Comprehension exposures to words in sentence contexts impact spoken word production

Abstract

Comprehension or production of isolated words and production of words embedded in sentence contexts facilitated later production in previous research. The present study examined the extent to which contextualized comprehension exposures would impact later production. Two repetition priming experiments were conducted with Spanish–English bilingual participants. In Experiment 1 (N = 112), all encoding stimuli were presented visually, and in Experiment 2 (N = 112), all encoding stimuli were presented auditorily. After reading/listening or translating isolated words or words embedded in sentences at encoding, pictures corresponding to each target word were named aloud. Repetition priming relative to new items was measured in RT and accuracy. Relative to isolated encoding, sentence encoding reduced RT priming but not accuracy priming. In reading/listening encoding conditions, both isolated and embedded words elicited accuracy priming in picture naming, but only isolated words elicited RT priming. In translation encoding conditions, repetition priming effects in RT (but not accuracy) were stronger for lower-frequency words and with lower proficiency in the picture-naming response language. RT priming was strongest when the translation response at encoding was produced in the same language as final picture naming. In contrast, accuracy priming was strongest when the translation stimulus at encoding was comprehended in the same language as final picture naming. Thus, comprehension at encoding increased the rate of successful retrieval, whereas production at encoding speeded later production. Practice of comprehension may serve to gradually move less well-learned words from receptive to productive vocabulary.

Introduction

Our everyday conversations are more than a simple exchange of information between family, friends, and colleagues; what is not always apparent to us is that language learning also takes place. The language that is experienced in everyday settings comes in the form of sentences, and comprehension is focused on the messages conveyed rather than individual words. Despite the contextualized nature of linguistic input in reading and listening, prior research on the effects of exposures to words using repetition-priming methodology has focused predominantly on processing of isolated words. The present study begins to connect these two phenomena by examining how reading, listening to, or translating sentences impacts later word production. We use a repetition priming protocol to investigate whether and how these contextualized word exposures facilitate the production of words several minutes later, how this facilitation varies with word frequency and language proficiency, and whether patterns of performance generalize across presentation modalities.

Repetition priming

Repetition priming is an item-specific change in speed, accuracy, or bias based on previous experience (Gabrieli, 1998). For example, both picture naming and translation are faster for repeated items, and these effects are durable across delays of several days or more (Cave, 1997; Francis & Sáenz, 2007; Mitchell & Brown, 1988; Wiggs et al., 2006), indicating sustained learning (Francis, 2014). Evidence that priming in picture naming lasts for several days even in patients with global amnesia (Cave & Squire, 1992) indicates that it does not require support from explicit memory and instead represents an implicit, nonhippocampal form of memory (e.g., Gabrieli, 1998).

Processes and repetition priming in picture naming and word translation

Picture naming requires access to the concept before production of the appropriate name in both monolinguals (Durso & Johnson, 1979; Potter & Faulconer, 1975; Smith & Magee, 1980) and bilinguals (Chen & Leung, 1989; Francis et al., 2003; Potter et al., 1984). Thus, object identification and word production (word retrieval and articulation) are the two major sets of processes required for picture naming. Response times are faster when naming repeated pictures (e.g., Bartram, 1974; Durso & Johnson, 1979). In picture naming, both object identification (e.g., Carroll et al., 1985; Lachman et al., 1980) and word production processes (e.g., Lee & Williams, 2001; Monsell et al., 1992) can be speeded with selective practice and make independent priming contributions (Francis et al., 2003; Francis et al., 2008). In bilinguals, repetition priming in the word production component of picture naming is stronger in the less proficient language (e.g., Francis et al., 2003; Francis et al., 2008).

In bilinguals who learn both languages at an early age, word translation is accomplished by comprehending, or accessing the concept of, a stimulus word in one language and production of its translation equivalent in the other language (e.g., De Groot, Dannenburg, & Van Hell, 1994; De Groot & Poot, 1997; Duyck & Brysbaert, 2004; Francis et al., 2003). Word translation is faster for words that have been translated recently relative to words that have not been translated recently, especially when responses are given in the less proficient language (Francis, Camacho, & Lara, 2014a; Francis et al., 2011; Francis & Gallard, 2005; Francis & Sáenz, 2007; Francis, Tokowicz, & Kroll, 2014b). Both comprehension and production processes in word translation can be speeded with selective practice and make independent priming contributions (Francis et al., 2014a; Francis et al., 2011; Francis & Gallard, 2005). Repetition priming effects in the production component of word translation are stronger when responses are given in the less proficient language (Francis et al., 2011; Francis & Gallard, 2005; Sholl et al., 1995), whereas priming effects in the comprehension component are larger when stimuli are presented in the less proficient language (Francis et al., 2014a; Francis et al., 2011; Francis & Gallard, 2005).

Of course, comprehension and production require multiple component processes to execute. We assumed a model of word production that requires access to both modality-general word representations, or lemmas, and modality-specific phonological representations prior to assembling individual phonetic units for overt articulation (e.g., Dell & O’Seaghdha, 1992; Levelt et al., 1999; Roelofs, 2020). We assumed a parallel approach to word comprehension in which perception of either a visual or auditory word stimulus results in access to orthographic or phonological representations, respectively, before access to modality-general lemmas and concepts (e.g., Dijkstra & Van Heuven, 2002; Green, 1998; Roelofs, 2020).

Picture naming, translation, and priming effects in picture naming and translation are sensitive to word frequency and participant proficiency. In both picture naming and translation, production responses are slower for low-frequency words than for high-frequency words (e.g., De Groot, 1992; Gollan et al., 2008; Griffin & Bock, 1998) and slower for words in the less proficient language than in the more proficient language (Kroll & Stewart, 1994; Potter et al., 1984; Sholl et al., 1995). According to the frequency-lag hypothesis (Gollan et al., 2008; Gollan et al., 2011), these effects arise because both low-frequency words and words in a less proficient language have weaker links to their corresponding concepts. Because learning episodes have diminishing returns, the benefits of repetition tend to be greater when words are less well learned. Specifically, stronger priming effects are observed for low-frequency words (Griffin & Bock, 1998; Wheeldon & Monsell, 1992) and words in a less proficient language (Francis et al., 2003; Francis et al., 2008; Francis & Sáenz, 2007; Sholl et al., 1995), even when words are embedded in sentences at encoding (Francis et al., 2014a).

Transfer-appropriate processing explanations

The patterns of repetition-priming effects described can be accommodated by the principle of transfer-appropriate processing (Morris et al., 1977; Roediger & Blaxton, 1987), the idea that common processes engaged during encoding and retrieval are a critical factor in memory performance. In a repetition priming context, the idea is that the basis of priming is the processes shared by the encoding and test tasks, with more shared processes leading to larger priming effects. It is less clear how this principle might apply when a word is comprehended at encoding before producing it at test, because the flow of information appears to be in opposite directions.

For example, picture naming is facilitated by reading words silently or aloud (Barry et al., 2001; Brown et al., 1991; Durso & Johnson, 1979), listening to words (Brown et al., 1991), and lexical decision (Van Assche et al., 2016). Also, in bilinguals, translation from Spanish to English facilitated later picture naming in Spanish, and translation from English to Spanish facilitated later picture naming in English (Francis et al., 2008). The reading, listening, and lexical decision tasks often result in but do not require conceptual access (e.g., Coltheart et al., 2001; Dell et al., 2007), whereas translation does require conceptual access (e.g., De Groot et al., 1994). Mere conceptual repetition is insufficient to account for the effects, because concept-level repetition alone does not facilitate picture naming (Monsell et al., 1992). Therefore, these effects do not initially appear to be transfer appropriate.

When considering the processes of comprehension and production more closely, the concept is not necessarily the only shared representation. Models of lexical processing often include modality-general lemma representations, and the lemmas are assumed to be the same for comprehension and production (e.g., Green, 1998; Hanley & Nickels, 2009; Levelt et al., 1999; Roelofs, 2020). Recent evidence of transfer from lexical decision to picture naming was interpreted as further evidence that lemmas are shared by comprehension and production modalities (Van Assche et al., 2016). Similarly, both reading and listening to words silently or under articulatory suppression facilitate later picture naming (Tsuboi et al., 2021). Thus, production requires retrieving the lemma from the concept, and comprehension requires retrieving the concept from the lemma, and a possible locus of priming from comprehension to production would be in the links between concepts and corresponding lemmas. In bilinguals, lemmas are assumed to be language-specific, whereas concepts are assumed to be shared across languages (Dijkstra & Van Heuven, 2002; Green, 1998; see Francis, 1999, 2005, for reviews).

Repetition Priming for Words Comprehended or Produced in Context

Findings indicating that exposures to individual words result in sustained learning suggest the possibility that repetition priming in comprehension and production plays an important role in language acquisition. If so, the effects observed for exposures to isolated words would extend to words that are embedded in sentences at encoding. In a variety of repetition-priming paradigms, embedding words in larger contexts at encoding has reduced or eliminated repetition-priming effects (MacLeod, 1989; Oliphant, 1983; Smith, 1991; Speelman et al., 2002). Most pertinent to the present study is that, relative to translating isolated words, translating words embedded in sentences at encoding elicited substantial but weaker repetition-priming effects in both picture-naming and translation test tasks (Francis et al., 2014a).

When words were read in sentence contexts at encoding and tested in isolation, repetition priming was more robust when low frequency words were used (Nicolas, 1996), when reading was made more difficult (Nicolas, 1998), or when testing participants who had low reading proficiency (Bourassa et al., 1998). These findings from test tasks such as word fragment completion, noun association, and reading words aloud suggest that transfer from contextualized words to picture naming will also be stronger for words with lower frequency and for participants with lower proficiency. Indeed, this pattern emerged when participants translated sentences at encoding and named pictures at test with the same response language (Francis et al., 2014a). It is unknown whether the comprehension of words presented in sentence contexts at encoding will facilitate their later production or whether any such facilitation is moderated by prior experience, as indicated by word frequency and language proficiency.

The present study

The main purpose of the present study was to examine whether and how well comprehension exposures to sentences would facilitate later production of key words and to better understand the factors that influence such facilitation. We reasoned that words comprehended in sentence contexts at encoding would exhibit weaker priming effects in later production than words encoded in isolation, because the tasks are less similar. However, the degree of transfer from embedded words to production of isolated words at test was expected to be greater when the final production task was more difficult, either because the words were of lower frequency or because the speaker was less proficient in the task language. In previous research using different test tasks, transfer from sentence-embedded words to isolated words was greater for less skilled readers, more difficult reading tasks, and lower-frequency words (Bourassa et el., 1998; Nicolas, 1996, 1998). Based on the frequency-lag hypothesis, we predicted that for both isolated words and words embedded in sentences, frequency and proficiency would exhibit parallel effects on repetition priming and that these effects would interact.

A second set of questions was about the degree to which comprehension exposures of two types would facilitate later production, relative to production exposures. To the extent that comprehension facilitates production, facilitation in repeated production can be attributed to speeded retrieval of a lemma common to comprehension and production. We hypothesized that comprehension at encoding would elicit facilitation in later production but not as much as production at encoding. Therefore, encoding tasks that require word production in the picture-naming response language would elicit greater facilitation in production. Specifically, translating to the eventual picture-naming response language, nontarget–target translation, was expected to elicit stronger priming effects than translating from the picture-naming language, target–nontarget translation. To the extent that target–nontarget translation facilitates later production, speeded lemma retrieval contributes to repetition priming in repeated production.

We also hypothesized that a more active and conceptual comprehension task would elicit more priming in later production than would a more passive comprehension task, because more shared representations would be accessed, and more shared processes would be engaged. Target–nontarget translation was considered to be a strong comprehension encoding task, because it requires access to both language-specific lemma and language-general conceptual representations (e.g., De Groot et al., 1994). Reading or listening to target words silently was considered to be a weak comprehension encoding task, because simply reading or listening often results in but does not require conceptual access (e.g., Coltheart et al., 2001; Dell et al., 2007). We predicted larger priming effects in final picture naming for strong than for weak comprehension encoding conditions. Another reason to include two encoding conditions meant to practice comprehension is that reading and listening to sentences is more like everyday language use than translation, but translation better ensures access to the shared lemma and conceptual representations.

Third, we examined patterns of repetition priming in accuracy to determine the impact of comprehension and production exposures on the probability of successful retrieval. If comprehension exposures increase the likelihood that a word will be successfully retrieved for later production, these exposures could serve as a mechanism for the gradual incorporation of words in receptive vocabulary into productive vocabulary. Previous research indicates that comprehension encoding exposures increase accuracy in later production (Brown et al., 1991; Francis et al., 2008; Tsuboi et al., 2021), but there has been little systematic investigation. There have been no comparisons of the effects of stronger and weaker comprehension exposures and no well-powered comparisons of the effects of comprehension and production exposures on accuracy priming. Also, it remains unknown how sentence contexts or language proficiency might impact how comprehension exposures improve later production accuracy. Word frequency did not impact accuracy priming in production when comprehension exposures were weak (reading or listening to isolated words; Tsuboi et al., 2021). In most reports of repetition priming in picture naming, error rates are either not reported or not analyzed in detail, in part because error rates were very low in studies with monolingual participants naming sets of relatively high-frequency words. With the early studies that used tachistoscopic presentation, fewer trials could be executed in a session, so accuracy was maximized to avoid losing trials for analysis. In contrast, with bilingual participants and word sets with lower average frequency, error rates are substantially higher, and with efficient trial execution, hundreds of pictures can be named in a single session. Thus, in the present study, we compared and contrasted patterns of accuracy priming (i.e., error rate reductions) with those of RT priming

Finally, we investigated whether patterns of effects would generalize across visual and auditory encoding modalities. Experiment 1 used visual stimulus presentation, with silent reading or reading with a spoken translation response. Experiment 2, conducted concurrently, used auditory stimulus presentation, with silent listening or listening with a spoken translation response. Based on the preceding logic, there was not a strong theoretical reason to expect differences between visual and auditory presentation conditions. However, in a previous study, while visual comprehension of words was facilitated with repetition, auditory comprehension of words was not (Francis et al., 2014a), presumably because the processes were so overlearned. However, it is still possible that auditory comprehension of words and sentences could facilitate the processes of spoken production, which is not overlearned. Also, we wanted to replicate the same set of tests in both visual and auditory modality to determine whether reading sentences and listening to sentences, whether passively or for translation, affect later production in a similar manner.

Experiment 1

At encoding, individual words and sentences with target words embedded were presented visually in English or Spanish, and bilingual participants were asked to read them silently or translate them aloud. In a final picture-naming test, participants named pictures corresponding to target words encountered at encoding and new target words not encountered at encoding.

Method

Power and sample size

Effects were to be tested in two completely within-subjects designs. Because power/sample size estimation for mixed-effects regression are not well established, sample size was initially estimated based on a repeated-measures analysis of variance (ANOVA). Although 34 participants would have yielded 80% power to detect medium-sized effects, constraints for complete counterbalancing required sample sizes of 112 for Experiments 1 and 2. These sample sizes allow 80% power to detect effect sizes as small as d = .27. In each experiment, with 112 participants and 20 items per condition, there were 2,240 trials/condition, far exceeding the recommendation of 1,600 to detect small effects in linear mixed-effects regression with crossed random factors (Brysbaert & Stevens, 2018).

Participants

Participants were 112 Spanish–English bilinguals recruited from the University of Texas at El Paso, a southwestern university bordering Juarez, Chihuahua, Mexico. Participants were compensated with either research credit for their Psychology course or payment of $20 for a 2-hour experimental session. Participants self-identified as bilingual and their proficiency was confirmed in both English and Spanish, using the Woodcock-Muñoz Language Survey–Revised (WMLS-R; Woodcock et al., 2005). The WMLS-R is a standardized objective assessment that includes four subtests: picture vocabulary, verbal analogies, letter-word identification, and dictation. These component scores are used to compute a composite measure of broad ability in English and in Spanish. In the present study, participants had to obtain an age equivalency score of at least 10 on the broad ability measure in both languages; the language with the higher age-equivalency score was considered to be the dominant language. Based on these scores, 56 participants were classified as English dominant, and the other 56 as Spanish dominant. The median age was 20, and 96% of participants reported Hispanic ethnicity. Table 1 provides additional participant information.

Table 1 Participant characteristics

Design

The experiment had a 3 (encoding task) × 2 (encoding context) × 2 (test language) within-subject design with a new-item control condition in each language. The encoding tasks were target reading, target–nontarget translation, or nontarget–target translation. The encoding context conditions were isolated words and words embedded in sentences. Half of the pictures at test were named in English and half were named in Spanish. For final English naming, examples for each encoding condition are given in Table 2. The dependent variables were picture-naming RT and accuracy.

Table 2 Encoding tasks for final English picture naming in Experiments 1 and 2

Stimuli

The stimuli were 280 single words and 140 short sentences that each contained two target words (see Appendix 1). The picture stimuli were 280 black-and-white normed line drawings (selected primarily from Abbate, 1984; Snodgrass & Vanderwart, 1980). The median frequency of the words in English was 13.5 (Brysbaert & New, 2009) and the median frequency in Spanish was 11.6 (Cuetos et al., 2011). (Note that a small proportion of words were used locally, but not in Spain and were replaced with their counterparts to estimate frequency.) The mean word length for English was 5.5 (SD = 2.0) and 6.3 (SD = 1.8) for Spanish. The 280 items were randomly divided into 14 sets with 20 items in each set. The sets assigned to each condition were counterbalanced across participants using a Latin square to control for specific-item effects.

Each sentence included exactly two critical words, which could appear in any position except first. Many different verbs were used, and when additional nouns were needed, we added pronouns, words for people (e.g., grandmother, student), or other words that were not target words. We did not control for predictability of the nouns from the verb, specific syntactic structures, thematic roles, or factors like phrasal stress or focus, because (a) every word and sentence structure appeared in every condition equally often, and (b) we were interested the basic isolated word versus sentence manipulation, not properties of particular words or sentences.

Apparatus

The experiment was programmed using PsyScope X software (Cohen et al., 1993). Visual stimuli were presented on an iMac computer monitor. RTs were registered using a microphone attached to an ioLab Systems button box. Participants’ verbal responses were recorded using a Sony digital voice recorder to allow verification of responses after the experimental session had ended.

Procedure

Participants were tested individually by a bilingual experimenter in a 2-hour session. After informed consent, the experimenter administered the WMLS-R language assessments in both English and Spanish, and the participant completed language background and demographic questionnaires.

The computerized experiment had an encoding phase and a test phase. During the encoding phase, participants completed eight different encoding tasks: reading isolated words or sentences in English, reading words or sentences in Spanish, translating English words or sentences to Spanish, and translating Spanish words or sentences to English. Reading was silent and required no participant response (because the target word is also not produced in target–nontarget translation). In the translation tasks, isolated words were presented in one language (e.g., apple) and the participants responded into the microphone with their translation equivalent (e.g., manzana); a vocal response was required for the next word to be presented. Sentences were presented in a similar fashion, and the participant responded aloud and then pushed a button to trigger the presentation of the next sentence. Each block of encoding trials began with three practice trials followed by trials involving the critical words. The reading blocks contained 20 words each, and the translation blocks contained 40 words each. Thus 240 words were presented, leaving the remaining 40 words to be presented as new items in the test phase. The order of tasks was counterbalanced across participants. For a description of the encoding tasks, see Table 2.

In the test phase, participants were instructed to name aloud pictures that were displayed one at a time on the computer monitor as quickly and accurately as possible. All 280 pictures were presented. Each picture-naming block began with an instruction indicating the appropriate response language, and three practice trials were completed before each of the four blocks of 70 experimental trials. The participant named 140 pictures in English, and 140 pictures in Spanish, with the order of languages counterbalanced across participants. In both encoding and test blocks, the experimenter verified correct responses and noted unexpected responses and voice relay misfires on a worksheet containing the expected responses.

Results

Data processing

Analysis focused on valid trials in the picture-naming test phase. All items and trials were included in the accuracy analysis except for one that had to be excluded for all participants as an extreme frequency outlier. The RT analysis focused on trials with correct responses and valid timing. From the 280 picture-naming trials at test, 15% were excluded due to incorrect naming responses (including unexpected and “don’t know” responses), and 1.2% were excluded because of voice-relay misfires. Spoiled trials (7.9%) were those that had an unexpected response at encoding (6.4%), those with invalid timing at encoding (0.4%), and those given as error responses to other items on an earlier trial (1.1%). Finally, trials with RTs greater than 5,000 ms, less than 200 ms, or more than two standard deviations above or below a participant’s condition mean (5.1%) were excluded as outliers. Thus, 70.9% of the items were retained for the analyses of RT priming.

Approach to analysis

RTs were analyzed using linear mixed-effects and accuracy using logistic mixed-effects regression models within the lme4 package of R. When included in a model, encoding task, context, repetition status, word frequency, and language proficiency were treated as within-subjects fixed factors. Fixed factor structures were never reduced; all possible interactions were retained in the final regression model. Participants and items were treated as random factors. Models included random intercepts for participants and items and random slopes for categorical fixed factors across participants and items if the model converged.

Objective language proficiency scores in English and Spanish and word frequencies for English and Spanish words were treated as continuous variables. For the purposes of analysis, English proficiency scores were entered for English naming trials and Spanish proficiency scores were entered for Spanish naming trials. Here, instead of using the age-equivalency scores shown in Table 1, we used W scores because of their better psychometric properties (Woodcock et al., 2005). Similarly, English word frequencies were entered for English naming trials and Spanish frequencies for Spanish naming trials, and frequencies were log transformed. These continuous predictors were standardized (M = 0, SD = 1) across the full set of scores. All categorical design variables were treated in a binary manner and centered using deviation coding (−.5, +.5) to make interactions among the fixed effects orthogonal.

Analyses required two steps, because the design was not fully factorial. First, an analysis of the full data set examined the effects of word frequency and participant language proficiency, the effects of repetition, and the interactions of word frequency and proficiency with repetition. Second, in repeated-item trials, the effects of encoding context and encoding task were examined along with their interactions with frequency and proficiency. Two separate mixed-effects regression models were used, one to compare the target reading and target–nontarget translation encoding conditions and one to compare the target–nontarget and nontarget–target translation conditions. Complete reports of fixed and random effects are given in Appendix 2.

Response time analyses

Mean RTs are given in Table 4. Repetition priming scores were obtained for each participant by subtracting RTs in repeated conditions from those of corresponding new-item conditions and are illustrated in Fig. 1. RTs for new and repeated conditions are plotted as a function of frequency in Fig. 2 and as a function of proficiency in Fig. 3.

Fig. 1

Repetition priming in RT and accuracy in Experiment 1 as a function of encoding task and context. Error bars indicate standard error of the mean (by participants). Weak comprehension = reading in target language; strong comprehension = translation with comprehension in target language; production = translation with production in target language

Fig. 2

Picture-naming RT in Experiments 1 and 2 as a function of encoding context, encoding task, and word frequency. Frequency is the log₁₀ of frequencies per million in SUBTLEX-US or SUBTLEX-ESP, consistent with the language of the naming task. TR = translation

Fig. 3

Picture-naming RT in Experiments 1 and 2 as a function of encoding context, encoding task, and proficiency in the response language. Proficiency scores are W scores from WMLS-R Broad Ability composite in the language of the naming task. TR = translation

Repeated items versus new items

A linear mixed-effects regression analysis was conducted with repetition status (repeated vs. new), word frequency, and participant proficiency as fixed factors, random intercepts for participants and items, and random slopes for repetition status across participants (see Table 5 in Appendix 2). Picture naming was slower for lower frequency words, b = −77.13, SE = 9.73, t = 7.929, p < .001, and for less proficient speakers, b = −129.63, SE = 6.90, t = 18.776, p < .001. These effects interacted, with stronger frequency effects for less proficient speakers, b = 11.36, SE = 4.90, t = 2.316, p = .021. The repetition priming effect was reliable overall, b = −100.79, SE = 13.71, t = 7.353, p < .001. Repetition priming effects were stronger for lower-frequency words, b = 29.17, SE = 9.83, t = 2.966, p = .003, and for less proficient speakers of the naming language, b = 49.196, SE = 11.617, t = 3.977, p < .001. The effects of word frequency and participant proficiency on repetition priming did not interact, b = 13.50, SE = 9.80, t = 1.377, p = .168. For each repeated-item condition, we conducted an analysis that included only that repeated-item condition and the new-item control condition, with repetition status, frequency, and proficiency as the fixed factors and random intercepts for participants and items and random slopes for repetition status across participants. Repetition priming was significant in every repeated condition (ps < .001) except for the sentence reading condition (p > .1; see Supplemental Materials for details).

Repeated items

To examine the effects of encoding task and encoding context on repetition priming in picture-naming RTs, data from repeated-item conditions were submitted to analyses with encoding task, encoding context, word frequency, and participant proficiency as fixed factors, random intercepts for participants and items, and random slopes for encoding task across participants (see Table 6 in Appendix 2). Because the different encoding-task and encoding-context conditions had the same new-item control conditions for comparison, differences among the RTs for these conditions can be interpreted as differences in the strength of the repetition priming effect.

The first analysis compared the repetition priming effects in naming RT elicited by the two tasks meant to practice comprehension and included only target reading and target–nontarget translation encoding conditions. Naming was faster in the target–nontarget translation than in the target reading condition, b = −79.24, SE = 11.74, t = 6.750, p < .001, indicating a larger priming effect following comprehension in translation. Naming was faster in the word than in the sentence condition, b = −26.48, SE = 9.32, t = 2.840, p = .005, indicating that priming was weaker for sentences than for isolated words. These effects did not interact (t < 1). The benefit of isolated relative to sentence-embedded word exposures was greater for less proficient speakers, b = 37.96, SE = 9.41, t = 4.036, p < .001, and this effect was stronger in target reading conditions, as indicated by a three-way interaction, b = −37.79, SE = 18.80, t = 2.010, p = .044 (see Fig. 3). No other effects involving encoding task or context were reliable (ps > .1).

The second analysis compared the repetition priming effects in naming RT elicited by comprehension and production exposures at encoding and included only the two translation encoding conditions. Naming was faster in the nontarget–target translation condition than in the target–nontarget translation condition, b = −29.19, SE = 9.82, t = 2.971, p = .004, indicating a larger priming effect when the target word was produced at encoding. Naming was faster in the word condition than in the sentence condition, b = −26.02, SE = 9.55, t = 2.725, p = .008, indicating that priming was weaker for sentences than for isolated words. The effects of encoding task and context did not interact (t < 1). Speakers with lower proficiency benefitted more from nontarget–target relative to target–nontarget translation encoding, b = 23.84, SE = 9.33, t = 2.556, p = .011, and this effect was stronger in sentence context conditions, b = −42.76, SE = 17.32, t = 2.470, p = .014 (see Fig. 3). No other effects involving encoding task or context were reliable (ps > .2).

Accuracy analyses

Mean accuracy scores are given in Table 3. Repetition priming scores were obtained by subtracting accuracy scores in repeated conditions from those of corresponding new-item conditions and are illustrated in Fig. 1.

Table 3 Mean RTs (milliseconds) and accuracy (%) for test phase picture naming for Experiment 1

Repeated items versus new items

A logistic mixed-effects regression analysis was conducted with repetition status (repeated vs. new), word frequency, and participant proficiency as fixed factors, random intercepts for participants and items, and random slopes for repetition status across participants (see Table 5 in Appendix 2). Naming was less accurate for lower frequency words, b = .548, SE = .049, z = 11.283, p < .001, and for less proficient speakers, b = .507, SE = .030, z = 16.767, p < .001. The effect of frequency was stronger for less proficient speakers, b = −.154, SE = .026, z = 5.830, p < .001. Repetition priming was reliable overall, b = .444, SE = .053, z = 8.394, p < .001, but repetition status did not enter into any interactions with frequency or proficiency (zs < 1). For each repeated-item condition, we conducted an analysis with fixed and random effects structure as for RT. The repetition priming effect in accuracy was significant in every repeated condition (ps < .005; see Supplemental Materials for details).

Repeated items

To examine the effects of encoding task and encoding context on repetition priming in picture-naming accuracy, accuracy data from repeated-item conditions were submitted to analyses with encoding task, encoding context, word frequency, and participant proficiency as fixed factors, random intercepts for participants and items, and random slopes for encoding task across participants (see Table 7 in Appendix 2). The first analysis compared repetition priming effects in naming accuracy elicited by the two comprehension encoding tasks and included only the target reading and target–nontarget translation conditions. Accuracy was higher following target–nontarget translation than following target reading, b = .534, SE = .057, z = 9.359, p < .001, indicating a larger priming effect following comprehension in translation at encoding. This effect was stronger for lower-frequency words, b = −.168, SE = .052, z = 3.208, p = .001. Context did not affect accuracy priming (z < 1). No other effects involving encoding condition or context were reliable (ps > .1), except for an unexpected four-way interaction, b = −.266, SE = .107, z = 2.496, p = .013.

The second analysis compared repetition-priming effects in naming accuracy for the two translation tasks. Accuracy was higher in the target–nontarget than in the nontarget–target translation encoding condition, b = −.314, SE = .056, z = 5.589, p < .001, indicating a larger priming effect following strong target comprehension than following target production. This effect was stronger for low-frequency words, b = .246, SE = .055, z = 4.462, p < .001. Context did not affect accuracy (z < 1), but a marginal interaction suggested that isolated word encoding confers a greater benefit relative to sentence encoding at lower proficiency levels, b = .107, SE = .057, z = 1.879, p = .060, No other effects involving encoding condition or context were reliable (zs < 1).

Discussion

When words were presented in sentence contexts at encoding, repetition priming was attenuated in RT but not in accuracy. Isolated encoding conditions were more beneficial for less proficient speakers, particularly in read-only conditions. RT priming effects were larger for production (nontarget–target translation) than for strong comprehension (target–nontarget translation) conditions and larger for strong than for weak comprehension (target reading) conditions. The benefit of production relative to comprehension encoding was greater for less proficient speakers, particularly in sentence contexts. In contrast, accuracy priming effects were larger for strong comprehension than for the other encoding tasks, particularly for lower-frequency words, and they did not differ for word and sentence encoding. Generally, repetition priming in RT (but not accuracy) was stronger with lower frequency and lower proficiency.

Experiment 2

The purpose of Experiment 2 was to investigate whether similar patterns of repetition priming in the comprehension and production processes of picture naming would persist when all word and sentence stimuli were presented in the auditory modality at encoding. Experiment 2 replicates the design of Experiment 1, with the words and sentences presented auditorily for listening and translation tasks at encoding.

Method

Participants

The participants were 112 Spanish–English bilinguals from the same population as in Experiment 1, but none had participated in Experiment 1. Participants self-identified as bilingual and their proficiency was confirmed using the WMLS-R in English and Spanish. Fifty-six participants were classified as English dominant and the other 56 as Spanish dominant. The median age was 20, and 93% of participants reported Hispanic ethnicity. Table 1 provides additional participant information.

Design, materials, apparatus, and procedure

The design was the same as in Experiment 1, except that target listening conditions replaced the target reading conditions. The same words and sentences were used, and they were recorded by a female native speaker of English and Spanish for auditory presentation. Sound files were edited using Praat software (Boersma & Weenik, 2017). The only change to the apparatus was the addition of headphones for listening to the auditory stimuli during the encoding phase. The procedure was the same as in Experiment 1, except that all words and sentences in the encoding phase were presented in the auditory modality, so participants listened to or translated auditory words and sentences.

Results

Data processing and approach to analysis

Data were processed in the same manner as for Experiment 1. All items and trials were included in the accuracy analysis except for three items that had to be excluded for all participants, one because it was an extreme frequency outlier and two because of a sound file error. From the 280 picture-naming trials at test, 15% were excluded due to incorrect naming responses, and 3% were excluded because of voice-relay misfires. Another 7% of trials were excluded as spoiled (due do unexpected response at encoding, invalid timing at encoding, or being given as error responses to other items on an earlier trial). Finally, trials with RTs greater than 5000-ms, less than 200-ms or more than two standard deviations above or below a participant’s condition mean (4.7%) were excluded as outliers. Thus, 69% of the items were retained for the analyses of RT priming. The approach to analysis was the same as in Experiment 1, and full reports of fixed and random effects are given in Appendix 2.

Response time analyses

Mean RTs are given in Table 4. Repetition priming scores were obtained by subtracting RTs in repeated conditions from those of corresponding new-item conditions and are illustrated in Fig. 4. RTs for new and repeated conditions are plotted as a function of frequency in Fig. 2 and as a function of proficiency in Fig. 3.

Table 4 Mean RTs (milliseconds) and accuracy (%) for test phase picture naming in Experiment 2

Fig. 4

Repetition priming in RT and accuracy in Experiment 2 as a function of encoding task and context. Error bars indicate standard error of the mean (by participants). Weak comprehension = listening in target language; strong comprehension = translation with comprehension in target language; production = translation with production in target language

Repeated items versus new items

Priming scores in RTs are illustrated in Fig. 4. A linear mixed-effects regression analysis was conducted with repetition status (repeated vs. new), word frequency, and participant proficiency as fixed factors, random intercepts for participants and items, and random slopes for repetition status across participants (see Table 8 in Appendix 2). Picture naming was slower for lower frequency words, b = −69.81, SE = 9.02, t = 7.743, p < .001, and for less proficient speakers, b = −63.97, SE = 6.52, t = 9.813, p < .001. A marginal interaction was in the direction of stronger frequency effects for participants with lower proficiency, b = 8.52, SE = 4.50, t = 1.894, p = .058. The repetition priming effect was reliable overall, b = −70.39, SE =10.65 t = 6.609, p < .001. Repetition priming was stronger for lower-frequency words, b = 27.53, SE = 8.76, t = 3.142, p = .002, and marginally stronger for less proficient speakers, b = 17.48, SE = 9.80, t = 1.784, p = .075. These effects did not interact (t < 1). Analyses using the same models as in Experiment 1 showed that the repetition priming effect was significant in all translation conditions (ps < .005) and marginal in the word listening condition (p < .1) but not in the sentence listening condition (p > .2; see Supplemental Materials for details.)

Repeated items

Data from repeated-item conditions were submitted to analyses with encoding task, encoding context, word frequency, and participant proficiency as fixed factors, random intercepts for participants and items, and random slopes for encoding task across participants (see Table 9 in Appendix 2).

The first analysis compared the repetition priming effects in naming RT elicited by the two comprehension encoding tasks and included only target listening and target–nontarget translation conditions. Naming was faster in the target–nontarget translation than in the target listening condition, b = −50.54, SE = 10.34, t = 4.889, p < .001, indicating a larger priming effect following comprehension with translation at encoding. The advantage for target–nontarget translation encoding was greater for less proficient speakers, b = 23.21, SE = 9.61, t = 2.415, p = .016. Overall, naming was not slower in the sentence than in the word condition (p > .2), but priming effects were weaker following sentence encoding in target–nontarget translation but not in target listening conditions, b = −36.42, SE = 17.08, t = 2.132, p = .033. This interaction of context and encoding task was stronger for lower-frequency words, b = 37.48, SE = 17.18, t = 2.182, p = .029 (see Fig. 2). No other effects involving encoding task or context were reliable (ps > .1).

The second analysis compared the repetition priming effects in naming RT elicited by comprehension and production exposures at encoding and included only the target–nontarget and nontarget–target translation conditions. Naming was faster in the nontarget–target translation condition than in the target–nontarget translation condition, b = −54.32, SE = 9.71, t = 5.598, p < .001, indicating a larger priming effect when the target word was produced at encoding, and this effect was stronger for lower-frequency words, b = 21.57, SE = 7.68, t = 2.809, p = .005. Naming was faster in the word condition than in the sentence condition, b = −45.19, SE = 9.95, t = 4.542, p < .001, indicating that priming was weaker for sentences than for isolated words, and this effect was stronger for low-frequency words, b = 17.85, SE = 7.70, t = 2.318, p = .020. No other effects involving encoding task or context were reliable (ps > .2).

Accuracy analyses

Mean accuracy scores are given in Table 4. Repetition priming scores were obtained by subtracting accuracy scores in repeated conditions from those of corresponding new-item conditions and are illustrated in Fig. 4.

Repeated items versus new items

A logistic mixed-effects regression analysis was conducted with repetition status (repeated vs. new), word frequency, and participant proficiency as fixed factors, random intercepts for participants and items, and random slopes for repetition status across participants (see Table 8 in Appendix 2). Naming was less accurate for lower frequency words, b = .691, SE = .055, z = 12.546, p < .001, and for less proficient speakers, b = .386, SE = .032, z = 12.219, p < .001. Less proficient speakers exhibited stronger frequency effects, b = −.109, SE = .026, z = 4.170, p < .001. Repetition priming was reliable overall, b = .262, SE = .052, z = 4.998, p < .001, but did not enter into any interactions with frequency or proficiency (zs < 1). Analyses using the same models as in Experiment 1 showed that repetition priming in accuracy was significant in all translation encoding conditions (ps < .005), but not in the listening conditions (ps > .1; see Supplemental Materials for details).

Repeated items

Accuracy data from repeated-item conditions were submitted to analyses with encoding task, encoding context, word frequency, and participant proficiency as fixed factors, random intercepts for participants and items, and random slopes for encoding task across participants (see Table 10 in Appendix 2). The first analysis compared repetition priming effects in naming accuracy elicited by the two comprehension encoding tasks and included only the target listening and target–nontarget translation conditions. Accuracy was higher for the target–nontarget translation than for the target listening conditions, b = .436, SE = .058, z = 7.466, p < .001, indicating a larger priming effect following comprehension in translation at encoding. This advantage for translation encoding was stronger for lower frequency words, b = −.111, SE = .051, z = 2.184, p = .029. Context did not affect accuracy priming (z < 1). No other effects involving encoding condition or context were reliable (ps > .2).

The second analysis compared repetition-priming effects in naming accuracy for the two translation conditions. Accuracy was higher for the target–nontarget translation than for the nontarget–target translation condition, b = −.170, SE = .053, z = 3.187, p = .001, indicating a larger priming effect following strong comprehension than following production of the target word. This effect was stronger for low-frequency words, b = .196, SE = .052, z = 3.801, p < .001. Context did not affect accuracy, b = .079, SE = .052, z = 1.497, p = .134. No other effects involving encoding condition or context were reliable (ps > .2).

Discussion

In the two translation conditions, repetition priming in RT was attenuated when words were presented in sentence contexts at encoding, but this effect was not observed in target listening conditions or in accuracy. RT priming effects were larger for production (nontarget–target translation) than for strong comprehension (target–nontarget translation) and larger for strong than for weak comprehension (target listening). The effects of both context and encoding task in RT priming were stronger for low-frequency words, and the effect of context was stronger for less proficient speakers. In contrast, accuracy priming effects were larger for target–nontarget translation than for the other encoding tasks, particularly for lower-frequency words.

General discussion

The present experiment investigated the impact of contextualized comprehension exposures on later spoken word production, with repetition priming in picture-naming RTs as the primary dependent variable. Translating visual and auditory words for production speeded later spoken picture naming, whether they were translated in isolation or in sentence contexts, consistent with previous research (Francis et al., 2003; Francis et al., 2014a; Francis et al., 2008). Translating visual and auditory words from the target to nontarget language speeded spoken picture naming, consistent with previous research (Francis et al., 2008), and this effect replicated for words embedded in written or spoken sentences. Silently reading or listening to isolated target words speeded spoken picture naming (as in Brown et al., 1991), but reading or listening to target words embedded in sentences did not. The patterns of repetition priming observed following visual and auditory presentation modality at encoding were very similar, but accuracy priming was stronger in the visual condition. Although listening for translation did not facilitate itself in a previous study (Francis et al., 2014a), perhaps because it is so well learned, listening for target–nontarget translation did facilitate picture naming in Experiment 2. In the following sections, we discuss the impact of sentence contexts on repetition priming, encoding tasks and transfer appropriate processing, and explanations for word frequency and language proficiency effects.

The impact of sentence contexts on repetition priming

When words were embedded in sentences at encoding, repetition priming in picture-naming RTs was attenuated relative to when words were encoded in isolation, consistent with previous research (Francis et al., 2014a; Levy & Kirsner, 1989; Oliphant, 1983). Two explanations for these effects in previous research cite aspects of sentence processing that might change the way that individual words are processed at encoding, thus making the encoding task less transfer-appropriate than for a test task involving the individual words. One explanation was that words in sentences are integrated into a larger conceptual framework, which elicits higher-order conceptual processing (Levy & Kirsner, 1989; MacLeod, 1989; Oliphant, 1983). The other explanation was that the sentence context does not allow the individuated and distinctive encoding of component words (Masson & MacLeod, 2000). In contrast to the attenuation of repetition priming effects in RT, embedding words in sentence contexts did not decrease repetition priming in picture-naming accuracy. The accuracy priming results therefore appear to be at odds with the preceding explanations.

These repetition priming explanations focus on how context impacts processing of individual words, with isolation being the default encoding, but maybe this thinking should be reversed. Maybe it would be more fruitful to consider how isolating words at study might cause them to be processed differently than they would be in a more natural sentence context. Including the less natural isolated-word conditions for comparison may also change the way the sentences are processed and reduce transfer for target words embedded in sentences. Here, we might expect RT priming to be more sensitive than accuracy priming to these changes in processing. Stronger support for these speculative explanations will require further research.

Encoding tasks and transfer-appropriate processing

The present results show that comprehending through reading, listening, and translation elicits learning that transfers to production several minutes later, as evidenced by speeded RT, increased accuracy, or both. Patterns of performance across the encoding tasks were consistent with expectations based on the principle of transfer-appropriate processing (Morris et al., 1977; Roediger & Blaxton, 1987). When target words were produced in nontarget–target translation at encoding, picture naming was speeded more than when they were comprehended in target–nontarget translation at encoding. This effect is transfer-appropriate, given that production encoding shares more processes than comprehension encoding with a later production episode.

While comprehension requires retrieving the concept from the lemma, production requires retrieving the lemma from the concept. Transfer from comprehension or identification tasks to production tasks is possible under transfer-appropriate processing only if two conditions are met. First, the lemmas used for comprehension and production would have to be one and the same, a contention supported by previous research (Tsuboi et al., 2021; Van Assche et al., 2016). Second, comprehension would have to involve feedback loops, interactivity, or top-down processing, which are properties of many single-language and bilingual lexical processing models (e.g., Dell & O’Seaghdha, 1992; Dijkstra & Van Heuven, 2002; Green, 1998; Shook & Marian, 2013). As explained in the introduction, priming from comprehension to production appears to occur in the associations between concepts and lemmas. Indeed, the present results provide further evidence for this contention. Specifically, the priming effect in strong comprehension (target–nontarget translation) encoding conditions was 87% and 66% of the priming effect in production (nontarget–target translation) encoding conditions in Experiments 1 and 2, respectively. These percentages constitute the minimum proportion of production priming that can be attributed to speeded lemma selection. Translating isolated or embedded words in either direction elicited greater RT facilitation than simply reading or listening, which is transfer-appropriate if simply reading or listening does not consistently result in access to the concept. In these simple tasks, less time is required for each item, and processing may be more superficial. Participants may not have made as much effort to attend to and focus on meanings of the words and sentences that did not have to be translated.

Accuracy priming showed a different pattern across encoding tasks and was strongest in target–nontarget translation conditions. Relative to the weak comprehension in simple reading or listening, strong comprehension for target–nontarget translation is more likely to result in conceptual access and therefore more shared processes, and the difference is transfer-appropriate. In contrast, a transfer-appropriate processing explanation is less straightforward for the finding that translation elicited larger accuracy priming effects when the target was comprehended at encoding than when it was produced (as in Francis et al., 2008). It is tempting to attribute this effect to the simple fact that participants were exposed at encoding to the eventual correct naming responses. However, such exposure also occurred in the target reading and listening conditions, where accuracy priming effects were significantly smaller than in the target–nontarget translation conditions and no larger than in the nontarget–target translation conditions.

When the target word is produced at encoding, as in nontarget–target translation, only the names that a participant can generate are encountered, and there are no additional names that might provide an opportunity for transfer. The increased accuracy in this condition relative to new items may arise because some of the initial “don’t know” responses reflected tip-of-the-tongue states, failures to retrieve known words in a person’s productive vocabulary, which are more common for words with low frequency or in a less proficient language (Burke et al., 1991; Gollan & Silverberg, 2001). For some items, tip-of-the-tongue states may have been resolved by the time of the test trial.

In target–nontarget translation encoding conditions, accuracy priming was substantially stronger than in nontarget–target translation conditions. This additional facilitation indicates that some of the names presented and comprehended at encoding were words that the participant would not have been able to retrieve with only a picture cue, primarily low-frequency words or words in a less proficient language. Some of these words were known to the participant in receptive vocabulary but not yet consistently retrieved for production. Comprehension of these words at encoding increased the basis for successful retrieval in production attempts at test several minutes later. This learning phenomenon may therefore be a mechanism for moving words from receptive to productive vocabulary. Full conceptual access is consistent in the strong comprehension required for translation, which gives it a priming advantage over the weaker comprehension of simple reading or listening tasks, in which conceptual access is inconsistent. This pattern suggests that taking full advantage of the presented response requires conceptual access at encoding.

Word frequency and language proficiency effects

In nontarget–target translation conditions where the target word was produced at encoding, repetition priming effects in picture-naming RTs for isolated words were stronger for words with lower frequency (consistent with Griffin & Bock, 1998; Wheeldon & Monsell, 1992) and stronger with lower participant proficiency (consistent with Francis et al., 2003; Francis et al., 2008; Francis et al., 2014b). These frequency and proficiency effects replicated with words in sentences (see Supplemental Materials for details). For words embedded in sentences, repetition priming effects were stronger for words with lower frequency in the test language, consistent with previous research using other test tasks (Nicolas, 1996). Similarly, repetition priming in sentence conditions was stronger for participants with lower proficiency in the test language, as found with priming of production (Francis et al., 2014a). More generally, this finding is consistent with repetition-priming effects using other tests when reading was made difficult (Nicolas, 1998) or when participants had lower reading proficiency (Bourassa et al., 1998).

In target–nontarget translation conditions meant to practice comprehension at encoding, isolated words showed consistent frequency and proficiency effects in RT repetition priming (consistent with Francis et al., 2008). However, words in sentences exhibited frequency and proficiency effects only when presented in the visual modality (Experiment 1). In the auditory modality, the effects of frequency and proficiency for target–nontarget translation were weaker than when the target word was produced in nontarget–target translation. This pattern indicates that the locus of the frequency effect on repetition priming in production includes phonological selection. Frequency and proficiency effects on RT priming in simple reading/listening conditions were not consistent. Accuracy priming generally was not sensitive to word frequency or proficiency in the test language, and tests of frequency and proficiency effects were not reported when similar tasks were used in previous research.

According to the frequency-lag hypothesis, word frequency and participant proficiency effects on word production arise from a common mechanism (Gollan et al., 2008; Gollan et al., 2011). Specifically, because of fewer cumulative lifetime exposures, low-frequency words and words in a less proficient language have weaker associations with their concepts than high-frequency words and words in a more proficient language. By this logic, in bilinguals, an uneven division of lifetime usage between two languages gives rise to the differences in association strength for more and less proficient languages. Consistent with the frequency-lag hypothesis, the effects of frequency and proficiency generally followed similar patterns, but with the proficiency patterns somewhat weaker.

Another important component of the frequency-lag hypothesis is the assumption that the time required to access a word decreases in a non-linear manner, with diminishing returns for each subsequent exposure. This property of learning implies that frequency and proficiency effects will interact (Gollan et al., 2011), with larger frequency effects for less proficient speakers. Indeed, in final picture naming, both RT and accuracy exhibited this pattern (although the RT interaction did not reach significance in Experiment 2, p = .058). However, there were no parallel interactions of frequency with proficiency in the corresponding repetition-priming effects.

Conclusion

Reading or translating spoken or written words elicits learning that transfers to production several minutes later. Embedding target words to be read or translated in sentence contexts at encoding reduced repetition priming in later picture naming relative to isolated word encoding. Practicing comprehension at encoding facilitated production at test to about 76% of that observed with production practice at encoding, indicating that comprehension and production involve access to common lemma representations and that the primary locus of facilitation in repeated production is in lemma selection. Comprehension in translation elicited stronger priming effects than simple reading or listening in both RT and accuracy, suggesting that simple reading or listening tasks do not consistently result in conceptual access.

In RT, producing a target word for translation at encoding elicited stronger facilitation than comprehending it for translation, whereas in accuracy, the opposite pattern was observed. Thus, translation in either direction, whether the words are isolated or embedded in sentence contexts is good practice for later production. Comprehension at encoding increases the accuracy of production at test, whereas production at encoding speeds production at test. Practice of production is ideal for maximizing the speed of later production. However, practice of comprehension is more effective in preventing retrieval failures and may serve to move less well-learned words from receptive to productive vocabulary.

Author note

Wendy S. Francis, Bianca V. Gurrola, and Michelle Martínez, Department of Psychology, University of Texas at El Paso. This research was supported by National Science Foundation Grant BCS-1632283 to the first author. We gratefully acknowledge Priscilla Medellín and Joseph Negrón for assistance with data collection and Erika L. Guedea for assistance with data processing. Preliminary data were presented at ARMADILLO, the Southwest Cognition Conference and the Annual Meeting of the Psychonomic Society in 2018.

Appendices

Appendix 1: Stimulus sentences

English sentences	Spanish sentences
The handcuffs were attached to the chair.	Las esposas estaban atadas a la silla.
The telescope was inside the house.	El telescopio estaba dentro de la casa.
The vase was used to serve the wine.	El florero fue usado para servir el vino.
The fox ran past the bus.	El zorro rebasó al camión.
She used a bandage to put on the diaper.	Ella usó una curita para poner el pañal.
The scarf covered his ears.	La bufanda tapaba sus orejas.
One boot did not fit with the skis.	Una bota no quedaba con los esquís.
The puzzle formed an apple.	El rompecabezas formaba una manzana.
They put a strawberry in the mixer.	Ellos pusieron una fresa en la batidora.
The rooster has a lock attached to his ankle.	El gallo tiene un candado atado a su tobillo.
He connected the plug to the lamp.	Él conectó el enchufe a la lámpara .
An onion forms a circle.	Una cebolla forma un círculo.
The stroller is stitched with a thread.	La carreola está tejida con hilo.
The elephant played with the racket.	El elefante jugaba con la raqueta.
The milk was in the refrigerator.	La leche estaba en el refrigerador.
The zebra and the penguin are black and white.	La cebra y el pingüino son negro y blanco.
The bird got stuck in the chimney.	El pájaro se quedó atorado en la chimenea.
His beard covered his lips.	Su barba le tapaba los labios.
He fixed the door with a hammer.	Él arregló la puerta con un martillo.
She plays the flute in the city.	Ella toca la flauta en la ciudad.
The camel and the horse are ridden	El camello y el caballo se montan.
The kangaroo has a broken shoulder.	El canguro tiene un hombro quebrado.
The aquarium is next to the shower.	La pecera está al lado de la regadera.
His pants are hanging from the tree.	Su pantalón está colgando del árbol.
There is an ant in the drawer.	Hay una hormiga dentro del cajón.
The pyramid is not on the map.	La pirámide no está en el mapa.
The frog was in a cage.	La rana estaba en una jaula.
Her nose perceived the aroma of the cherry.	Su nariz percibió el aroma de la cereza.
The grasshopper wore a bow.	El chapulín usaba un moño.
There was a skeleton on the train.	Había un esqueleto en el tren.
The tank was parked next to the bench.	El tanque se estacionó al lado de la banca.
The scorpion ate the egg.	El alacrán se comió el huevo.
The kite flew over the mountain.	El papalote voló sobre la montaña.
The king was carrying his crown and his pipe.	El rey cargaba su corona y su pipa.
The hose filled the pool with water.	La manguera llenó la alberca con agua.
There is an orange next to the coffeepot.	Hay una naranja enseguida de la cafetera.
He ate his banana with a fork.	Él comió su plátano con un tenedor.
The tire was next to the ladder.	La llanta estaba enseguida de la escalera.
The vest has a star on it.	El chaleco tiene una estrella.
The snail was as long as the pencil.	El caracol era tan largo como el lápiz.
The castle has a tall fence.	El castillo tiene un cerco alto.
The man looked at himself in the mirror.	El hombre se miró en el espejo.
The rope wrapped around the cane.	El lazo se envolvió en el bastón.
The balloon was shaped like a turkey.	El globo tenía la forma de pavo.
The hanger was next to the cup.	El gancho estaba enseguida de la taza.
The bone was held in place with a screw.	El hueso se sostenía con un tornillo.
Her finger got stuck in the window.	Su dedo se quedó atorado en la ventana.
A potato and a pear are the same color inside.	Una papa y una pera son del mismo color por dentro.
The dolphin was transported in a truck.	El delfín fue transportado en una troca.
He put garbage in the ashtray.	Él puso la basura en el cenicero.
The bomb exploded when the bell rang.	La bomba explotó cuando la campana sonó.
He won a medal for playing chess.	El ganó una medalla por jugar ajedrez.
She stepped on the mouse with her heel.	Ella pisó al ratón con su tacón.
They took a picture of a shark with the camera.	Ellos tomaron una foto al tiburón con la cámara.
The sweater and the skirt matched.	El suéter y la falda combinaban.
He played music with only one hand.	Él tocaba música con una sola mano.
The rhinoceros balanced on a cube.	El rinoceronte se balanció en un cubo.
The bridge was connected to the island.	El puente se connectaba a la isla.
The duck ate a cookie.	El pato comió una galleta.
The knife and spoon are used to eat.	El cuchillo y la cuchara se usan para comer.
There was a rainbow behind the ship.	Había un arcoíris detrás del barco.
She always carried a safety pin and needle.	Ella siempre cargaba un seguro y una aguja.
Her long hair covered the pillow.	Su cabello largo cubría la almohada.
The backpack was left on the table.	La mochila se quedó en la mesa.
The butterfly has only one wing.	La mariposa tiene una sola ala.
The giraffe has a small tail.	La jirafa tiene una cola pequeña.
There is a flag on the mailbox.	Hay una bandera en el buzón.
He played the trumpet in the church.	Él tocó la trompeta en la iglesia.
She used a mop to clean the bathtub.	Ella usó el trapeador para limpiar la tina.
The hippopotamus was drying himself under the sun.	El hipopótamo se estaba secando debajo del sol.
She found a worm in the toaster.	Ella encontró un gusano en el tostador.
He made his mask from an arrow.	Él hizo su máscara de una flecha.
The dinosaur hid behind a rock.	El dinosaurio se escondió detrás de una piedra.
The guitar has an octopus painted on it.	La guitarra tiene un pulpo pintado en ella.
The shoe was inside the box.	El zapato estaba adentro de la caja.
She made a soup with corn.	Ella hizo una sopa con elote.
The rabbit likes the carrot.	Al conejo le gusta la zanahoria.
The candle was shaped like a lemon.	La vela tenía la forma de un limón.
The plant had a new tomato on it.	La planta tenía un tomate nuevo en ella.
The bear has a bad eye.	El oso tiene un ojo malo.
The monkey carried a flower.	El chango cargó una flor.
She broke the scissors with the iron.	Ella quebró las tijeras con la plancha.
The snake was far from her nest.	La víbora estaba lejos de su nido.
She could not find either her belt or her glasses.	Ella no pudo encontrar ni su cinto ni sus lentes.
They served the lobster with peas.	Ellos sirvieron la langosta con chícharos.
She swept the popcorn with the broom.	Ella barrió las palomitas con la escoba.
The cat played with the ball.	El gato jugó con la pelota.
The cow has a roof for shade.	La vaca tiene un techo para la sombra.
She offered tea and cake to them.	Ella les ofreció té y pastel.
The squirrel ate the grapes.	La ardilla se comió las uvas.
The deer has an injured leg.	El venado tiene una pierna herida.
The microscope was kept next to the fan.	El microscopio fue guardado al lado del ventilador.
He observed the whale under an umbrella.	Él observaba a la ballena debajo de un paraguas.
The magnet got stuck to the rocket.	El imán se quedó pegado al cohete.
The peacock was a gift for the princess.	El pavoreal fue un regalo para la princesa.
He wears a glove to carry the sword.	Él usa un guante para cargar la espada.
They kept the pineapple in the jar.	Ellos guardaron la piña en el frasco.
The fish ate a mushroom.	El pez comió un hongo.
She pressed the button to answer the telephone.	Ella apretó el botón para contestar el teléfono.
He has a cork stuck to his sock.	Él tiene un corcho pegado al calcetín.
The coach has a whistle tied to his shirt.	El entrenador tiene un silbato amarrado a su camisa.
The basket was full of cheese.	La canasta estaba llena de queso.
She dried the lettuce with the towel.	Ella secó la lechuga con la toalla.
The baby wore a pretty hat.	El bebé usó un sombrero bonito.
He left the bottle on the desk.	Él dejó la botella en el escritorio.
A gun and a shovel can be used as weapons.	La pistola y la pala se pueden usar como armas.
The pig became friends with the goat.	El marrano se hizo amigos con el chivo.
The earring was under the bed.	El arete estaba debajo de la cama.
The match was right next to the brush.	El cerillo estaba enseguida del cepillo.
His wallet was the color of a pumpkin.	Su cartera era del color de una calabaza.
The band has only one drum and one microphone.	El grupo tiene sólo un tambor y un micrófono.
Her ring was the shape of a heart.	Su anillo tenía la forma de un corazón.
The parachute was struck by lightning.	Al paracaídas se le cayó un relámpago.
The car ran the traffic light.	El carro se pasó el semáforo.
The tiger was killed with an axe.	El tigre fue matado con una hacha.
He was riding his bicycle and ran over a skunk.	Él se paseaba en su bicicleta y machucó a un zorrillo.
He put the swing next to the decorative column.	Él puso el columpio al lado de la columna decorativa.
The peanut was crushed by the skateboard.	El cacahuate se apachurró con la patineta.
The sheep was resting under the moon.	La oveja estaba descansando bajo la luna.
She picked the cigarette up with the dustpan.	Ella recogió el cigarro con el recogedor.
He hit the can with the bat.	Él le pegó a la lata con el bate.
The dog was playing with the package.	El perro estaba jugando con el paquete.
The gorilla is not afraid of the lion.	El gorila no le tiene miedo al león.
The girl was reading a book.	La niña estaba leyendo un libro.
She wore a cross when she got on the airplane.	Ella se puso una cruz cuando subió al avión.
He got his tie wet in the rain.	Él se mojó la corbata en la lluvia.
He put his suitcase inside the tent.	Él puso su maleta adentro de la carpa.
She put the butter on a plate.	Ella puso la mantequilla en un plato.
The volcano was erupting fire.	El volcán eruptaba fuego.
She saw an owl with her binoculars.	Ella vió a un búho con su miralejos.
The turtle likes to eat celery.	A la tortuga le gusta comer apio.
Her dress will not fit if she eats the hamburger.	Su vestido no le va a quedar si se come la hamburguesa.
He ate a popsicle next to the fountain.	Él se comió una paleta al lado de la fuente.
The key for the helicopter was lost.	La llave del helicóptero estaba perdida.
She wears a coat when she rides her motorcycle.	Ella usa un abrigo cuando se sube a su motocicleta.
The lizard bit his arm.	La lagartija le mordió su brazo.
He put the saxophone in the barrel.	Él puso el saxofón en el barril.
There was a spider under the rug.	Había una araña debajo del tapete.
The vacuum was next to the stove.	La aspiradora estaba enseguida de la estufa.
The helmet was as big as a watermelon.	El casco era tan grande como una sandía.

1. Note. Words in bold print are the target words, but they did not appear in bold print when presented to participants.

Appendix 2: Detailed tables of primary analyses

Table 5 Experiment 1 analysis of proficiency, frequency, and repetition effects in RT and accuracy for all items

Table 6 Experiment 1 analysis of context and encoding condition effects in repeated-item RTs

Table 7 Experiment 1 analysis of context and encoding condition effects in repeated-item accuracy

Table 8 Experiment 2 analysis of proficiency, frequency, and repetition effects in RT and accuracy for all items

Table 9 Experiment 2 analysis of context and encoding condition effects in repeated-item RTs

Table 10 Experiment 2 analysis of context and encoding condition effects in repeated-item accuracy