Learning the two translations of translation-ambiguous words: Simultaneous vs. consecutive presentation

Abstract

Some words have more than one translation across languages. Such translation-ambiguous words are harder to learn, recognize, and produce for individuals across the language learning spectrum. Past research demonstrates that learning both translations of translation-ambiguous words on consecutive trials confers an accuracy advantage relative to learning them on separate sessions. We tested the proposal that presenting the two translations of translation-ambiguous words simultaneously on the screen could facilitate the formation of a more integrated mapping, because this would enable learners to make direct comparisons between them, whether implicitly or explicitly. We predicted that this facilitation would especially hold for translation-ambiguous words with related translations. Fifty native English speakers learned 48 German words with one or two translations that varied in the meaning similarity of their translations. Paired associate training took place on a Monday, and a first language (L1) to second language (L2) translation production test took place on Wednesday and Friday. Generally, higher translation similarity facilitated translation speed. In accuracy, training condition interacted with the similarity of the translations; translation accuracy was more affected by translation similarity in the simultaneous condition and went up as similarity increased. Overall, the consecutive condition demonstrated higher accuracy and faster reaction times than the simultaneous training condition, suggesting that learners may have been unable to successfully divide their study time between multiple words on the screen without explicit instruction.

Keywords

training order translation ambiguity translation semantic variability vocabulary learning

I Introduction

The bedrock of becoming proficient in a second language (L2) is learning a large number of vocabulary words in that language. However, just like many words within a language are ambiguous in that they have more than one meaning (Klein and Murphy, 2001), research has demonstrated that many words across languages are ambiguous in that they have more than one translation (e.g. Basnight-Brown and Altarriba, 2014; Prior et al., 2007; Tokowicz et al., 2002; Tseng et al., 2014; Wen and van Heuven, 2017). For example, the German word Kiefer translates to the English words ‘pine’ and ‘jaw’, whereas the German word Obst translates to the English word ‘fruit’. Words with multiple translations are referred to as ‘translation ambiguous’, and these words are more difficult to learn than ‘translation unambiguous’ words with a single translation¹ at the beginning stages of L2 learning for naive learners (Degani and Tokowicz, 2010; Degani et al., 2014). This difficulty with translation-ambiguous relative to translation-unambiguous words is referred to as the ‘translation-ambiguity disadvantage’; this disadvantage persists throughout the L2 proficiency continuum (for reviews, see Schweiter and Prior, 2020; Tokowicz, 2014), and applies to both naive learners and lifelong bilinguals (Tokowicz, 2014). Furthermore, the difficulty applies to performance on both translation recognition (Boada et al., 2013; Degani and Tokowicz, 2010; Degani et al., 2014; Eddington and Tokowicz, 2013; Laxén and Lavaur, 2010; Prior et al., 2013) and translation production tasks (Basnight-Brown and Altarriba, 2014; Basnight-Brown et al., 2020; Degani and Tokowicz, 2010; Degani et al., 2014; Prior et al., 2013; Tokowicz and Kroll, 2007).

In a study of translation-ambiguous word learning, Degani et al. (2014) manipulated the way that the two translations of translation-ambiguous words were presented to naive native English-speaking learners of Dutch to determine whether one of the presentation methods would yield a smaller translation-ambiguity disadvantage. In one condition, the two translations were presented on consecutive trials. In the other, the two translations were presented two days apart. Translations trained consecutively showed an accuracy advantage relative to those trained on separate sessions, such that the translation-ambiguity disadvantage (for translation-ambiguous relative to translation-unambiguous words) was eliminated for words trained consecutively. Degani et al. concluded that consecutive trial presentation benefited learners by enabling them to compare the two translations to each other. In addition, it signaled from the time of initial learning that the translation-ambiguous words were in fact translation ambiguous and not translation unambiguous. By contrast, for the words with translations trained on separate days, an initial assumption was made that these words were translation-unambiguous, and this had to be overridden when the learner was taught the second translation, thereby requiring remapping or reorganization in memory. Notably, there was a disadvantage for these second-learned translations in later tests.

Note that not all translation-ambiguous words have the same relationship between their translations. For example, in another manipulation in this study, the English stimuli had either two Dutch translations that reflected separate interpretations of the word (e.g. ‘sheet’ translates to laken for bed sheet or blad for sheet of paper) or two translations that meant roughly the same thing (e.g. ‘boot’ translates to laars and schoen). Several models have been proposed to describe the way that these different types of translation-ambiguous words may be represented in the bilingual mind. For example, according to the Revised Hierarchical Model of Translation Ambiguity (RHM-TA, Eddington and Tokowicz, 2013; see Figure 1), translation-ambiguous words like sheet have two Dutch lexical representations, and each are linked to distinct conceptual representations (Figure 1a). By contrast, translation-ambiguous words like boot have two Dutch lexical representations, but they are linked to a single, shared conceptual representation (Figure 1b). Translation-unambiguous words have only a single lexical representation in each language, which are linked to a shared conceptual representation (Figure 1c).

Figure 1.

The Revised Hierarchical Model of Translation Ambiguity (RHM-TA): (a) represents a translation-ambiguous word with unrelated translations; (b) represents a translation-ambiguous word with related translations; (c) represents a translation-unambiguous word.

Degani et al. (2014) found that accuracy was significantly lower for words that had translations that meant roughly the same thing (e.g. ‘boot’). These findings suggest that the translation-ambiguity disadvantage arises when it is not possible to create one or more one-to-one mappings in memory between the L2 target form and meaning (see also Bracken et al., 2017). In the case of a word with multiple translations with unrelated meanings (e.g. ‘sheet’/laken/blad), the learner can create two one-to-one mappings: one between ‘sheet’ and laken, and one between ‘sheet’ and blad. In the case of boot, this is not possible.

However, it is important to note that the meaning similarity of the translations of translation-ambiguous words is more likely continuous than categorical. Therefore, in subsequent research, we began examining this relationship in a continuous fashion using a variable called translation semantic variability (TSV; Bracken et al., 2017). TSV is the semantic similarity of the meanings of the translations of a word, as rated by a separate set of native speakers. It is rated on a scale from 1 (unrelated) to 7 (related); higher TSV indicates higher meaning similarity. Words like ‘boot’ would be rated as higher in TSV than words like ‘sheet’. The model of translation ambiguity presented by Laxén and Lavaur (2010) captures this continuous relationship nicely (see Figure 2).

Figure 2.

Depiction of translation-ambiguous words with more (top) and less (bottom) related translations.

Figure 2 demonstrates the difference between translation-ambiguous words that are relatively more (top panel) or less (bottom panel) related in meaning. This model uses a distributed features framework based on the Distributed Conceptual Feature and Distributed Representation Models of bilingual memory (e.g. de Groot, 1992a, 1992b; van Hell and de Groot, 1998), in which word meanings are represented as sets of distributed features. In the case of translation-ambiguous words with more overlapping meanings, the French words mari and époux are connected to more of the same semantic features (top panel). In the case of translation-ambiguous words with less overlapping meanings, the French words ongle and clou are connected to fewer of the same semantic features (bottom panel).

II The present study

In the present study, we taught native English speakers German–English translation-unambiguous and translation-ambiguous translation pairs. These pairs varied in the semantic relatedness of the two English translations of the translation-ambiguous words. We tested the proposal that presenting the translations of translation-ambiguous words simultaneously on the screen could facilitate the formation of a more integrated mapping. Seeing words simultaneously on the screen allows for direct comparisons between them (either implicitly or explicitly), and that process is what allows the relationships among the words to be more apparent to the learner.

Research on text integration supports the idea that proximity of information facilitates structural integration (i.e. integration that occurs during learning; Walker and Meyer, 1980). This idea is also supported by previous research on the spacing effect for learning vocabulary from context, which suggests that spacing presentation is not beneficial when trying to derive meanings from context (Elgort and Warren, 2014; Webb and Chang, 2015), and instead presenting exemplars more closely together may assist learners in forming coherent, integrated representations. A final example of learning that benefits from placing information more closely together comes from demonstrations in the inductive learning literature, in which learners benefited from closer placement of information that needed to be assimilated, particularly when the information that was being assimilated was not exact (e.g. two translations) (e.g. Appleton-Knapp et al., 2005; Glover and Corkill, 1987).

As noted previously, a study by Degani et al. (2014) demonstrated that presenting the two translations of translation-ambiguous words on consecutive trials during the same session led to superior performance relative to presenting the translations on separate sessions. It was concluded that presentation in the same session enabled learners to build an appropriate representation sooner than was possible when the translations were presented on separate sessions. They further concluded that when translations were presented on separate sessions, learners initially created a one-to-one mapping that later had to be overridden.²

Here, we tested the prediction that simultaneous presentation of the two translations may facilitate formation of an appropriate mapping even earlier than in the previous study. This is because the learner would be aware from the earliest possible point in learning that the German target word has two English translations. We contrasted this with the consecutive presentation condition from the previous study.

We further tested the idea that simultaneous presentation may be more beneficial for translation-ambiguous words that have translations that are more related in meaning (i.e. that are higher in TSV). For example, the German word Tüte, translates to ‘bag’ and ‘sack’, and this could be integrated into a one-to-one mapping: Tüte→ bag/sack. (Note here that bag and sack represent the meanings rather than the English words.) However, for words that have quite different meanings (i.e. are lower in TSV), such as Kiefer, which translates to ‘pine’ and ‘jaw’, this would not be possible, and simultaneous presentation could instead amplify the differences between the translations (see Figure 3). Indeed, previous research suggests that when processing stimuli under difficult conditions it is beneficial to have the stimuli presented together, if the learner will benefit from comparing the stimuli in some way (Loess and Duncan, 1952). There is also evidence in the literature of integrated representations being formed when information is presented simultaneously but not sequentially. For example, Congdon et al. (2017) found that gesture supports speech during instruction only when it is presented simultaneously with speech, forming an integrated representation. A similar finding of support for the idea that bodily gestures support speech comes from work by Liu et al. (2007) who observed that visual speech (speech with the articulators simultaneously visualized via a simulation) were superior for learning in an adult population. This type of simulation permits the learner to integrate the sound of the word and the visualization of the articulators.

Figure 3.

Depiction of German words with a one-to-one and one-to-two mapping.

To quantify translation similarity, we used the TSV measure mentioned above. The TSV ratings were obtained for the semantic similarity of the meanings of the English translations of the German words, as rated by a separate set of native English speakers in a published normative study. Note that this examines the effect of an existing relationship in the first language (L1), i.e. in English.

The stimuli in this study varied in terms of number of translations, and for translation-ambiguous words, the relationship between their translations. The stimuli also varied in terms of their word-specific characteristics (e.g. frequency, concreteness). Although we matched the important sub-groups of stimuli on these important characteristics, we also included these characteristics in our analyses so that we could determine the extent to which they influenced our findings. Word concreteness may be of particular interest because it plays an important role in some models of bilingual memory representation. Specifically, as noted above, the Distributed Feature Model (e.g. de Groot, 1992b) and Distributed Representation Model (van Hell and de Groot, 1998) represent meanings as sets of distributed features that are linked to lexical representations. These models propose that translations of words higher in concreteness share more semantic features across languages than translations of words lower in concreteness, and as a result words higher in concreteness will be translated more quickly and accurately (e.g. de Groot, 1992b; de Groot et al., 1994; van Hell and de Groot, 1998).

We hypothesized that we would observe an interaction between TSV and training condition, such that simultaneous training of translations would improve accuracy on the translation production task when TSV was higher. This should be the case because the relationship between the two English translations of the German words would be most salient when they were trained simultaneously. We also expected that for translation pairs lower in TSV, simultaneous training would lower accuracy because it would highlight the differences in meaning, which would be detrimental to learning. We also hypothesized that consecutive presentation would not yield a translation-ambiguity disadvantage in accuracy consistent with previous research (Degani et al., 2014).

In addition to the effects of these variables, we examined the effects of word-specific characteristics known to influence translation performance. These included word length (English and German), word frequency (English), and word concreteness (English). Based on previous research, we expected that reaction time would be faster, and accuracy would be higher for words shorter in length (e.g. de Groot, 1992b; de Groot et al., 1994), higher in frequency (e.g. de Groot, 1992b; de Groot et al., 1994), and higher in concreteness (e.g. Basnight-Brown and Altarriba, 2016; de Groot, 1992b; de Groot et al., 1994).

III Method

1 Participants

Fifty students (23 females, 22 males, 5 of undisclosed gender; mean age = 19.04 years) from an American midwestern university participated in this study towards an Introduction to Psychology research participation requirement. All participants were right-handed native speakers of English and had no previous knowledge of Dutch or German. 11 additional participants were tested but not included in the analyses due to audio recording problems or failure to return for one of the sessions. Participants filled out a language history questionnaire (Tokowicz et al., 2004) after participation in the final session. All participants gave written informed consent and the study was approved by the University’s Institutional Review Board.

2 Design

The study used a two ambiguity status (translation ambiguous vs. translation unambiguous) within-participants design. Additionally, training condition was manipulated within participants: translation-ambiguous words were trained either simultaneously (both translations presented at once) or consecutively (translations presented on consecutive trials).

3 Stimuli

Forty-eight German words were trained during this experiment (24 translation-ambiguous German words and 24 translation-unambiguous German words). These German words and their translations were chosen from a set of previous translation norms (Eddington et al., 2022; for stimuli, see Appendices 1 and 2, and https://osf.io/kuz9c). Translation-ambiguous and unambiguous words were matched (ts < 1) on German and English length, concreteness of translations (Brysbaert et al., 2014) and translation log SUBTL_US word frequency (Brysbaert and New, 2009). Furthermore, the two English translations of the German words were matched along all of these dimensions as well (Fs < 1). Translation-ambiguous and unambiguous words were matched on orthographic neighborhood (number of orthographic neighbors; from the English Lexicon Project; Balota et al., 2007), F (1, 53) = 1.85, p = .18, and the two English translations of the German words were matched on this dimension, F (1, 27) = 2.91, p = .10. For an overview of the stimuli, see Table 1.

Table 1.

Stimulus characteristics by translation number.

	Translation 1	Translation 2
German length	6.54	n/a
English length	5.44	5.63
Frequency (log SUBTL_US)	3.22	3.29
Concreteness	3.50	3.45
Orthographic neighborhood	5.87	4.86

The translation-ambiguous words were split into two lists. The two lists of translation- ambiguous words were counterbalanced for the training portion of the study (yielding four training versions), such that each translation was presented an equal number of times as both the first and second translation. Additionally, each translation was presented an equal number of times in the consecutive and simultaneous conditions. These lists were matched according to the means of German length, English translation length, log English word frequency, TSV, and concreteness (Fs < 1). An error in counterbalancing resulted in the translation of the German word Versuch being presented in the same order for three versions instead of two.

4 Procedure

The experiment took place over the course of three sessions on Monday, Wednesday, and Friday of the same week (e.g. Terrazas, 2018). This spacing is used to encourage memory consolidation during sleep (e.g. Dumay and Gaskell, 2007), and to facilitate scheduling. During the first session, participants were instructed to learn the German–English translation pairs (“Your task is to learn these German words and their translations”). They were alerted that some German words may have more than one translation into English. E-Prime (Psychology Software Tools, Sharpsburg, PA) was used for stimulus presentation. For translation-unambiguous words, each trial began with a screen showing only a fixation cross until the participant pressed the space bar to initiate the training trial. After a 100 ms blank screen, a German word and its corresponding translation appeared (German word – English word) for 8 seconds, at which time the trial ended, and the fixation cross reappeared.

The translation-ambiguous word pairs were trained in either consecutive or simultaneous conditions. For the consecutive condition, the same structure described above was used, with the only difference being that each trial consisted of two translation pairs, presented such that there was a fixation cross, then translation pair 1, following by a fixation cross, then translation pair 2. Each pair was presented for 8 seconds. No additional information was provided to the participant, but they likely realized that the same German word was shown twice in a row. For the simultaneous condition, each trial consisted of a fixation cross, followed by a screen that showed both translation pairs on the same screen (German word – English word), with one pair arranged above the other; the German word was repeated. This screen was shown for 16 seconds, to match the amount of time, per translation, of the consecutive training condition.

The full training session consisted of three cycles, resulting in each translation pair being presented three times. All trials were intermixed in random order determined by EPrime, other than the intentional consecutive order for that training condition. Participants were not informed of the training difference between consecutive and simultaneous trials, though sample items of these types were included in practice trials preceding the actual training.

a Free recall

Participants completed a free recall task immediately after training to reinforce learning (e.g. Karpicke and Roediger, 2008). Because this task was used solely to reinforce learning, these data were not analysed and this task will not be discussed further. This task was done using a Microsoft Excel spreadsheet with a German Word and an English Word column. The instructions indicated that participants should type every German word that could be recalled, regardless of whether the English translation could also be recalled. There was no time limit for this task.

b Translation

A German to English translation production task was conducted on Sessions 2 and 3 to test retention. This task is often used with beginning learners (e.g. Degani and Tokowicz, 2010; Degani et al., 2014; Kaushanskaya and Marian, 2009). It was selected for use in this study because it permits us to examine the connections that are formed between L1 and L2 words during training, while also gaining more data than the reverse direction of translation because participants respond in their L1. Furthermore, this direction of translation is sensitive to meaning variables such as word concreteness (e.g. de Groot et al., 1994).

EPrime was used for word presentation and the recording of reaction time. All German items were included in the task; the order of German words to be translated was randomized by E-Prime. Unambiguous word trials began with the presentation of a fixation cross centered on the computer screen. Participants then pressed a button to advance to a screen with the German word. Participants vocally produced a translation into a microphone and digital voice recorder simultaneously. As soon as the participant started to produce the translation, the word disappeared and a fixation cross for the next trial appeared. Reaction time for this task was recorded, by EPrime, as the amount of time that elapsed from the word presentation to the beginning of utterance production. Accuracy coding was performed offline by two independent coders.

For ambiguous items, a similar procedure was used, with the only difference being that participants were told that they would be tested on both translation-ambiguous and unambiguous words. They were also informed that words with more than one translation would be designated by a ‘1’ and ‘2’ on consecutive trials, indicating the first and second opportunity to translate the German words with multiple translations; they were told to respond only with one translation per trial. For example, the German word Boden would be presented as Boden 1 during the participant’s first opportunity to provide a translation, with the next German word being Boden 2 (for scoring information, see Section IV1a). We gave participants the opportunity to provide both translations because we wanted to determine if they had learned both translations of the word, particularly because accuracy tends to be a more sensitive measure for beginning learners (e.g. Degani and Tokowicz, 2010; Degani et al., 2014).³

IV Results

Data were analysed using linear mixed effects models in R Studio (RStudio Team, 2016) with the lmer and glmer commands of the lmerTest package (Kuznetsova et al., 2017). Variance inflation factors (VIF) were checked to verify that we did not have an unacceptable level of multicollinearity in the models (VIF below 10 considered acceptable, Belsley et al., 1980; all VIFs ⩽ 9.78). We initially built all models using the maximal random effects structure (e.g. Barr et al., 2013) but removed random slopes when necessary to permit models to converge.

During the translation production task, the voice key was not triggered correctly by the participant on a small number of trials, either because an extraneous noise was made prior to the actual response (e.g. a cough, sigh, ‘um’, etc.), or because the participant didn’t speak loudly enough to trigger the voicekey and therefore repeated the response (2% of trials). For the purposes of the accuracy analyses, such responses were treated as correct responses because the correct response was in fact given, whereas for the reaction time analyses, such responses were not included because a valid reaction time was not available. Reaction time analyses were conducted on correct trials only. We also focused our reaction time analyses on the first translation provided for the translation-ambiguous words because the second translation could have been prepared as part of the response to the first word, and therefore the reaction time to the second word may have been artificially deflated.

1 Accuracy

a Analysis comparing translation-ambiguous and unambiguous words

This analysis describes the extent to which participants were able to correctly translate the German words into English. For translation-unambiguous words, this meant providing a single translation, whereas for translation-ambiguous words, this meant providing two translations, one on each of two consecutive trials. To be considered correct, we used a conservative criterion whereby both translations of a translation-ambiguous word had to be provided. This variable was coded as 0 for correct and 1 for incorrect. This analysis examined the potential differences between translation-ambiguous and unambiguous words, as well as differences as a function of training condition (simultaneous vs. consecutive) and session.

The fixed effects of the analysis were German word length, English word length, English word frequency, English word concreteness, and the interaction between session and training condition. The English word variables were averaged across the two translations for translation-ambiguous words; numerical predictors were centered. The random effects were participants and the German word, and the model included random slopes for all fixed effects except Session. The model equation is given in Table 2. This analysis demonstrated that German words of longer length yielded lower accuracy, β = - 0.25, z = −4.14, p < .01, and English words of higher concreteness yielded higher accuracy, β = 0.41, z = 3.03, p < .01. Furthermore, there was a main effect of training condition such that the simultaneous translation-ambiguous condition had lower accuracy than the translation-unambiguous condition, β_simultaneous = −0.73, z = −2.35, p < .05, whereas the consecutive condition was not significantly different from the translation-unambiguous condition, β_consecutive = −0.35, z = −1.07, p = .28 (see Figure 4).

Table 2.

Fixed effects for the model for translation production accuracy for all words.

Fixed effect	β	SE	z value	Pr(>\|z\|)
(Intercept)	−1.65	0.28	−5.90	< .001**
GermanLength.cen	−0.25	0.06	−4.14	< .001**
EnglishAvLen.cen	−0.14	0.09	−1.44	0.15
EnglishAvFreq.cen	−0.06	0.33	−0.17	0.86
EnglishAvConc.cen	0.41	0.13	3.03	< .001**
Session2	0.08	0.11	0.70	0.48
TrainingConditionTwoConsec	−0.35	0.32	−1.07	0.28
TrainingConditionTwoSimult	−0.74	0.31	−2.35	0.02*
Session2:TrainingConditionTwoConsec	−0.05	0.20	−0.24	0.81
Session2:TrainingConditionTwoSimult	−0.04	0.21	−0.21	0.83

Notes. glmer(EntireAcc~1+GermanLength.cen+EnglishAvLen.cen+EnglishAvFreq.cen+EnglishAvConc.cen + Session * TrainingCondition + (1|Participant) + (1|GermanWord) + (0+GermanLength.cen+EnglishAvLen.cen+EnglishAvFreq.cen+EnglishAvConc.cen+TrainingCondition||Participant),data=ZacTranslationAcc,family=binomial,glmerControl(optimizer = ‘bobyqa’, optCtrl = list(maxfun = 10000))). * p < .05. ** p < .01. Number of observations: 4,800.

Figure 4.

Translation production accuracy as a function of training condition.

b Analysis of translation-ambiguous words only

This analysis examined the translation-ambiguous words only. We also included TSV in this analysis and explored its potential effect. The fixed effects of the analysis were German word length, English word length, English word frequency, English word concreteness, and the interaction between session and training condition. This analysis also included TSV to examine a potential main effect of this factor on accuracy; the potential interaction of TSV and training condition was also examined because the relationship between the two English translations of a German word may be more salient when the two words are trained on the screen at the same time rather than on consecutive trials. Random effects were participants and the German word; random slopes for the interaction between session and training condition, and the interaction between TSV and training condition were excluded from the model.

The fixed effects of the analysis and the model equation are given in Table 3. Longer German words were associated with lower accuracy, β = −0.22, z = −1.92, p = .051. In addition, there was an interaction between TSV and training condition in the simultaneous condition, β_simultaneous = 0.21, z = 1.95, p = .05. Although this result has to be interpreted somewhat cautiously because it is significant at p = .05, it demonstrates that at lower levels of TSV, there is an accuracy advantage for consecutive trials, whereas at higher levels of TSV there is a very small difference between the two training conditions (see Figure 5).

Table 3.

Fixed effects for the model for translation production accuracy for ambiguous words.

Fixed effect	β	SE	z value	Pr(>\|z\|)
(Intercept)	−1.84	0.43	−4.29	< .001**
GermanLength.cen	−0.22	0.11	−1.92	0.06.
EnglishAvLen.cen	−0.24	0.19	−1.21	0.23
EnglishAvFreq.cen	0.19	0.52	0.36	0.72
EnglishAvConc.cen	0.30	0.22	1.35	0.18
TSV.cen	0.10	0.17	0.57	0.57
TrainingConditionTwoSimult	−0.56	0.40	−1.40	0.16
Session	0.05	0.17	0.26	0.79
TSV.cen:TrainingConditionTwoSimult	0.21	0.11	1.95	0.05.
TrainingConditionTwoSimult:Session	−0.01	0.25	−0.05	0.96

Notes. glmer(EntireAcc~1+GermanLength.cen+EnglishAvLen.cen+EnglishAvFreq.cen + EnglishAvConc.cen+TSV.cen*TrainingCondition+Session*TrainingCondition+(1|Participant) + (1|GermanWord)+(0+GermanLength.cen+EnglishAvLen.cen+EnglishAvFreq.cen + EnglishAvConc.cen+TSV.cen||Participant),data=ZacTranslationAcc.AmbigOnly, family=binomial,glmerControl(optimizer = ‘bobyqa’, optCtrl = list(maxfun = 100000))). p < .10. ** p < .01. Number of observations: 2,400.

Figure 5.

Translation production accuracy for translation-ambiguous words as a function of training condition and TSV.

2 Reaction time

Reaction times on correct trials that were shorter than 100 ms and longer than 2.5 SDs from the mean (8944 ms) were not analysed; this constituted 4.3% of the data.

a Analysis comparing translation-ambiguous and unambiguous words

This analysis describes the speed with which participants were able to correctly translate the German words into English. This analysis examined the potential differences between translation-ambiguous and unambiguous words, as well as differences as a function of training condition (simultaneous vs. consecutive).

The fixed effects of the analysis and the model equation are given in Table 4. This analysis demonstrated that German words of longer length yielded slower reaction times, β = 114.99, t = 3.16, p < .01. English words of higher concreteness yielded faster reaction times, β = −156.60, t = −2.15, p < .05. Responses were faster on the second session than the first, β = −662.85, t = 4.93, p < .01. Furthermore, there was an effect of training condition (see Figure 6), β_consecutive = 684.01, t = 1.83, p = .06, β_simultaneous = 1371.60, t = 3.47, p < .01. This effect demonstrates that reaction times for the simultaneous condition were significantly slower than those in the unambiguous condition, and reaction times in the consecutive condition were marginally slower than those in the unambiguous condition (see Figure 6).

Table 4.

Fixed effects for the model for translation production reaction time for all words.

Fixed effect	Estimate	Standard Error	df	t	Pr(>\|t\|)
(Intercept)	3576.72	264.15	398.88	13.54	< .001**
GermanLength.cen	114.99	36.34	40.43	3.16	< .001**
EnglishAvLen.cen	74.25	54.11	29.11	1.37	0.18
EnglishAvFreq.cen	−271.24	177.82	45.97	−1.53	0.13
EnglishAvConc.cen	−156.6	73.02	40.25	−2.15	0.04*
Session	−662.85	134.42	1046.16	−4.93	< .001**
TrainingConditionTwoConsec	684.01	373.27	820.28	1.83	0.07.
TrainingConditionTwoSimult	1371.6	395.9	868.57	3.47	< .001**
Session:TrainingConditionTwoConsec	−60.16	224.78	1043.91	−0.27	0.79
Session:TrainingConditionTwoSimult	−339.97	236.56	1042.17	−1.44	0.15

Notes. lmer(Amb1UnamRTTrim~1+GermanLength.cen+EnglishAvLen.cen+EnglishAvFreq.cen+EnglishAvConc.cen+Session*TrainingCondition+(1|Participant)+(1|GermanWord)+(0+GermanLength.cen+EnglishAvLen.cen||Participant),data=ZacTranslationRT,control=lmerControl(optimizer = ‘bobyqa’, optCtrl = list(maxfun = 100000))). . p < .10. * p < .05. ** p < .01. Number of observations: 1,774.

Figure 6.

Translation production reaction time as a function of training condition.

b Analysis of translation-ambiguous words only

For the translation-ambiguous words only, we examined the reaction time for the first translation given and explored potential differences as a function of training condition and TSV. The potential interaction of TSV and training condition was also examined because, as mentioned above, the relationship between the two English translations of a German word may be more salient when the two words are trained on the screen at the same time than on consecutive trials.

The fixed effects of the analysis and the model equation are given in Table 5. Longer German words yielded longer reaction times, β = 152.29, t = 2.55, p < .05. In addition, reaction times were faster on the second session, β = −704.41, t = −3.88, p < .01. Finally, reaction times were shorter for translation pairs higher in TSV (i.e. with more related meanings), β = −192.18, t = −2.03, p < .05.

Table 5.

Fixed effects for the model for translation production reaction time for ambiguous words.

Fixed effect	Estimate	Standard error	df	t value	Pr(>\|t\|)
(Intercept)	4323.16	327.62	411.97	13.20	< .001**
rtGermanLength.cen	152.29	59.71	26.31	2.55	0.02*
rtEnglishAvLen.cen	57.55	93.77	22.64	0.61	0.55
rtEnglishAvFreq.cen	−30.89	394.68	27.93	−0.08	0.94
rtEnglishAvConc.cen	−170.61	103.53	16.92	−1.65	0.12
TSV.cen	−192.18	94.71	45.79	−2.03	0.05.
TrainingConditionTwoSimult	643.44	440.61	522.93	1.46	0.14
Session	−707.41	182.46	488.67	−3.88	< .001**
TSV.cen:TrainingConditionTwoSimult	115.99	124.39	272.37	0.93	0.35
TrainingConditionTwoSimult:Session	−288.76	269.09	489.61	−1.07	0.28

Notes. lmer(Amb1UnamRTTrim~1+GermanLength.cen+EnglishAvLen.cen+EnglishAvFreq.cen+EnglishAvConc.cen+TSV.cen*TrainingCondition+Session*TrainingCondition+(1|Participant)+(1|GermanWord)+(0+rtGermanLength.cen+rtEnglishAvLen.cen+rtEnglishAvFreq.cen||Participant),data=ZacTranslationRTAmbigOnly,control=lmerControl(optimizer = ‘bobyqa’, optCtrl = list(maxfun = 100000))).

p < .10. * p < .05. ** p < .01. Number of observations: 595.

V Discussion

In this study, we asked whether presenting the translations of translation-ambiguous words simultaneously on the screen could facilitate the formation of a more integrated mapping: at least for some translation-ambiguous words. As reviewed in the introduction, our hypothesis that simultaneous training of translations would improve accuracy on the translation production task when TSV was higher was based on past research that the proximity of information in space and time facilitates structural integration and the formation of integrated representations. Furthermore, the specific prediction that TSV and training condition would interact is consistent with the models of translation ambiguity in bilingual language representation we presented in the introduction (e.g. Eddington and Tokowicz, 2013; Laxén and Lavaur. 2010). When the two to-be-trained words are presented simultaneously on the screen, the extent to which they overlap at the conceptual level is evident and highlights the (dis)similarity of the translations. Translation overlap is much less evident when the to-be-trained words are presented on separate trials. In line with this, we did indeed find an interaction between TSV and training condition as hypothesized (though this was significant only at p = .05), such that translation accuracy was more affected by TSV in the simultaneous condition: accuracy increased with TSV. When the translations’ meanings are more related, they are easier to integrate into a one-to-one mapping (e.g. Tüte/bag/sack), whereas this is not possible when their translations are less similar in meaning (e.g. Kiefer/pine/jaw), where the differences are made more salient by the simultaneous presentation (see Figure 3). This suggests that learners, under at least some circumstances, do some comparison of the translations, and this is to their detriment if they are very different from each other, and particularly so if this difference is highlighted.

Notably, however, performance in the consecutive presentation condition was superior to the simultaneous condition, both in terms of accuracy and reaction time. This could be due to the fact that participants had to determine how to divide their attention among the two translations on the screen in the simultaneous presentation condition. We did not provide participants with instructions regarding strategies to use when studying the material on the training screen, therefore it is possible that by presenting multiple translations on the screen simultaneously overwhelmed them. To try to mitigate this, we allowed additional time: the same amount of time as the two consecutive trials together. However, without direction, we do not know how participants managed their time, and unfortunately, we did not ask them following the experiment. In a future experiment, it would be useful to ask participants what strategies they used, or to give varied instructions (e.g. study each word for half the time). Perhaps a more useful instruction may be to help learners try to integrate the two translations (e.g. the keyword method), or to try to find similarities or differences between them; in other words, to ask learners to do something active during the study task. Nevertheless, when TSV was highest, the difference between the training conditions disappeared (see Figure 5), demonstrating the powerful role that translation similarity played during training.

Additionally, it may be that learners with particular cognitive characteristics may be better able to take advantage of simultaneous presentation of multiple translation pairs without additional direction. For example, as noted above, we did not provide learners with specific directions regarding how they should allocate their time. This could have overwhelmed learners who have a tendency to become distracted. Including measures in future research such as the Stroop (Stroop, 1935) or working memory task (e.g. Turner and Engle, 1989) would permit us to examine whether there are particular cognitive skills or characteristics (e.g. better ability to ignore task-irrelevant information, higher working memory) that go along with an ability to benefit from certain training manipulations (e.g. Martin and Ellis, 2012; Terrazas, 2018).

As expected, performance for consecutively-trained words was not different from translation-unambiguous words, replicating the findings of Degani et al. (2014). This observation is important given prior observations of a translation-ambiguity disadvantage (e.g. Basnight-Brown et al., 2020; Degani and Tokowicz, 2010; Laxén and Lavaur, 2010). We also observed that performance got faster from Session 2 to Session 3, and that performance was faster overall for words with more related translations (e.g. Bracken et al., 2017). We found additional results that are consistent with past research as well, including faster and more accurate responses for shorter L2 words and more concrete words (e.g. de Groot, 1992b).

A novel finding in this study is that TSV affected performance in response time (RT) and accuracy in a translation production task. This variable was previously shown to affect both accuracy and RT performance in translation recognition, so this serves as a generalization of this variable to production. The pattern of results across the two studies are very similar; higher TSV is associated with higher accuracy and faster RT in both studies, although it affected the consecutive condition to a lesser degree in accuracy in the present study. It is interesting that TSV interacted with the training manipulation in accuracy but there was a main effect for reaction time. This is most likely because translation production is a more difficult task for beginning learners, and therefore we see effects of our training manipulating most easily in the accuracy measure. It is also important to note that in previous research on this topic, TSV outperformed a dichotomous measure that separated words with translations that were for separate meanings of the word (Kiefer = pine/jaw) vs. words with two translations that meant roughly the same thing (Tüte = bag, sack) (Bracken et al., 2017). This finding, along with other details from that study, demonstrate that the psychological reality of translation similarity is continuous rather than dichotomous.

Footnotes

Appendix

Appendix 2.

Training and testing stimuli: Translation-ambiguous items.

German word	English translation 1	English translation 2
Bedarf	need	permission
Besitzung	ownership	occupation
Boden	floor	ground
Ergebnis	result	answer
Folge	episode	result
Friede	peace	joy
Gegner	opponent	enemy
Geschichte	history	story
Gnade	mercy	grace
Herbst	autumn	fall
Kiefer	jaw	pine
Kiste	box	chest
Meer	sea	ocean
Meldung	message	report
Reifen	tire	wheel
Rohr	pipe	tube
Schlägerei	fight	brawl
Schuld	guilt	fault
Tat	deed	act
Tütte	bag	sack
Verabredung	date	appointment
Veranstaltung	event	production
Versuch	attempt	try
Zahl	number	count

Acknowledgements

The authors thank members of the PLUM Lab at the University of Pittsburgh for their assistance with this study. We thank Amy Crosson, Victor Ferreira, and Tessa Warren for input on the project

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: We thank the University of Pittsburgh Honors College for funding.

ORCID iD

Natasha Tokowicz

Notes

References

Appleton-Knapp

Bjork

Wickens

(2005) Examining the spacing effect in advertising: Encoding variability, retrieval processes, and their interaction. Journal of Consumer Research 32: 266–76.

Balota

Yap

Hutchison

, et al (2007) The English lexicon project. Behavior Research Methods 39: 445–59.

Barr

Levy

Scheepers

, et al (2013) Random effects structure for confirmatory hypothesis testing: Keep it maximal. Journal of Memory and Language 68: 255–78.

Basnight-Brown

Altarriba

(2014) Number of translation differences in Spanish and Chinese bilinguals. In: Cooper

Ratele

(eds) Psychology serving humanity: Proceedings of the 30th International Congress of Psychology: Volume 2: Western Psychology. Abington: Psychology Press, pp. 240–51.

Basnight-Brown

Altarriba

(2016) Multiple translations in bilingual memory: Processing differences across concrete, abstract, and emotion words. Journal of Psycholinguistic Research 45: 1219–45.

Basnight-Brown

Kazanas

Altarriba

(2020) Translation ambiguity in Mandarin–English bilinguals: Translation production differences in concrete, abstract, and emotion words. Linguistic Approaches to Bilingualism 10: 559–86.

Belsley

Kuh

Welsch

(1980) Regression diagnostics: Identifying influential data and sources of collinearity. New York: John Wiley.

Boada

Sánchez-Casas

Gavilán

García-Albea

Tokowicz

(2013) Effect of multiple translations and cognate status on translation recognition performance of balanced bilinguals. Bilingualism: Language and Cognition 16: 183–97.

Bracken

Degani

Eddington

Tokowicz

(2017) Translation semantic variability: How semantic relatedness affects learning of translation-ambiguous words. Bilingualism: Language and Cognition 7: 783–94.

10.

Brysbaert

New

(2009) Moving beyond Kučera and Francis: A critical evaluation of current word frequency norms and the introduction of a new and improved word frequency measure for American English. Behavior Research Methods 41: 977–90.

11.

Brysbaert

Warriner

Kuperman

(2014) Concreteness ratings for 40 thousand generally known English word lemmas. Behavior Research Methods 46: 904–11.

12.

Congdon

Novack

Brooks

, et al (2017) Better together: Simultaneous presentation of speech and gesture in math instruction supports generalization and retention. Learning and Instruction 50: 65–74.

13.

de Groot

AMB

(1992a) Bilingual lexical representation: A closer look at conceptual representations. In: Frost

Katz

(eds) Orthography, phonology, morphology, and meaning. Amsterdam: Elsevier, pp. 389–412.

14.

de Groot

AMB

(1992b) Determinants of word translation. Journal of Experimental Psychology: Learning, Memory, and Cognition 18: 1001–18.

15.

de Groot

AMB

Dannenburg

van Hell

(1994) Forward and backward word translation by bilinguals. Journal of Memory and Language 33: 600–29.

16.

Degani

Tokowicz

(2010) Ambiguous words are harder to learn. Bilingualism: Language and Cognition 13: 299–314.

17.

Degani

Tseng

Tokowicz

(2014) Together or apart? Learning of ambiguous words. Bilingualism: Language and Cognition 17: 749–65.

18.

Dumay

Gaskell

(2007) Sleep-associated changes in the mental representation of spoken words. Psychological Science 18: 35–39.

19.

Eddington

Tokowicz

(2013) Examining English–German translation ambiguity using primed translation recognition. Bilingualism: Language and Cognition 16: 442–57.

20.

Eddington

Degani

Tokowicz

(2022) English and German translation norms: The role of proficiency in translation ambiguity. Manuscript in revision.

21.

Elgort

Warren

(2014) L2 vocabulary learning from reading: Explicit and tacit lexical knowledge and the role of learner and item variables. Language Learning 64: 365–414.

22.

Glover

Corkill

(1987) Influence of paraphrased repetitions on the spacing effect. Journal of Educational Psychology 79: 198–99.

23.

Karpicke

Roediger

HLI

(2008) The critical importance of retrieval for learning. Science 319: 966–68.

24.

Kaushanskaya

Marian

(2009) The bilingual advantage in novel word learning. Psychonomic Bulletin and Review 16: 705–10.

25.

Klein

Murphy

(2001) The representation of polysemous words. Journal of Memory and Language 45: 259–82.

26.

Kuznetsova

Brockhoff

Christensen

(2017) lmerTest package: Tests in linear mixed effects models. Journal of statistical software 82: 1–26.

27.

Laxén

Lavaur

J-M

(2010) The role of semantics in translation recognition: Effects of number of translations, dominance of translations and semantic relatedness of multiple translations. Bilingualism: Language and Cognition 13: 157–83.

28.

Liu

Massaro

Chen

, et al (2007) Using visual speech for training Chinese pronunciation: An in-vivo experiment. In: Speech and Language Technology in Education (SLaTE). Farmington, PA, USA, 1–3 October 2007, pp. 29–32. Farmington, PA: SLaTE.

29.

Loess

Duncan

(1952) Human discrimination learning with simultaneous and successive presentation of stimuli. Journal of Experimental Psychology 44: 215–21.

30.

Martin

Ellis

(2012) The roles of phonological short-term memory and working memory in L2 grammar and vocabulary learning. Studies in Second Language Acquisition 34: 379–413.

31.

Prior

Kroll

MacWhinney

(2013) Translation ambiguity but not word class predicts translation performance. Bilingualism: Language and Cognition 16: 458–74.

32.

Prior

MacWhinney

Kroll

(2007) Translation norms for English and Spanish: The role of lexical variables, word class, and L2 proficiency in negotiating translation ambiguity. Behavior Research Methods 39: 1029–38.

33.

R Studio Team (2016) RStudio: Integrated development for R. Boston, MA: RStudio, Inc.

34.

Schweiter

Prior

(2020) Translation ambiguity. In: Heredia

Cieślicka

(eds) Bilingual lexical ambiguity resolution. Cornwall: Cambridge University Press, pp. 96–125.

35.

Stroop

(1935) Studies of interference in serial verbal reactions. Journal of Experimental Psychology 18: 643–61.

36.

Terrazas

(2018) Alleviating the translation-ambiguity disadvantage: Using a placeholder to signal an upcoming translation. Unpublished master’s thesis, University of Pittsburgh, Pittsburgh, PA, USA.

37.

Tokowicz

(2014) Translation ambiguity affects language processing, learning, and representation. In: Miller

Martin

Eddington

, and et al. (ed.) Selected Proceedings of 2012 Second Language Research Forum: Building bridges between disciplines. Somerville, MA: Cascadilla Press, 170–80.

38.

Tokowicz

Kroll

(2007) Number of meanings and concreteness: Consequences of ambiguity within and across languages. Language and Cognitive Processes 22: 727–79.

39.

Tokowicz

Michael

Kroll

(2004) The roles of study-abroad experience and working-memory capacity in the types of errors made during transition. Bilingualism: Language and Cognition 7: 255–72.

40.

Tokowicz

Kroll

de Groot

AMB

van Hell

(2002) Number-of-translation norms for Dutch–English translation pairs: A new tool for examining language production. Behavior Research Methods, Instruments, and Computers 34: 435–51.

41.

Tseng

Chang

L-Y

Tokowicz

(2014) Translation ambiguity between English and Mandarin Chinese: The role of proficiency and word characteristics. In: Schwieter

Ferreira

(eds) The development of translation competence: Theories and methodologies from psycholinguistics and cognitive science. Newcastle upon Tyne: Cambridge Scholars Publishing, pp. 107–65.

42.

Turner

Engle

(1989) Is working memory capacity task dependent? Journal of Memory and Language 28: 127–54.

43.

van Hell

de Groot

AMB

(1998) Conceptual representation in bilingual memory: Effects of concreteness and cognate status in word association. Bilingualism: Language and Cognition 1: 193–211.

44.

Walker

Meyer

(1980) Integrating different types of information in text. Journal of Verbal Learning and Verbal Behavior 19: 263–75.

45.

Webb

Chang

(2015) Second language vocabulary learning through extensive reading with audio support: How do frequency and distribution of occurrence affect learning? Language Teaching Research 19: 667–86.

46.

Wen

van Heuven

WJB

(2017) Chinese translation norms for 1,429 English words. Behavior Research Methods 49: 1006–19.