacLeod, 1989). Similarly, word fragment completion priming for words read in sentences or generated in response to a sentence was substantial but weaker than priming for words read in isolation (Smith, 1991). This was true for both the dominant and the nondominant languages in bilinguals. Thus, the effects in several forms of repetition priming are reduced or eliminated when context is provided at study.

Initial explanations of this effect were that integration of the word into the larger conceptual framework of the sentence causes conceptual processing that is not transfer appropriate for later processing in isolation (Levy & Kirsner, 1989; MacLeod, 1989; Oliphant, 1983). However, Masson and MacLeod (2000) showed that the context does not have to be meaningful to reduce priming. Even unrelated word sequences caused enough contextual binding to reduce priming. Thus, they proposed that the encoding task must allow a word to be individuated and distinctively encoded for transfer to occur. Priming from words in context at encoding to isolated words at test was more robust when low-frequency words were used (Nicolas, 1996), when the reading process was made more difficult (Nicolas, 1998), or when the task was done with participants who had low reading proficiency (Bourassa, Levy, Dowin, & Casey, 1998). On the basis of the hypothesis that bilingual vocabulary is of lower functional frequency in L2 than in L1 (Gollan et al., 2008) and because reading is more difficult in L2 than in L1, these findings suggest that transfer from contextualized words will be stronger in L2 than in L1.

When words were presented in context with exactly the same task and context at encoding and test, priming was strong. Text passages were read faster on the second read (Levy & Kirsner, 1989). Repeated readings also resulted in substantial, but attenuated, between-language transfer (Kolers, 1975). In these cases, it is not clear to what extent the individual words benefited from their own repetition and to what extent the response time (RT) reduction was due to the predictability of the word sequence afforded by the prior reading. In an ERP study with each critical word presented as the final word of a sentence, repetition effects in neural response were found when the context was the same, but not when it changed, even if the meaning of the word was preserved (Besson & Kutas, 1993). This result suggests that repetition effects in comprehension are tied to context and that the individual words do not benefit simply by virtue of being repeated.

The present study

The primary goal of the present study was to find out whether words translated in the context of a sentence at encoding would exhibit facilitation in comprehension and production at test and how this facilitation would compare with that found for words translated in isolation at encoding. According to transfer-appropriate processing logic (Morris, Bransford, & Franks, 1977; Roediger & Blaxton, 1987), there should be substantial transfer from contextualized to isolated words, on the basis of processes common to contextualized and isolated word comprehension and production. However, transfer from contextualized to isolated words should be attenuated, relative to identical repetition, because of differences in the operations or processes engaged at encoding and test. Specifically, words comprehended and/or produced in the context of sentences may not achieve the same level of individuation as words comprehended and/or produced in isolation (Masson & MacLeod, 2000).

The weaker links hypothesis (Gollan et al., 2008) suggests that L2 words should operate similarly to low-frequency words, and low-frequency words are less adversely affected by the presence of context at encoding than are high-frequency words (Nicolas, 1996). Therefore, a second goal of the study was to find out whether repetition priming for contextualized words, as a proportion of priming for isolated words, would be greater for L2 words than for L1 words.

Words were presented at encoding either in isolation or in a sentence context and translated, which requires both comprehension in the stimulus language and production in the response language. At test, half of the words were translated in isolation, which again requires both comprehension and production. The remaining words were generated in a picture-naming task, which requires word production. Repetition priming was measured for words previously translated in isolation and for words previously translated in a sentence context. In Experiment 1, the words to be translated at encoding and test were presented visually, and in Experiment 2, the words to be translated at encoding and test were presented auditorially.

Experiment 1

The purpose of Experiment 1 was to find out whether words translated in the context of a written sentence would elicit repetition priming in comprehension and production at test. Spanish–English bilinguals translated written words at encoding either in isolation or in the context of sentences. At test, they translated words in isolation and named pictures.

Method

Participants

The research participants were 96 self-identified Spanish–English bilinguals (33 men, 63 women) ranging in age from 17 to 41 (median = 19). They were undergraduate students at the University of Texas at El Paso who participated for introductory psychology research credit or as unpaid volunteers. According to self-report, 50 % were English dominant and 50 % were Spanish dominant. The first language learned was Spanish for 88 % and English for 2 %, and 9 % learned both languages simultaneously from infancy. The median age of second language acquisition was 5 years. Usage over the preceding month was reported as 46.5 % English, 43.2 % Spanish, 9.7 % a mixture of English and Spanish, and 0.6 % other languages; this corresponded to using L1 51.0 % of the time and L2 38.7 % of the time. Nearly all (99 %) reported Hispanic ethnicity (primarily Mexican-American or Mexican national). Five additional participants were excluded because of excessive error rates or failure to follow instructions.

It should be noted that the University of Texas at El Paso is located less than 1 mile from the U.S.–Mexico border and that the city of El Paso, as well as the university, are very much bilingual double-immersion environments. Nevertheless, individual participants were able to identify one language as stronger than the other, and the error rate and RT data, as we will see, bear this out. The Spanish-dominant participants were primarily students who immigrated to the U.S. after age 10 and Mexican nationals who commute across the border daily to attend class.

Design

The design was a 3 (encoding condition) × 2 (test task) × 2 (response language) within-subjects design. The encoding conditions were word translation, sentence translation, and not presented. The test tasks were translation and picture naming. The response language was always consistent from encoding to test, and half of the items were assigned to each response language. The primary dependent variable was RT for translation or picture naming at test.

Materials

The experimental stimuli were 288 pictures and their English and Spanish names. These were words and pictures used in our previous bilingual research (Francis & Sáenz, 2007). There were no identical cognates. The median frequency of the words was 13 per million in English (Baayen, Piepenbrok, & Gulikers, 1995) and 12 in Spanish (Alameda & Cuetos, 1995). The 288 items were randomly assigned to 12 sets of 24, and these sets were rotated through the experimental conditions across participants, using a Latin square to control for specific-item effects.

For the sentence conditions, simple sentences were constructed that each contained two of the critical words. Although it is an interesting question whether particular combinations of nouns and verbs or particular types of sentences would be more salient than others, we were not interested in examining specific types of relationships, just embedding the words in some syntactic and semantic context and comparing that with isolated words. Note that every word appeared in a sentence for one third of participants, in isolation for one third of participants, and not at all for one third of participants.

Procedure

Participants were tested individually by a bilingual experimenter in sessions lasting approximately 1 h. After filling out a language background questionnaire and getting some practice in use of the microphone, the experimenter gave instructions for each task in the assigned response language. The encoding phase consisted of four blocks of trials: translating words from Spanish to English, translating sentences from Spanish to English, translating words from English to Spanish, and translating sentences from English to Spanish. The order of tasks and languages was counterbalanced across participants. Each word translation block consisted of 3 practice trials and 48 experimental trials. The word stimulus appeared in the center of the screen and remained until the participant responded aloud; once the voice relay registered a response, the next trial was initiated after a 1,200-ms intertrial interval. Each sentence translation block consisted of 3 practice trials and 24 experimental trials (each containing two critical words). The sentence stimulus appeared in the middle of the screen and remained until the participants had completed their response. At that point, a button was pressed to advance to the next sentence.

The test phase consisted of four blocks of trials: translating words from Spanish to English, naming pictures in English, translating words from English to Spanish, and naming pictures in Spanish. The order of tasks was counterbalanced across participants, and the response language order matched that of the encoding phase. Each block consisted of 3 practice trials and 72 experimental trials (24 items previously translated in isolation, 24 items previously translated in sentence context, and 24 items not previously presented). The translation trials were done in exactly the same manner as the translation trials for words in the encoding phase. On picture-naming trials, the picture would appear in the center of the screen and remain until a vocal naming response was registered; after a 1,200-ms delay, the next trial was initiated. Throughout the computerized tasks, the experimenter noted any unexpected responses or voice relay misfires on a worksheet preprinted with the sequence of expected responses.

Results

Data processing

Analysis focused on valid RTs in the test phase. Four problematic items were removed for all participants. On average, 11.7 % of trials were removed as naming or translation response errors (including “don’t know” responses), 0.6 % were removed as machine timing errors, and 6.5 % were removed as spoiled trials. Spoiled trials included trials for which the corresponding encoding phase trial had an inconsistent but acceptable response (1.1 %), a response error (4.1 %), or a machine timing error (0.4 %) and trials for which the correct response was given in error to a different item on a previous trial (0.9 %). Items with RTs greater than 5,000 ms, less than 200 ms, or more than 2 standard deviations from the condition mean were removed as outliers, which resulted in the exclusion of 5.5 % of the trials. Thus, on average, 74.4 % of the test phase trials were retained for the RT analysis, which left a mean of about 18 items per condition per participant.

Encoding phase performance

Encoding phase error rates and RTs are given in Table 1. Responses were recoded to the dominant and nondominant languages according to the self-reported dominant language of each participant. Error rates were submitted to a 2 (context) × 2 (translation direction) repeated measures ANOVA. Error rates were higher for L1–L2 than for L2–L1 translation, F(1, 95) = 39.17, MSE = 3.06, p < .001. Error rates did not differ reliably for isolated and contextualized words, F(1, 95) = 2.21, MSE = 1.84, p = .140. The effects of translation direction and context interacted, with a bigger effect of direction in the sentence context, F(1, 95) = 4.19, MSE = 2.20, p = .044. RTs were measured only for the isolated words, and the effect of translation direction did not approach significance, t < 1.

Table 1 Mean encoding phase response times (RTs) and error rates in Experiments 1 and 2
Full size table

Test phase performance

Test phase error rates and RTs are given in Table 2. New-item RTs were submitted to a 2 (task) × 2 (response language) repeated measures ANOVA. Overall, translation RTs were longer than picture-naming RTs, F(1, 95) = 19.08, MSE = 82,747, p < .001, and responses were slower when given in L2, F(1, 95) = 21.40, MSE = 90,999, p < .001. These effects were qualified by a significant interaction, F(1, 95) = 8.75, MSE = 73,634, p = .004, such that picture-naming RTs were faster in L1 than in L2, t(95) = 5.22, p < .001, but translation RTs did not differ significantly across translation directions, t(95) = 1.52, p = .132.

Table 2 Mean response times (RTs) and repetition priming in Experiment 1
Full size table

Priming scores were obtained by subtracting repeated condition RTs from the corresponding new-item RTs and are shown in Fig. 1. Repetition priming was statistically significant in every cell (ps < .001). Repetition-priming scores were submitted to a 2 (task) × 2 (context) × 2 (response language) repeated measures ANOVA. A main effect of task showed that overall, priming was greater when translation was repeated from encoding to test than when the task changed to picture naming, F(1, 95) = 14.33, MSE = 176,673, p < .001. A main effect of context showed that priming was greater when the word was isolated at both encoding and test than when it went from contextualized to isolated, F(1, 95) = 67.32, MSE = 22,138, p < .001. An interaction of task and context showed that the effect of context was stronger for translation than for picture naming, F(1, 95) = 11.46, MSE = 34,268, p = .001. A main effect of response language showed greater priming in L2 than in L1, F(1, 95) = 19.84, MSE = 100,564, p < .001. Response language interactions with the other variables were not statistically reliable, ps > .05.

Fig. 1
figure 1

Repetition priming in Experiment 1 as a function of language, test task or process shared, and encoding context

Full size image

Production and comprehension priming estimates were derived for a second analysis. With identical repetition, priming of picture naming is based on both object identification and word production processes, but because word translation and picture naming share only word production processes, priming of picture naming from prior translation was assumed to be based solely on repetition of word production processes. (Note that priming in picture naming does not occur when only the concept is repeated from encoding to test; Monsell, Matthews, & Miller, 1992.) Priming of translation was assumed to be based on repetition of both word comprehension and word production processes. Therefore, the difference in priming between translation and picture naming represents priming in word comprehension processes. Specifically, L1–L2 translation and L2 picture naming have L2 word production processes in common. However, L1–L2 translation also involves L1 comprehension. To capture the priming in L1 comprehension, we subtract the priming obtained for L2 picture naming from that obtained for L1–L2 translation. As indicated by priming in picture naming, priming in production was significant in each context × language combination, ps < .001. In comprehension, no priming was observed for words comprehended in an L1 sentence context, t < 1, but significant priming was observed in the other three conditions (ps < .005).

Production and comprehension priming estimates were submitted to a 2 (process) × 2 (context) × 2 (language) repeated measures ANOVA. A main effect of process showed more priming for production than for comprehension, F(1, 95) = 7.983, MSE = 379,561, p = .006. A main effect of context showed greater priming for words originally translated in isolation, F(1, 95) = 44.47, MSE = 19,186, p < .001. A main effect of language showed more priming for L2 processes than for L1 processes, F(1, 95) = 8.02, MSE = 333,937, p = .006. The effects of context and language were qualified by a significant interaction, F(1, 95) = 4.62, MSE = 69,789, p = .034. Specifically, context effects were smaller in L2 than in L1. In fact, in L2 production, the effect of context did not approach significance, t < 1. No other interactions approached significance, ps > .10.

Discussion

As in our previous research, individual word translation at study primed both word translation and picture naming at test. The priming effect in picture naming indicates that word production in the response language was facilitated. The priming effect in word translation was larger than the priming effect in picture naming, indicating that comprehension processes in the stimulus language were also facilitated. Priming in both processes was stronger for L2 than for L1, consistent with the weaker links hypothesis.

The new findings were in the priming of word translation and picture naming following sentence translation at encoding. Words translated in a sentence context exhibited priming when translated in isolation at test. Words translated in a sentence context also exhibited priming when their pictures were named at test, indicating substantial priming in the word production component of translation. This component was stronger in L2 than in L1. The word comprehension component, derived by subtraction, also exhibited facilitation that was stronger in L2 than in L1. Both of these effects were consistent with predictions of the weaker links hypothesis. In fact, word comprehension priming was not reliable in L1. This last condition is the one that most resembles the tasks that did not exhibit priming from words in context in previous monolingual repetition priming studies.

Another question was whether embedding the words in a sentence context would reduce priming, relative to words presented in isolation. Overall, priming was reduced for words translated in context at encoding, relative to the words translated in isolation at encoding. However, the degree of reduction was different for comprehension and production components. The reduction was not as great for production as for comprehension. Also, the degree of reduction was different for L1 and L2; the reduction was greater for L1, consistent with predictions of the weaker links hypothesis.

Experiment 2

The purpose of Experiment 2 was to find out whether words translated in the context of a spoken sentence at study would produce repetition priming in comprehension and production at test. Experiment 2 was a replication of Experiment 1, with the words and sentences presented auditorially for translation at encoding and test.

Method

Participants

The research participants were 96 self-identified Spanish–English bilinguals (38 men, 58 women) with a median age of 19 years and recruited from the same sources as in Experiment 1. According to self-report, 56 % were English dominant and 44 % were Spanish dominant. The first language learned was Spanish for 90 % and English for 5 %, and 5 % learned both languages simultaneously from infancy. The median age of second language acquisition was 6 years. Usage over the preceding month was reported as 49 % English, 41 % Spanish, and 10 % a mixture of English and Spanish; this pattern corresponded to using L1 57 % of the time and L2 33 % of the time. All reported Hispanic ethnicity (primarily Mexican-American or Mexican national). Thirteen additional participants were excluded because of excessive error rates.

Design, materials, apparatus, and procedure

The design was the same as that for Experiment 1, except that all verbal stimuli for translation at encoding and test were presented auditorially. The materials were the same, except that several of the sentences were edited for length to make them easier to remember for translation. Auditory recordings were made by a female native speaker of English and Spanish and edited using SoundEdit software. The apparatus was the same except for the addition of a set of headphones for listening to the auditory stimuli. The procedure was the same as that in Experiment 1.

Results

Data processing

Analysis focused on valid RTs in the test phase. Out of a total of 288 test phase trials on average, 14.3 % were removed as naming or translation response errors (including “don’t know” responses), 3.6 % were removed as machine timing errors, and 9.2 % were removed as spoiled trials. Outlier removal criteria were the same as in Experiment 1 and resulted in the exclusion of 5.8 % of the trials. Thus, on average, 67.1 % of the test phase trials were retained for the RT analysis, which left a mean of 16 items per condition per participant.

Encoding phase performance

Encoding phase error rates and RTs are given in Table 1. Responses were recoded to the dominant and nondominant languages according to the self-reported dominant language of each participant. Error rates were submitted to a 2 (context) × 2 (translation direction) repeated measures ANOVA. Error rates were higher for L1–L2 than for L2–L1 translation, F(1, 95) = 45.31, MSE = .009, p < .001. Error rates were higher for words that appeared in the context of a sentence than for isolated words, F(1, 95) = 15.55, MSE = .006, p < .001. The effects of translation direction and context on error rates did not interact, F < 1. RTs were measured only for the isolated words, and L1–L2 translation was slower than L2–L1 translation, t(95) = 4.39, p < .001.

Test phase performance

Test phase error rates and RTs are given in Table 3. New-item RTs were submitted to a 2 (task) × 2 (response language) repeated measures ANOVA. Overall, translation RTs were longer than picture-naming RTs, F(1, 95) = 127.31, MSE = 60,444, p < .001, and responses were slower when given in L2, F(1, 95) = 20.81, MSE = 117,831, p < .001. These effects did not interact, F(1, 95) = 1.714, MSE = 56,447, p = .194.

Table 3 Mean response times (RTs) and repetition priming in Experiment 2
Full size table

Priming scores were obtained by subtracting repeated condition RTs from the corresponding new-item RTs and are illustrated in Fig. 2. Repetition priming was statistically significant in every cell (ps < .02). Repetition priming scores were submitted to a 2 (task) × 2 (context) × 2 (response language) repeated measures ANOVA. There was no main effect of task, F(1, 95) = 1.15, MSE = 117,244, p = .286, so there was no evidence that priming was greater when translation was repeated from encoding to test than when the task changed to picture naming. A main effect of context showed that priming was greater when the word was isolated at both encoding and test than when it went from contextualized to isolated, F(1, 95) = 7.67, MSE = 73,280, p = .007. A main effect of response language showed greater priming when responding in L2 than when responding in L1, F(1, 95) = 6.62, MSE = 136,508, p = .012. The interaction of context and response language was not significant, F(1, 95) = 2.39, MSE = 30,560, p = .126, nor were there any other interactions, Fs < 1.

Fig. 2
figure 2

Repetition priming in Experiment 2 as a function of language, test task or process shared, and encoding context

Full size image

Production and comprehension priming estimates were derived as in Experiment 1. As indicated by priming in picture naming, priming in production was significant in each context × language combination, ps < .02. In comprehension, no priming was observed in any context × language combination. Production and comprehension priming estimates were submitted to a 2 (process) × 2 (context) × 2 (language) repeated measures ANOVA. A main effect of process showed more priming for production than for comprehension, F(1, 95) = 6.926, MSE = 309,707, p = .01. A main effect of context showed greater priming for words originally translated in isolation, F(1, 95) = 7.772, MSE = 25,991, p = .006. A marginally significant main effect of language showed more priming for L2 processes than for L1 processes, F(1, 95) = 2.947, MSE = 282,309, p = .089. The process × language interaction was not significant, F(1, 95) = 1.520, MSE = 71,525, p = .221, nor were any of the other interactions, Fs < 1.

Discussion

As in our previous research using translation of words presented visually, translation of individual words presented auditorially at study primed both word translation and picture naming at test. The priming effect in picture naming indicates that word production was facilitated, and this facilitation was stronger in L2 than in L1, consistent with the weaker links hypothesis. The priming effect in word translation was not reliably larger than the priming effect in picture naming, indicating that auditory comprehension processes were not detectably facilitated.

The priming of word translation and picture naming following sentence translation at encoding replicates the findings of Experiment 1 with auditory stimulus presentation. Words translated in a sentence context at encoding exhibited priming when translated in isolation at test. Words translated in a sentence context at encoding also exhibited priming when their pictures were named at test, indicating substantial priming in the word production component of translation. This component was stronger in L2 than in L1, consistent with the weaker links hypothesis. The word comprehension component, derived by subtraction, did not exhibit significant facilitation in either language.

Another question was whether embedding the words in a sentence context would reduce priming, relative to words presented in isolation. Overall, priming was reduced for words translated in context at encoding, relative to the words translated in isolation at encoding.

General discussion

In two experiments, we examined repetition priming based on words encoded either in isolation or in a sentence context. In Experiment 1, words to be translated were presented visually, and in Experiment 2, words to be translated were presented auditorially. In both experiments, words that were translated in isolation at encoding produced priming in identical translation at test, as in our previous research (Francis et al., 2011; Francis & Gallard, 2005; Francis & Sáenz, 2007; Francis et al., 2014). Words translated in isolation also produced priming in picture naming with the same response language, as in our previous research (Francis et al., 2003; Francis et al., 2008; Francis & Sáenz, 2007), indicating that production processes were facilitated. In Experiment 1, priming in translation was stronger than priming in picture naming with the same response language, indicating that visual word comprehension processes in translation were also facilitated. However, in Experiment 2, priming for translation and picture naming did not differ, indicating that auditory word comprehension processes in translation were not facilitated.

We demonstrated for the first time that words translated in sentence contexts at encoding produce priming in both translation and picture naming. These effects were observed in both experiments, although the priming effects were generally reduced, relative to the effects found for words translated in isolation. In Experiment 1, visual word comprehension priming for words translated in context was reduced, relative to priming for words translated in isolation, consistent with previous monolingual results showing reduced comprehension priming for words read in a sentence context at encoding (Levy & Kirsner, 1989; MacLeod, 1989; Oliphant, 1983; Smith, 1991; Speelman et al., 2002). In fact, the words translated in a sentence context produced priming in visual word comprehension only in L2.

This study is the first to measure translation RTs for auditorially presented stimulus words. Translation RTs were longer and translation error rates higher for the auditorially presented words in Experiment 2 than for the visually presented words in Experiment 1. This difference is most likely due to the serial nature of auditory stimulus presentation and the fact that there is only one chance to correctly hear the auditory stimulus. It is unlikely to be a result of differences in proficiency between Experiment 1 and Experiment 2 participants, because new-item picture naming RTs are comparable across experiments.

This study is also the first to examine repetition priming in translation with auditory presentation of the stimulus words. Unfortunately, priming of auditory word comprehension processes was so weak as to be undetectable even with 96 participants. This result was unexpected. Models of auditory speech comprehension, such as the bilingual interactive model of lexical access (Grosjean, 1988), describe a number of component processes that must transpire to identify a word presented auditorially. Any of these processes was a potential locus of facilitation when a word was repeated. More specifically, we expected that there would be substantial facilitation in the process of accessing the concept on the basis of the phonological word form, but in fact no facilitation was observed. One possible explanation to consider is that the processes were overlearned; that is, they had been practiced so much that the impact of additional exposures on speed of processing were too small to detect.

It is unknown how long repetition priming for words encoded in a sentence context might last. Repetition priming for isolated words in picture naming lasts for at least 1 year (Cave, 1997), and repetition priming for isolated words in translation lasts for at least 1 week (Francis & Sáenz, 2007). The procedural learning exhibited in repeated readings of text passages also lasts for over 1 year (Kolers, 1976). These findings suggest that priming from the translation of contextualized words may also last over a long period of time, and this is an empirical question for future research.

Language dominance effects on transfer and their possible mechanisms

Translation RTs were longer and error rates were higher when responding in L2 than when responding in L1. The RT effect did not reach significance in Experiment 1, but it did in Experiment 2, thus exhibiting the typical translation asymmetry (e.g., Kroll & Stewart, 1994; Sholl et al., 1995). Across experiments, repetition priming for translation and picture naming was stronger when responses were given in L2 than when responses were given in L1. The effect in picture naming indicates that production processes were primed more in L2 than in L1. In Experiment 1, comprehension processes in the visual modality were primed more in L2 than in L1.

The effects of encoding context on priming differed for L1 and L2. Specifically, in Experiment 1, context effects in comprehension and production were smaller in L2 than in L1, indicating a greater degree of transfer for contextualized, relative to isolated, words in L2 than in L1. Even if considered proportionally, the proportion of transfer for contextualized, relative to isolated, words in L2 was greater than in L1. There are at least two possible reasons for the greater proportion of transfer in L2 than in L1.

First, the major models of bilingual lexical processing, including the bilingual interactive activation model (BIA+; Dijkstra & Van Heuven, 2002), the inhibitory control model (IC; Green, 1998), and the revised hierarchical model (RHM; Kroll & Stewart, 1994), are in agreement that L2 words have weaker links to their concepts than do L1 words, just as low-frequency words have weaker links to their concepts than do high-frequency words. In fact a weaker-links hypothesis has been used to describe both frequency effects and language dominance effects using the same mechanism (Gollan et al., 2008). On the basis of the similarity between language dominance and word frequency effects, we would expect the transfer from contextualized words at encoding to isolated words at test to be greater in L2 than in L1, paralleling the greater transfer observed for low-frequency words than for high-frequency words (Nicolas, 1996). Thus, the results are consistent with the weaker-links hypothesis.

Another explanation for the greater transfer from contextualized words at encoding to isolated words at test in L2 than in L1 is that reading is more difficult in L2 than in L1 or, similarly, that bilinguals are less skilled readers in L2 than in L1. Making a reading task at encoding more difficult increased transfer to isolated words at test (Nicolas, 1998). Similarly, poor readers showed greater transfer than more skilled readers (Bourassa et al., 1998). Thus, bilinguals may show greater priming from words encoded in context to isolated words at test because of the lower functional frequency of L2, relative to L1, words or because of the greater difficulty of reading in L2, relative to L1.

Other implications for models of bilingual lexical access

This study adds to the existing evidence that translation of isolated words is concept mediated in both directions in fluent early bilinguals (e.g., De Groot et al., 1994; Duyck & Brysbaert, 2004; Miller & Kroll, 2002; Potter et al., 1984). Specifically, the findings that L1–L2 translation primes L2 picture naming and that L2–L1 translation primes L1 picture naming support the conclusion that translation is concept mediated. This finding extends to words translated in a sentence context at encoding. Therefore, we can draw the new conclusion that concept mediation extends to words translated in a sentence context. Concept mediation of contextualized translation in proficient bilinguals can be accommodated in current models of bilingual lexical access, including the RHM (Kroll & Stewart, 1994), the BIA+ (Dijkstra & Van Heuven, 2002), and the IC (Green, 1998).

According to the BIA+ model, the recognition of words in a sentence context is different from the recognition of isolated words, in that it is sensitive to syntactic and semantic context information (Dijkstra & Van Heuven, 2002). This sensitivity to syntactic and semantic context may explain, in part, why the operations performed on words in the context of a sentence are somewhat different from those performed on isolated words. That is, this context may cause the words to be contextually bound such that less individuating information is encoded than for words presented and processed in isolation (Masson & MacLeod, 2000). The syntactic and semantic frames for the sentences in the present study were relatively simple; one might expect that with more complex sentences at encoding, transfer to the translation of isolated words at test would be reduced.

A proceduralist account of the results

According to the proceduralist view described by Kolers and Roediger (1984), a person who is less skilled at the encoding task must apply more detailed analytical effort to acquire a stimulus. This additional analysis leads to a more extensive long-term representation than that acquired by a more skilled encoder. Therefore, less-skilled individuals will show better memory for the encoded items. This view may explain three aspects of the present results. First, language skill level varied systematically within individuals, such that comprehension and production were less skilled in L2 than in L1. Consistent with the implications of this view, repetition priming was stronger in L2 comprehension and production than in L1 comprehension and production. Second, within this view, one might speculate that the reason for the reduction in priming with context at encoding is that contextualized words are easier to encode and, therefore, produce less transfer. Finally, translation is a task that requires more detailed analytic effort than does reading, which may explain why the repetition priming effects based on contextualized translation at encoding are stronger (relative to isolated words) than those observed in previous research following contextualized reading at encoding.

Conclusions

Words translated in a sentence context elicited repetition priming in visual word comprehension processes and in spoken word production processes. However, the priming elicited by words translated in a sentence context was only reliable for L2 in visual word comprehension, and priming in auditory word comprehension was not reliable in either language. Patterns of priming in partial repetition conditions were compatible with the principle of transfer-appropriate processing (Morris et al., 1977; Roediger & Blaxton, 1987). The finding that priming was stronger in L2 processes than in L1 processes was consistent with Kolers and Roediger’s (1984) proceduralist view of task transfer. The finding that priming in L2 processes was stronger than priming in L1 processes and the finding that priming from contextualized words was proportionally stronger in L2 than in L1 were compatible with the weaker-links hypothesis (Gollan et al., 2008). Translation for both isolated and contextualized words was concept mediated, consistent with modern models of bilingual lexical access. The results suggest that the long-term learning indicated by repetition priming could occur with natural language exposures to words, particularly comprehension exposures to written words and self-generated spoken production exposures.