Language-Minority Children’s Sensitivity to the Semantic Relations between Words

Abstract

The purpose of this study was to examine automatic language processing among Spanish-speaking language-minority children. Seventy-three children (mean age = 90.4 months) completed two measures of semantic priming (an auditory lexical decision task and a looking-while-listening task). It was hypothesized that within- and cross-language semantic priming effects would occur but that translation priming effects would not occur. Results from vocabulary assessments indicated that language-minority children in this study were more proficient in English than they were in Spanish. Limited evidence for semantic priming effects within English and from English to Spanish emerged. Additionally, substantial evidence for translation priming from Spanish-to-English and from English-to-Spanish emerged. Given the lack of within-Spanish semantic priming effects and the presence of translation priming effects from Spanish to English, the results of this study indicated that Spanish-speaking language-minority children rely on translation from their less-proficient to their more-proficient language to access meaning. There was partial evidence that language-minority children’s two languages are active simultaneously, indicating that early in life children develop semantic knowledge that is associated with words known in both languages.

Children in the U.S. who speak a language other than English at home are often referred to as language-minority (LM) children because the majority of the population of the country in which they live does not speak their home language. In contrast to terminology that designates limited proficiency in one language (e.g., English language learners) or specifies the order in which languages are acquired (e.g., sequential bilinguals), LM simply indicates that children speak a language at home that is not the societal language, and LM children could have any range of language proficiency or could have sequential or simultaneous exposure to their two languages. In the U.S. the largest and fastest growing subgroup of LM children is children who are native Spanish speakers. As of 2011, over 37 million people in the U.S. (among individuals 5 years of age or older) spoke Spanish or a Spanish creole at home (U.S. Census Bureau, 2011). Furthermore, in recent years this population has expanded due to continued immigration from Latin America. Despite the cultural, linguistic, and economic diversity present within this population, Spanish-speaking LM children are at an elevated risk for struggling academically (Hemphill, Vanneman, & Rahman, 2011). An understanding of the cognitive processes that underlie the acquisition of academic skills could help identify children at risk for struggling academically and prevent those children from falling behind their peers.

Evidence indicates that children’s vocabulary knowledge is significantly related to subsequent academic outcomes (e.g., Storch & Whitehurst, 2002). According to the lexical quality hypothesis (Perfetti & Hart, 2002), children’s comprehension of spoken and written language is dependent, in part, on the complexity of children’s knowledge of word meanings and the ease with which children can access this knowledge. A substantial body of evidence exists regarding the development of language skills among LM children. Studies frequently report that LM children have significantly less vocabulary knowledge in their first (L1) or second language (L2) than do monolingual children (e.g., Hoff et al., 2012; Mancilla-Martinez & Vagh, 2013). However, it appears that this is because language exposure is distributed across two languages for LM children and not because of an inherently lower propensity for the acquisition of language skills (Bedore & Peña, 2008). Whereas most prior studies have evaluated LM children’s language skills through measures of explicit vocabulary knowledge, a novel method of examining LM children’s language skills is using cognitive tasks that capture automatic language processing across L1 and L2. One method of measuring automatic language processing is through semantic priming.

Semantic Priming among Monolinguals

Semantic priming tasks allow a test of whether presentation of one word activates other words related to the presented word through meaning. Semantic priming effects are often indicative of underlying language proficiency. Individuals with greater language proficiency typically demonstrate stronger semantic priming effects than individuals with less language proficiency (e.g., Kotz & Elston-Güttler, 2004). Although early studies of semantic priming were primarily conducted with adults and this body of literature has continued to grow (see Lucas, 2000, for a review), more recently research has begun to examine semantic priming in monolingual children. Results of this research indicate that the degree to which semantic priming effects occur is related to developmentally important outcomes, such as reading comprehension (e.g., Cremer & Schoonen, 2013). Some evidence indicates that semantic networks begin to develop as early as 21 months of age (Arias-Trejo & Plunkett, 2009), and early development of conceptual connections in the lexicon may be important for the development of reading comprehension (Nation & Snowling, 1999). Furthermore, semantic priming effects do not emerge in children with reading disability (Betjemann & Keenan, 2008). Considering these findings, it may be important to examine how automatic language processing occurs and develops early in life to identify children at risk for developing academic difficulties.

Theories of Bilingual Language Processing

Research on semantic priming effects among bilinguals is primarily centered on describing cross-language connections between words and concepts (e.g., Altarriba & Basnight-Brown, 2009; Pavlenko, 2009); however, the majority of this research has focused on adults learning a second language. Cross-language semantic priming effects (i.e., presentation of a word in L1 causing activation of a related word in L2, or vice versa), would indicate that the mental representations of L1 and L2 are simultaneously active. In other words, presentation of a word in one language allows access to meaning for that word, and activation of meaning allows for subsequent activation of a word in the other language that shares or is related in meaning.

Evidence indicates that cross-language lexical and conceptual connections in bilingual adults develop as a function of proficiency in L2 (see Kroll & de Groot, 1997, for a review). Specifically, cross-language connections are only initially made at the lexical level, not at the conceptual level. As proficiency in L2 increases, so does the strength of the link between L2 words and their underlying concepts. Once a direct connection between L2 words and their underlying concepts is established, L2 words can trigger activation of those concepts and ultimately lead to activation of L1 words. These findings have led to the formulation of several models of bilingual lexical and conceptual processing: the word association model, the concept mediation model (Potter, So, Von Eckardt, & Feldman, 1984), and the revised hierarchical model (Kroll & Stewart, 1994; see the Online Supplementary Materials for detailed descriptions of the word association model and concept mediation model).

To capture the developmental shift that occurs as individuals become more proficient in L2, Kroll and Stewart (1994) proposed the revised hierarchical model (see Figure S3 of the Online Supplementary Materials). According to this model, lexical and conceptual links between L1 and L2 vary in magnitude. Specifically, the lexical connections from L2 to L1 are stronger than are the lexical connections from L1 to L2 because translation from L2 to L1 was relied upon to access meaning when initially acquiring L2. Similarly, connections between words and their underlying concepts are stronger in L1 than in L2 because they have been established for a longer period of time. Several studies of priming effects among bilingual adults report evidence in support of the revised hierarchical model (see Kroll, Van Hell, Tokowicz, & Green, 2010, for a review). For example, in a study comparing L1 speakers of English who had either high or low proficiency in Spanish (L2), Sunderman and Kroll (2002) reported that exposure to Spanish words only activated English translation equivalents for participants with lower levels of proficiency in Spanish.

Priming Effects among Language-Minority Children

Spanish-speaking LM children in the U.S. differ from adult second language learners in a number of ways. First, adult learners of a second language presumably have a strong foundation of L1 skills upon which to build when they begin to learn L2. There is a substantial amount of knowledge both at the lexical and conceptual levels that has already been acquired because of the development of L1 proficiency. For example, as hypothesized by the word association model, adults with low L2 proficiency can rely on translating from L2 to L1 to access the meaning of L2 words. In contrast, LM children often do not have a strong L1 knowledge base upon which to build when first exposed to L2. Prior research on vocabulary knowledge among LM children indicates that their word knowledge is distributed across L1 and L2, with multiple studies reporting that approximately 70% of words were known uniquely in L1 or L2 and 30% of words were known in both languages (Goodrich & Lonigan, in press; Peña, Bedore, & Zlatic-Giunta, 2002). Because of a lack of overlap in L1 and L2 vocabulary knowledge, LM children cannot rely solely on translation from L2 to L1 to access meaning associated with L2 words. However, the extent to which LM children’s vocabulary knowledge is distributed across languages may vary as a function of several factors (e.g., age, language exposure). In contrast to the results of Peña et al., Umbel, Pearson, Fernández, and Oller (1992) reported that between 59% and 67% of words were known in both languages. Although the ages of children were similar across each of these studies (i.e., preschool to first grade), the sample selected for the Umbel et al. study came from middle- to upper-SES backgrounds in South Florida whereas the samples in Peña et al. and Goodrich and Lonigan came from lower SES backgrounds, potentially explaining the discrepant findings.

Second, many Spanish-speaking LM children in the U.S. come from lower socioeconomic status backgrounds and are, therefore, at a higher than normal risk of developing language difficulties in both of their languages. As research on monolingual English-speaking children indicates (e.g., Betjemann & Keenan, 2008; Nation & Snowling, 1999), the sensitivity to the semantic relations between words may differ for typically developing children and children with poor reading skills. Finally, the majority of the population of the country in which LM children live does not speak the same L1 as those children. Therefore, most L1 input will occur at home, with input in other environments (e.g., school) occurring primarily in L2. These relatively isolated language-learning environments may lead to the distributed nature of LM children’s vocabularies and may necessitate direct access to meaning for words in each language.

To date, only one study has evaluated semantic or translation priming effects in bilingual children. Singh (2014) evaluated within- and cross-language semantic priming effects using a preferential looking paradigm in 21 toddlers (2.5 years old) from Singapore learning English and Mandarin Chinese from birth. Results indicated that cross-language semantic priming effects occurred for bilingual toddlers; however, these effects emerged only when the prime was in the child’s dominant language. Similarly, within-language priming effects only occurred in the dominant language. Singh (2014) argued that these findings were similar to predictions derived from the revised hierarchical model; that is, cross-language semantic priming should be stronger from L1 to L2 than from L2 to L1 and within-language semantic priming should be stronger in L1 than in L2.

Although only one study to date has examined semantic priming in bilingual children, studies of other types of priming effects among children learning more than one language provide evidence for simultaneous co-activation of L1 and L2. Von Holzen and Mani (2012) demonstrated that cross-language phonological priming through translation occurred among bilingual toddlers for whom German was the L1 and English was the L2. Specifically, presentation of English words led to activation of German words that were phonologically similar to the German translation equivalent of the presented English word. For example, presentation of leg primed the German target word stein (English “stone”) because the German translation equivalent of leg is bein, which rhymes with the target word. This study provided evidence that simultaneous co-activation of L1 and L2 occurs for young bilingual children, at least to the extent that exposure to L2 words activates L1 translation equivalents. This finding is consistent with the notion that individuals learning an L2 initially rely upon translation from L2 to L1 to access meaning, as described in the revised hierarchical model (Kroll & Stewart, 1994).

Current Study

Although prior research has primarily focused on examining cross-language priming effects in the context of either infants or toddlers exposed to more than one language (e.g., Singh, 2014) or adult learners of a second language (e.g., Sunderman & Kroll, 2006), researchers have called for further examination of language processing throughout development, from infancy to adulthood (e.g., DeAnda, Poulin-Doubois, Zesiger, & Friend, 2016). Therefore, the goal of this study was to extend prior studies of semantic and translation priming among monolingual and bilingual adults and children to school-age Spanish-speaking LM children in the U.S. To evaluate priming effects among LM children, an auditory lexical decision task and a looking-while-listening task were used. Auditory lexical decision tasks have been used in prior research to evaluate priming effects among children (e.g., Betjemann & Keenan, 2008), and the looking-while-listening task is similar in design to the preferential looking paradigm used by Singh (2014). Although Singh’s findings represent an important starting point in the study of language processing among children learning more than one language, they may not replicate in a sample of LM children for a number of reasons (e.g., differences in age of acquisition of L1 and L2, isolation of L1 and L2 environments among LM children). Evidence of cross-language semantic priming in LM children would indicate that children’s L1 and L2 are simultaneously active during spoken language processing.

Considering results of prior research on semantic and translation priming as well as key features of the population of LM children, it was hypothesized that the pattern of results from this study would primarily resemble the concept mediation model (Figure S2). First, translation priming effects were not expected to occur for LM children, as prior evidence indicates that the majority (70%) of the vocabulary knowledge of LM children is known uniquely in one language (e.g., Peña et al., 2002). Because of the distributed nature of LM children’s vocabularies, it should not be possible to prime translation equivalents that are not known. Second, it was expected that within-language priming effects would occur in both L1 (Spanish) and L2 (English). One reason for the distributed nature of LM children’s vocabulary knowledge may be because input in each language is unique to a specific language-learning environment (i.e., input in L1 primarily occurs at home and input in L2 primarily occurs at school). For this reason, children need direct access to meaning from words in each language, as there may not be known translation equivalents on which children can rely to access meaning. It was also expected that cross-language semantic priming would occur from both L1 to L2 and from L2 to L1, as activation of concepts from input in one language should lead to activation of related concepts for which the word may be known in the other language. Finally, it was expected that the magnitude of within-language priming effects would be larger than the magnitude of cross-language priming effects. Because LM children’s language-learning environments are relatively isolated, words known in L1 may be words that are more relevant for use at home and words known in L2 may be words that are more relevant for use at school. Therefore, there should be greater overlap of semantic knowledge across words known within languages than there is for words known across languages.

Method

Participants

Participants were recruited from seven elementary schools across three school districts in North Florida and South Georgia. At the time of recruitment into the study, teachers and schools were asked to identify Spanish-speaking children who were eligible for the study, resulting in signed consent forms returned for 78 children who were identified as Spanish-speaking LM children. Children completed standardized measures of English and Spanish vocabulary knowledge to determine the extent of their English and Spanish language skills. Based on results of vocabulary assessments, it was determined that five children did not speak Spanish. These children were excluded from all analyses, resulting in a total of 73 children in the final sample. These children ranged in age from 75 to 108 months (M = 90.4 months, SD = 7.8 months). Thirty-nine participants were female (53.4%). According to parent report, all children except one were Latino, and all children came from homes in which Spanish was spoken to some degree. Therefore, for descriptive purposes Spanish was considered the L1 for children in this sample. Parent report indicated that Spanish was the language most commonly spoken at home for 51% of the sample; English was the language most commonly spoken at home for 6% of the sample; and Spanish and English were spoken equally at home for 43% of the sample. According to parent report, the mean annual household income for the sample was $15,000, indicating that the majority of children in this sample came from low-income backgrounds.

Measures

Vocabulary knowledge

The Picture Vocabulary subtest of the Woodcock-Johnson III Tests of Achievement (WJ-III; Woodcock, McGrew, & Mather, 2001) was used to assess children’s receptive and expressive vocabulary knowledge. On this task, children are required to either recognize a named picture (receptive) or name pictures (expressive). Participants also completed the Picture Vocabulary subtest of the Woodcock-Muñoz language survey (Muñoz-Sandoval, Woodcock, McGrew, & Mather, 2005). This test is a Spanish language adaptation of the WJ-III Picture Vocabulary subtest. Internal consistency reliability for the English and Spanish Picture Vocabulary subtests ranged from adequate to high in this sample of children (α =.72 for English, α = .90 for Spanish).

Semantic priming tasks

Participants completed two measures of semantic priming, a lexical decision task and a looking-while-listening task. Detailed descriptions of the logic behind the two semantic priming tasks used in this study are provided in the Online Supplemental Materials. Prior to administration of the semantic priming tasks, examiners spoke with participants to determine children’s preferred conversational language (i.e., English or Spanish). Instructions for the task were given in the language in which the child was most comfortable speaking. Based on brief conversations with children prior to testing, it was determined that for 27 children the preferred conversational language was Spanish and for 46 children the preferred conversational language was English.

Lexical decision task

For this study, a computer-based auditory lexical decision task was developed to measure semantic and translation priming effects. This task was a bilingual adaptation of the lexical decision task used by Nation and Snowling (1999) and the majority of real words used in the task were adapted from Nation and Snowling. Additional trials in which the two words were synonyms were included to provide an analogue to the translation equivalent trials for the within-language trial blocks. On each trial of the task, children listened to a spoken prime followed by a spoken target. In half of the trials the spoken target was a real word and in half of the trials the spoken target was a nonword. Similarly, in half of the trials the spoken prime was a real word and in half of the trials the spoken prime was a nonword. Consistent with the procedure used by Nation and Snowling, children were asked to make a lexical decision for every stimulus (i.e., children were instructed to push one button if the stimulus was a real word and a different button if the stimulus was a nonword). Responses to primes were followed by a brief pause of 500 milliseconds (ms) before presentation of the target. Responses to targets were followed by a slightly longer delay of 1000 ms to indicate a shift to the next trial. To ensure that children understood the nature of the task, children completed four practice trials in which corrective feedback was provided.

Order of administration of trials was randomized within trial blocks (e.g., L1-prime-L1-target trials; for a list of specific trials, see Table 1). See Table S1 of the Online Supplementary Materials for English translation equivalents of Spanish stimuli. Order of administration of trial blocks was counterbalanced across participants. Prime-target word pairs were randomly assigned to trial type. This design yielded 20 total trials in each of the six conditions; thus, each participant made 240 total lexical decisions. English and Spanish nonwords were created using Wuggy (Keuleers & Brysbaert, 2010), with the real words from the lexical decision task used as reference words for generating nonwords that were matched for length with the real words in the task. All English and Spanish nonwords were pronounceable in English and Spanish, respectively, and were pronounced according to the conventions of their respective language.

Table 1.

Word pairs used in the lexical decision task.

Within English (L2)	Within Spanish (L1)	English-Spanish (L2–L1)	Spanish-English (L1–L2)	English-Spanish (L2–L1) Translation	Spanish-English (L1–L2) Translation
Semantically Related
Hat-Cap	Casa-Hogar	Water-Tomar	Hermano-Sister	Glove-Guante	Camión-Truck
Chef-Cook	Ruidoso-Alboroto	King-Reina	Lago-Mountain	Beach-Playa	Granja-Farm
Hospital-Doctor	Padre-Papá	Table-Silla	Tetera-Tea	See-Ver	Carnicero-Butcher
Pants-Skirt	Cuchillo-Pan	Pig-Caballo	Película-Theater	Coat-Abrigo	Desayuno-Breakfast
Semantically Unrelated
Nose-Home	Mujer-Ave	Boot-Plato	Baño-Chicken	Pencil-Perro	Cocina-Pen
Party-Lady	Pájaro-Dama	Market-Pelota	Cama-Food	Dog-Lápiz	Pluma-Kitchen
House-Head	Vaca-Lluvia	Cup-Zapato	Tenedor-Sofa	Comb-Martillo	Ciudad-Nose
Woman-Music	Paragüas-Caballo	Bat-Verduras	Pájaro-Kitchen	Hammer-Peine	Nariz-City
Nonword-Nonword
Hule-Houd	Maba-Gomar	Manret-Secona	Fexpano-Curner	Glone-Gonste	Harez-Cery
Han-Fap	Muvor-Ebe	Rup-Kavito	Laro-Brocken	Beals-Aceto	Cizcad-Noke
Wesan-Rubic	Mafle-Maba	Kint-Riasa	Necero-Rea	Pengyl-Mecho	Canmetero-Belcher
Chasp-Clook	Mafiro-Vasa	Fagle-Tirra	Necador-Sogo	Hadder-Piase	Demaagmo-Broatbust
Fobbital-Puctor	Seiposo-Lierte	Pog-Cacirro	Babo-Moontail	Ree-Her	Cocaro-Pem
Pands-Skith	Fada-Pluvio	Buit-Flado	Taron-Fied	Dag-Fabiz	Cadien-Truns
Noke-Hoth	Paramin-Cacirro	Bap-Forcuras	Mafiro-Kenchen	Kime-Dertallo	Grarza-Fage
Marby-Dady	Cullarro-Ban	Galer-Sosar	Meleluva-Dreaker	Reet-Allazo	Crusa-Kenchen
Word-Nonword
Fish-Ned	Gorra-Dimplero	Bike-Mestar	Bote-Brip	Christmas-Hapadad	Techo-Wams
Bow-Crither	Madre-Paba	Zoo-Minsiono	Jugar-Moop	Arm-Plajo	Pan-Clead
Nonword-Word
Tasser-Dad	Encopo-Piso	Cadhit-Lechuga	Sallalla-Toast	Slill-Salvar	Aacetrazo-Airplane
Artby-Mad	Fitro-Revista	Moop-Estrella	Gorbo-Potato	Vionad-Violín	Heliboco-Telephone

Because various lexical characteristics of words have been shown to impact response times on lexical decision tasks (McDonough & Trofimovich, 2009), analyses were conducted to determine whether words in the various conditions of this study differed from each other on length and frequency. Results indicated that word frequency (as indexed by SUBTLEX-US and SUBTLEX-ESP word frequency values; Brysbaert & New, 2009; Cuetos, Glez-Nosti, Barbón, & Brysbaert, 2011) did not differ across trial blocks, F(2, 108) = 2.14, p = .12, English and Spanish words, F(1, 108) = .56, p = .46, or prime and target words, F(1, 108) = 2.71, p = .10. Word length did not differ across trial blocks, F(1, 108) = 1.84, p = .16, or prime and target words, F(1, 108) = .42, p = .52); however, Spanish words were longer than English words, F(1, 108) = 31.25, p < .001.

Looking-while-listening task

The looking-while-listening task used in this study was a Spanish-English adaptation of the preferential looking paradigm used by Singh (2014). The majority of the words used in the task were adapted from the task used by Singh, and others were included based on word association norms. Some trials were adapted from Von Holzen and Mani (2012) as additional, unrelated word pairs. In this task, children were situated directly in front of two side-by-side 16.3″ x 15.8″ flat-screen computer monitors. A video camera was situated between the two monitors to record children’s eye movements during the task and trained research assistants who were blind to the condition of the trial and location of the target image coded eye movements offline. At the onset of each trial, a fixation cross appeared at the center of the monitors to indicate to the child that the trial was beginning and center children’s attention to the point between the monitors to ensure that children were not systematically looking toward one monitor or the other at the onset of visual stimuli. An audio recording of the prime word was then played. Two hundred ms after the prime word was played, an audio recording of the target word was played. Finally, 400 ms after the target word was played, the target and distractor images appeared (one on each screen) and persisted for 2,500 ms. Images presented were color “clipart” drawings that were visual depictions of words used in the task. Trials were randomly assigned to one of four trial blocks (i.e., L1–L1, L2–L2, L1–L2, L2–L1), with the constraint that there were six semantically related and six unrelated trials within each block. This design resulted in 48 trials and 96 words in the entire task (for specific trials, see Table 2). See Table S1 of Online Supplementary Materials for English translation equivalents of Spanish stimuli. Word frequency did not differ across trial blocks, F(1, 88) = 2.72, p = .10, English and Spanish words, F(1, 88) = .75, p = .39, or prime and target words, F(1, 88) = .01, p = .93. Word length did not differ across trial blocks, F(1, 88) = .31, p = .58, or prime and target words, F(1, 88) = 1.91, p = .17; however, Spanish words were longer than English words, F(1, 88) = 14.95, p < .001.

Table 2.

Prime-target pairs and distractor images used in the looking-while-listening task.

Within English (L2)			Within-Spanish (L1)			English-Spanish (L2–L1)			Spanish-English (L1–L2)
Prime	Target	Distractor	Prime	Target	Distractor	Prime	Target	Distractor	Prime	Target	Distractor
Semantically Related
Train	Bus	Spoon	Gorra	Camisa	Cheese	Ball	Pelota	Telephone	Pájaro	Bird	Soup
Pencil	Pen	Cat	Ratón	Elefante	Jacket	Cat	Perro	Button	Mano	Hand	Bed
Bear	Tiger	Ball	Oveja	Vaca	Flower	Sun	Luna	Hand	Violín	Guitar	Cheese
Table	Chair	Flower	Zapato	Calcetín	Elephant	Kitchen	Horno	Slide	Silla	Chair	Car
Pants	Button	Apple	Granja	Puerco	Book	Cheese	Queso	Arrow	Cinturón	Pants	Tiger
Hammer	Nail	Feather	Cuchara	Tazón	Rat	Rain	Lluvia	Bird	Parque	Tree	Cup
Semantically Unrelated
Banana	Dog	Car	Queso	Pierna	Pig	Button	Tigre	Fan	Fuego	Egg	Clock
Window	Ball	Candy	Árbol	Taza	Sheep	Window	Oído	Tree	Pelota	Dress	Cookie
Elephant	Telephone	Candy	Mano	Sol	Bicycle	Pig	Camisa	Moon	Luna	Cat	Bowl
Bus	Horse	Bed	Pato	Bolígrafo	Train	Soup	Muñeca	Flower	Azul	Cow	Bed
Nose	Egg	Clock	Botella	Bolsa	Plant	Mouth	Perro	Key	Luna	Chicken	Sheep
Balloon	Fan	Bottle	Pierna	Piedra	Hat	Bottle	Oído	Tree	Nariz	Umbrella	Horse

Looking-while-listening tasks (Fernald, Zangl, Portillo, & Marchman, 2008) are functionally similar to preferential looking paradigms such as that used by Singh (2014). The primary difference between a preferential looking paradigm and a looking-while-listening task is how the outcome is coded. For a preferential looking paradigm, the outcome for each individual trial is coded as the proportion of time the participant looks to the target image out of total time looking at the target and distractor images. For a looking-while-listening task, the outcome for each individual trial is coded as the participants’ latency to look to the target image. As children get older, their language processing efficiency increases and interest in continuously looking at named pictures decreases (Fernald et al., 2008). Because preferential looking effects diminish as a function of child age, in this study, children’s eye movements were coded as reaction times (RTs) to look to the target image, defined by the point at which eye movement toward the target image began. To determine inter-rater reliability 25% of the data was double coded. When an error margin of one video frame (i.e., approximately 33 ms) was used as the criterion to compute reliability, the two raters were in agreement on 91% of all trials. Data were coded as missing on 10.9 percent of all trials. Data for individual trials were coded as missing for several reasons. First, trials were coded as missing when the child was not attending to the task (e.g., looking away from the computer monitors). Second, trials were also coded as missing when the child was already looking at the screen on which the target image appeared prior to the onset of the target image. In this instance, it was not possible to compute RTs to look to the target image because the child’s gaze was already fixated on the location at which the target image would appear. Finally, trials were coded as missing when the child never looked to the target image.

Procedure

Approval was obtained from the institutional Human Subjects Committee prior to implementation of the study (HSC No. 2016.19033; Examining Bilingual Children’s Sensitivity to the Semantic Relations Between Words). Informed consent was obtained from children’s parents or guardians prior to inclusion in the study, and child assent was obtained prior to each individual testing session. An adult female speaker proficient in English and Spanish (L1 English) recorded all audio clips for the looking-while-listening and lexical decision tasks. All measures were administered in a quiet area in the child’s school or classroom by the first author and trained bilingual research assistants. All research assistants were bilingual, native Spanish speakers, with the exception of the first author, who is a native English speaker proficient in Spanish. To avoid fatigue effects, children were given brief breaks when needed. Children were offered a small reward (e.g., sticker, eraser) for their participation at the end of each testing session. During the course of testing, 10 children were unable to complete the lexical decision task due to an inability to attend to the task or follow instructions (e.g., pressing the same button on every trial prior to hearing complete words). Data from these children were excluded from analyses for the lexical decision task. Examination of differences between children who did and did not complete the lexical decision task revealed that children who did not complete the lexical decision task were significantly younger (F[1,71] = 5.00, p < .05) and had lower Spanish (F[1,70] = 6.47, p < .05) and English (F[1,70] = 6.85, p < .05) vocabulary knowledge than did children who completed the lexical decision task. The average age of children who completed the lexical decision task was 91.16 months (SD = 7.32 months) and approximately 53% of children who completed the lexical decision task were female. Among children who completed the lexical decision task, parent report indicated that English was the language most commonly spoken at home for 3.4% of children, Spanish was the language most commonly spoken at home for 57.6% of children, and Spanish and English were spoken equally at home for 39% of children. Additionally, computer errors rendered data from the looking-while-listening task unusable for two participants. This resulted in a final sample of 63 participants for the lexical decision task and 71 participants for the looking-while-listening task.

Data Analytic Plan

Trials of interest for data analytic purposes were those trials in which the target was a real word (either related or unrelated to the prime word). Only correct responses were used for the lexical decision task, as is typical in semantic priming research (McDonough & Trofimovich, 2009). RTs to targets were analyzed as the dependent variable using a two-level mixed effects repeated measures ANCOVA to evaluate semantic and translation priming effects. In this model, a 3 × 2 × 2 design with two covariates (English and Spanish vocabulary knowledge) was used. The first factor reflected trial type (i.e., translation equivalents, semantically related, or unrelated trials). The second factor reflected whether the prime word was in English or Spanish. The third factor reflected whether the prime and target words were in the same or different languages. In these models, trials were nested within participants. Participants were treated as a random factor to model between-subjects differences in RTs to words. Trials were treated as a repeated factor (i.e., analogous to an effect in the marginal model) to model the within-subject covariance across trials after controlling for the random effect of participants¹. Thus, the correlations between the residuals for different trials were uniquely estimated (the correlation between the residuals for trials that are closer together [e.g., trials 1 and 2] may be different than correlation between the residuals for trials that are further apart [e.g., trials 1 and 10]). Effect sizes (ES) for all contrasts were computed as Hedges’ g. All analyses were conducted using SPSS Version 23.

Results

Descriptive Statistics

Data on English and Spanish vocabulary knowledge were available for 72 out of 73 children². Children’s mean standard scores for English and Spanish vocabulary knowledge were in the below-average to well-below-average range (English M = 83.31, SD = 8.19; Spanish M = 65.73, SD = 19.93). Because of the low Spanish vocabulary knowledge of children in this sample, the interaction between Spanish vocabulary knowledge and trial type was included in analyses. It was expected that semantic and translation priming effects would only emerge for those children with higher levels of Spanish vocabulary knowledge. Additionally, analyses were conducted to examine priming effects in the context of children’s more-proficient language versus their less-proficient language, rather than English versus Spanish. Results of these analyses are reported in the Online Supplementary Materials (see Figure S4).

Lexical decision task

For the lexical decision task, response accuracy (i.e., whether children correctly indicated if each stimulus was a real word) to all words and non-words in the task was 72.2%. Response accuracy to targets that were real words (i.e., the primary outcome of interest) was 79.7%. ANOVA indicated that response accuracy did not differ as a function of trial type, F(2, 3744) = 1.49, p = .23. To avoid potential influence of outliers on results, RTs greater than three standard deviations above the grand mean RT on all trials (M = 1603.51 ms, SD = 978.57) were not included in analyses. Approximately 1.5% of all RTs were outliers. After outliers were removed from the data, mean RT for all trials was 1522.90 ms (SD = 643.00). Mean RT to the outcome of interest for data analysis (i.e., targets that were real words with correct responses) was 1368.97 ms (SD = 541.33).

Mean RTs for unrelated, semantically related, and translation equivalent trials within each trial block for the lexical decision task are reported in the upper panel of Table 3³. Results of analyses indicated that there was a significant effect of trial type, F(2, 2689.01) = 3.43, p < .05. Children were approximately 75 ms faster to respond to semantically related trials than they were to respond to translation equivalent trials (ES = .13), and children were approximately 49 ms faster to respond to unrelated trials than they were to respond to translation equivalent trials (ES = .10). Response times to unrelated and semantically related trials were not significantly different (ES = .04). This pattern of results indicated that neither translation nor semantic priming effects occurred on this task. There were no other significant main effects or interactions of trial-level factors. The main effect of English vocabulary knowledge was positive and statistically significant (b = 8.15, p < .05). Children with higher English vocabulary knowledge were slower to respond to targets that were real words than were children with lower English vocabulary knowledge. However, the zero-order correlation between RTs and English vocabulary knowledge was small (r = .11). Spanish vocabulary knowledge was not significantly related to RTs.

Table 3.

Unadjusted mean RTs (in milliseconds) and SDs to targets that were real words for all conditions in the lexical decision task and looking-while-listening task.

	Unrelated	Related	Translation
	Lexical Decision Task
Average	1361.93 (517.06)	1343.90 (540.65)	1415.29 (606.43)
Within English	1324.89 (497.23)	1305.28 (460.53)	1482.99 (574.74)a
Within Spanish	1319.40 (565.19)	1283.96 (510.55)	1343.20 (589.16)a
English-Spanish	1383.98 (483.66)	1351.40 (513.44)
Spanish-English	1355.50 (529.55)	1379.74 (614.15)
English-Spanish Translation	1394.64 (542.69)		1452.09 (708.72)
Spanish-English Translation	1387.32 (485.13)		1383.58 (521.83)
	Looking-while-listening task
Average	991.36 (463.29)	1004.34 (450.43)	854.60 (439.91)
Within English	1017.44 (438.98)	962.22 (435.54)
Within Spanish	1022.70 (475.89)	1086.24 (471.93)
English-Spanish	983.38 (387.99)	882.39 (441.38)	838.27 (419.15)
Spanish-English	939.75 (539.56)	1052.06 (408.30)	871.10 (460.49)

Note.

^aTranslation trials for within language conditions on lexical decision task were approximated by synonym word pairs.

Because Spanish vocabulary knowledge for children in this sample was low, the interaction between trial type and children’s Spanish vocabulary knowledge was added to the model. All significant results in the original model remained significant when the interaction between trial type and Spanish vocabulary knowledge was included in analyses. The interaction between trial type and children’s Spanish vocabulary knowledge was statistically significant, F(2, 2789.79) = 3.21, p < .05. Mean RTs for each trial type at varying levels of Spanish vocabulary knowledge are shown in Figure 1. Follow-up contrasts revealed that the difference in RTs for unrelated trials and translation equivalent trials and the difference in RTs for related trials and translation equivalent trials differed across varying levels of Spanish vocabulary knowledge. Differences in RTs were evaluated at the mean of Spanish vocabulary knowledge and at one SD above and below the mean of Spanish vocabulary knowledge. At one SD below the mean of Spanish vocabulary knowledge, the mean RT to unrelated trials was approximately 101 ms faster than was the mean RT to translation equivalent trials (ES = .19). At the mean of Spanish vocabulary knowledge, the mean RT to unrelated trials was approximately 48 ms faster than was the mean RT to translation equivalent trials (ES = .09). At one SD above the mean of Spanish vocabulary knowledge, the difference in mean RT (1.68 ms) to unrelated trials and translation equivalent trials was not statistically significant (ES = .00). Similarly, at one SD below the mean and at the mean of Spanish vocabulary knowledge the mean RT to related trials was faster than was the mean RT to translation equivalent trials (137.73 ms [ES = .25] and 66.73 ms [ES = .12] difference in RT at one SD below the mean and at the mean of Spanish vocabulary knowledge, respectively). At one SD above the mean of Spanish vocabulary knowledge, the difference in mean RT (1.15 ms) to related trials and translation equivalent trials was not statistically significant (ES = .00). RTs for unrelated and related trials did not differ across levels of Spanish vocabulary knowledge. This pattern of results indicated that children with very low Spanish vocabulary knowledge were especially slow to respond to target words when the prime words were the translation equivalents of the target words.

Figure 1.

Mean RTs for unrelated, related, and translation equivalent trials at varying levels of Spanish vocabulary knowledge for the lexical decision task.

Because language of task administration differed across children based on preferred conversational language, analyses were conducted to determine whether language of task administration influenced results. There was a significant main effect of language of task administration, such that reaction times were significantly faster for children who completed the task in English than they were for children who completed the task in Spanish, F(1, 67.61) = 4.24, p < .05. Additionally, the interaction between trial type and whether the prime was in English or Spanish was moderated by language of task administration, F(2, 2688.01) = 3.15, p < .05). The pattern of results for this interaction indicated that when the prime word was in English and language of test administration was English, reaction times to translation equivalents were significantly slower than were reaction times to related trials. Similarly, when the prime word was in English and language of test administration was Spanish, reaction times to translation equivalents were significantly slower than were reaction times to related and unrelated trials. No other interaction contrasts were statistically significant.

Looking-while-listening task

Mean RTs for unrelated, related, and translation equivalent trials within each trial block for the looking-while-listening task are reported in the lower panel of Table 4⁴. Results of the 3 × 2 × 2 mixed model revealed significant main effects of all trial-level factors (for trial type, F[2, 2876.80] = 8.90, p < .001; for language of the prime word, F[1, 1394.45] = 11.16, p < .01; for whether prime and target words were in the same language, F[1, 1320.20] = 10.97, p < .001). The mean RT to translation equivalent trials was significantly faster than was the mean RT to semantically related trials (mean difference = 135.64 ms, ES = .30) or unrelated trials (mean difference = 134.21 ms, ES = .29). RTs for unrelated trials were not significantly different than RTs for semantically related trials (ES = .00). This pattern of results suggested that translation priming effects occurred from both Spanish to English and from English to Spanish. RTs were 51.52 ms faster for trials in which the prime word was in English than they were for trials in which the prime word was in Spanish (ES = .13), and RTs were 92.93 ms faster for trials in which the prime and target words were in different languages than they were for trials in which the prime and target words were in the same language (ES = .20).

The only significant interaction was the interaction between trial type and the language of the prime word, F(2, 2867.35) = 13.98, p < .001. English and Spanish expressive vocabulary knowledge did not significantly predict RT to targets. The interaction between trial type and Spanish vocabulary knowledge was not statistically significant, F(2, 2894.28) = .05, p = .95. When the interaction between trial type and Spanish vocabulary was included in the model the main effect of trial type became non-significant, F(2, 2888.20) = 1.16, p = .32. Because the interaction between trial type and Spanish vocabulary was not statistically significant, this term was dropped from all subsequent models. There were no significant main effects or interactions of language of task administration, indicating that semantic and translation priming effects did not differ based on whether language of task administration was English or Spanish.

To determine the nature of the interaction between trial type and language of the prime word, contrasts for trial type were evaluated for trials in which the prime word was in English and Spanish separately. Results of this interaction are reported in Figure 2. Follow-up contrasts revealed that the mean RT to translation equivalent trials was 192.31 ms faster than was the mean RT to semantically related trials when the prime word was in Spanish (ES = .42) and it was 78.98 ms faster when the prime word was in English (ES = .18). Additionally, the mean RT to translation equivalent trials was 101.40 ms faster than was the mean RT to unrelated trials when the prime word was in Spanish (ES = .20), and it was 167.02 ms faster when the prime word was in English (ES = .40). This pattern of results indicated that translation priming effects occurred from both Spanish to English and from English to Spanish. The mean RT to semantically related trials was 88.04 ms faster than was the mean RT to unrelated trials when the prime word was in English (ES = .21); however, the mean RT to unrelated trials was 90.91 ms faster than was the mean RT to semantically related trials when the prime word was in Spanish (ES = −.19). This pattern of results indicated that within-English semantic priming effects and cross-language semantic priming effects from English to Spanish occurred for LM children. In contrast, there was a negative semantic priming effect for within-Spanish trials and Spanish-to-English trials.

Figure 2.

Mean RTs for unrelated, related, and translation equivalent trials when the prime word was in Spanish and when the prime word was in English for the looking-while-listening task.

Discussion

The purpose of this study was to evaluate within- and cross-language semantic and translation priming effects in a sample of first and second grade Spanish-speaking LM children. Semantic and translation priming effects were used to measure automatic language processing and to investigate whether LM children’s two languages are simultaneously active. It was hypothesized that the pattern of results would primarily resemble the concept mediation model (Potter et al., 1984); however, results obtained more closely approximated the revised hierarchical model (Kroll & Stewart, 1994). Despite no effects of semantic or translation priming on the lexical decision task, there were strong translation priming effects and some semantic priming effects on the looking-while-listening task. Specifically, translation priming effects occurred in both directions and semantic priming effects occurred within L2 and from L2 to L1. Although results were not consistent across tasks, there was partial evidence that exposure to words in L1 or L2 led to simultaneous activation of both languages. This finding suggests that LM children’s semantic knowledge is shared across L1 and L2, to some extent.

Translation Priming

It was expected that translation priming effects would not occur for LM children in this study. Evidence suggests that LM children often know words in one language and not the other (e.g., Peña et al., 2002). If this were the case and few translation equivalent words were known in L1 and L2, it would not be possible to prime words through their translation equivalents. However, results of this study suggested that translation priming did occur for this sample of children. The looking-while-listening task showed substantial translation priming effects, such that RTs to translation-equivalent word pairs were significantly faster than were RTs to unrelated word pairs.

In this sample, translation priming effects occurred from both Spanish to English and from English to Spanish. According to the revised hierarchical model (Kroll & Stewart, 1994), translation priming effects should be stronger from L2 to L1 than from L1 to L2; however, this model was developed to explain connections between the L1 and L2 lexicons in bilingual adults who were proficient in L1 prior to learning L2. Although Spanish was the L1 for children in this study, scores on English and Spanish vocabulary assessments indicated that English was these children’s more-proficient language at the time of the study. For LM children, it may be more appropriate to conceptualize the revised hierarchical model in terms of relative levels of language proficiency rather than L1 versus L2. It is possible that LM children rely on translation from their less- to their more-proficient language, regardless of which language they began to acquire first. However, the pattern of results obtained in this study was not entirely consistent with the revised hierarchical model, as translation priming effects from English to Spanish were somewhat stronger than were translation priming effects from Spanish to English.

Observed RTs for translation-equivalent word pairs were also significantly faster than were RTs to semantically related word pairs; however, this effect was more pronounced when the prime word was in Spanish. This finding indicated that established lexical links from Spanish to English were stronger than were connections between words and concepts in Spanish, a finding consistent with prior research that has indicated that bilingual adults with low L2 proficiency are faster at translating from L2 to L1 than they are at naming pictures in L2 (Chen & Leung, 1989). According to the revised hierarchical model, the relative strength of lexical and conceptual links is dependent on the usage of these links. For example, the lexical links from the less- to the more-proficient language should be stronger than the lexical links in the opposite direction because translation from the less- to the more-proficient language was initially relied upon to access meaning. In contrast, direct access between the lexical unit and its underlying conceptual referent already existed in the more-proficient language when the individual began acquiring a second language, eliminating the need to translate from the more- to the less-proficient language. The majority of formal language instruction for children in this sample took place in English. Therefore, it seems plausible that children would have direct access to the meanings of words in English and rely on translation from Spanish to English to access the meanings of Spanish words.

Results from the lexical decision task did not show effects of translation priming. In fact, children’s RTs to semantically related word pairs and unrelated word pairs were significantly faster than were their RTs to translation equivalent word pairs. Significant moderation of translation priming effects on the lexical decision task by children’s Spanish vocabulary knowledge indicated that children’s RTs to unrelated word pairs were faster than were their RTs to translation equivalent word pairs for children with low and average but not high Spanish vocabulary knowledge. It is possible that this effect was due to poor performance on the lexical decision task for children with low Spanish vocabulary knowledge; however, examination of the response accuracy rate suggested that this was not the case. Therefore, it is more likely that children with lower Spanish vocabulary knowledge simply had less overlap in words known in their two languages than did children with higher Spanish vocabulary knowledge. Translation priming effects may have emerged in a sample of older children or children with higher Spanish vocabulary knowledge. Results of one study suggest that children are more likely to acquire translation equivalents of words known in one language but not the other than they are to acquire words not known in either language (Goodrich, Lonigan, Kleuver, & Farver, 2016). As children’s language skills in Spanish and English continue to co-develop they should acquire more translation equivalents of words they already know in one language. Once children know an adequate amount of translation equivalents in English and Spanish it should be possible to detect translation priming effects.

Semantic Priming

An overall main effect of semantic priming within and across languages (i.e., RTs for semantically related word pairs versus RTs for unrelated word pairs) did not emerge for either the lexical decision task or the looking-while-listening task in this study. However, there were significant semantic priming effects within English and from English to Spanish on the looking-while-listening task. Mean RTs on the lexical decision task were trending in a similar direction for trials in which the prime word was in English, but this trend was not statistically significant. The overall lack of priming effects observed in this study is in contrast to a large body of research on semantic priming effects among adults learning a second language (e.g., Basnight-Brown & Altarriba, 2007; Kroll et al., 2010), and limited evidence among children exposed to more than one language (Singh, 2014). One plausible explanation for the lack of significant semantic priming effects is the pattern of relations between children’s English and Spanish language skills. For within-language semantic priming effects to occur, the strength of the link between words and their underlying concepts needs to be sufficient to activate other words that share semantic information. According to the revised hierarchical model (Kroll & Stewart, 1994), as proficiency in one language increases, the strength of the link between words and their underlying concepts should also increase. Despite relatively high levels of English proficiency when compared to Spanish proficiency, children in this study had below-average English proficiency. However, evidence from semantic priming research among monolingual children with poor reading comprehension suggests that some within-language priming effects should occur for children with below-average English proficiency (Nation & Snowling, 1999). For cross-language semantic priming effects to occur, a relatively high degree of proficiency in both languages is needed. This is supported by prior evidence consistent with the word association model for adults with low L2 proficiency (e.g., Chen & Leung, 1989; Kroll & Curley, 1988) and prior evidence consistent with the concept mediation model and revised hierarchical model for adults with high L2 proficiency (e.g., Basnight-Brown & Altarriba, 2007; Kroll et al., 2010). Additionally, Singh (2014) reported that cross-language semantic priming effects occurred when the prime word was in children’s dominant language.

The negative semantic priming effect for word pairs in which the prime word was in Spanish suggests that children did not have the language proficiency in Spanish necessary to establish well-developed semantic knowledge for Spanish words. In other words, although children had some Spanish vocabulary knowledge, there was not sufficient depth of semantic knowledge for exposure to Spanish words to lead to activation of related words in either Spanish or English. Although it would be expected that a lack of semantic knowledge associated with low Spanish language skills would result in no semantic priming rather than negative priming effects, it is possible that slow processing of Spanish prime words extended into the time at which the target word was presented, interfering with children’s processing of that word. In contrast, the interaction between trial type and language of the prime word for the looking-while-listening task indicated that children were able to access Spanish words when the underlying concepts were activated by English words. Perhaps it is easier for LM children to retrieve a known word in the less-proficient language when its underlying concept is already activated than it is to access meaning from a word in the less-proficient language without relying on translation to the more-proficient language. These results were somewhat consistent with the revised hierarchical model, as predictions derived from that model would suggest stronger semantic priming effects from the more- to the less-proficient language than vice versa; however, according to the revised hierarchical model, effects of semantic priming from the less- to the more-proficient language should be significant as well.

Theoretical Implications

The results of this study are most likely due to an unexpected pattern of relations in children’s English and Spanish vocabulary knowledge. It was expected that children’s Spanish language skills would be stronger than their English language skills, consistent with the typical conceptualization of LM children. However, children in this study had stronger English language skills than Spanish language skills--despite parental report that Spanish was the language most frequently spoken at home for 51% of the children. This pattern of language proficiency may have emerged, in part, because these children had received academic instruction in predominantly English-language environments. Consequently, these children were more similar to adult learners of a second language than was expected, and the patterns of semantic and translation priming effects reflected these similarities.

It was not expected that effects would emerge on one priming task but not the other. It is possible that the visual component of the looking-while-listening task assisted children with weaker oral language skills in recognizing stimuli. In contrast, the lexical decision task had no visual component to assist children with stimulus recognition. This may explain the significant moderation of priming effects by children’s Spanish vocabulary knowledge for the lexical decision task but not for the looking-while-listening task. This explanation is consistent with the finding of Chen and Leung (1989) that young children learning L2 rely on pictorial representations of L2 words to access meaning. Additionally, semantic priming tasks are designed to measure automatic language processing. However, when children (or adults) have weak language skills, a task that requires a decision regarding whether a stimulus was a real word may be more indicative of conscious evaluation of words than it is of automatic processing. Consistent with this explanation, higher English vocabulary knowledge was associated with slower responses to targets that were real words on the lexical decision task. This unexpected finding could be due to children with more vocabulary knowledge exerting a conscious effort to determine whether the auditory stimulus represented a real word. Therefore, differences in priming effects across the two tasks used in this study could be a result of fundamental differences in the aspects of language processing required by each task.

For LM children, a theoretical model of the conceptual and lexical links between the two languages that is based on relative levels of language proficiency may be more appropriate than is a model based on which language the child learned first or which language is most commonly spoken at home (i.e., L1 vs. L2). This model would be similar to the revised hierarchical model (see Figure S3). Specifically, L1 and L2 would be replaced with more- and less-proficient languages, respectively. The degree to which the effects hypothesized by the revised hierarchical model directly apply to LM children should be dependent on children’s degree of proficiency in each language. For children with a large discrepancy in relative levels of proficiency in their two languages, lexical links between L1 and L2 should be well established. Similarly, direct access from words to concepts should only exist in the more-proficient language. There is the potential for cross-language conceptual connections from the more- to the less-proficient language, depending on the extent to which children can retrieve words in the less-proficient language after the activation of the underlying conceptual referent. As LM children’s levels of L1 and L2 proficiency increase, the pattern of relations between the two languages should more closely resemble the revised hierarchical model.

Limitations

Although this study is one of the only studies to examine semantic priming effects in LM children and had some methodological strengths (e.g., two independent measures of priming effects), it had several limitations. The primary limitation of this study was that the majority of children in this sample had stronger English than Spanish language skills, limiting the generalizability of these findings to LM children with different patterns of strengths and weaknesses in L1 and L2. For example, it is plausible that the pattern of relations between children’s L1 and L2 would differ with a sample of children that are more typical of “English language learners” (i.e., average Spanish language skills, limited English proficiency) or with a sample of children who are balanced bilinguals. Future research examining the connections between children’s languages should include children with different patterns of L1 and L2 language skills. A second limitation of this study is that the lexical decision task may have been too difficult for first and second grade LM children. The correct response rate of LM children who participated in this study was lower than was the correct response rate of children who were of similar age or older in prior research (Betjemann & Keenan, 2008; Nation & Snowling, 1999; Plaut & Booth, 2000). However, correct response rate was not significantly correlated with child age in this study. Future research should evaluate semantic priming effects in older LM children or forego using an auditory lexical decision task to evaluate priming effects among LM children. Finally, we did not assess whether children knew the meanings of all of the words used in the semantic priming tasks. Because the tasks were adapted from those used in prior priming experiments with children, it was expected that words used on the priming tasks would be known by children in this study. However, the sample in Nation and Snowing (1999) was slightly older than were the children in this study, and therefore it is possible that words on the lexical decision task were too difficult for the first and second grade children in this study. Therefore, the fact that we did not assess knowledge of words on the priming task is an especially salient limitation for the lexical decision task, as it was not possible to determine if the lack of priming effects for that task was due to the nature of the task or if children simply did not know the words used on the task, preventing the possibility of priming them.

Some characteristics of the semantic priming tasks used in this study (and other studies in this area [e.g., Altarriba & Basnight-Brown, 2007; Singh, 2014]) limit the extent to which conclusions about semantic priming effects among LM children can be drawn. For example, the tasks in this study did not use the same target words across the different conditions of the task. Because of this, it is not possible to determine how different prime words affect processing of the target word. Additionally, translation priming could occur through lexical or conceptual links, and evidence from a single priming task cannot isolate the route through which translation equivalents were primed. Finally, we did not systematically manipulate phonological similarity of prime and target words across languages, and prior research indicates that significant phonological priming effects occur across languages for bilingual children, providing evidence of language co-activation (e.g., Von Holzen and Mani, 2012)⁵. Future research should design different types of priming tasks to more clearly describe how priming effects occur among LM children and ensure that priming effects are a result of automatic language processing and not confounding lexical characteristics.

Conclusions

Although results of this study did not support a priori hypotheses, limited evidence of semantic priming alongside more substantial evidence of translation priming suggested that L1 and L2 language processing in LM children unfolds in a similar manner as it does for adult learners of a second language. That is, the nature of language processing for individuals learning more than one language seems to be dependent on the degree of proficiency in L1 and L2. This study provided some evidence that LM children’s languages are simultaneously active. Additionally, results suggested that LM children rely on translation from their less- to their more-proficient language to access meaning. The results of this study provide evidence of how language skills develop among LM children and may be relevant for developing instructional methods that are designed to utilize children’s knowledge in one language when learning another language. Future research should continue to utilize measures of automatic language processing in addition to using explicit measures of language skills to advance knowledge of how LM children acquire L1 and L2.

Supplementary Material

supplement

Figure S1. According to the word association model, translation priming effects should occur from the second language (L2) to the first language (L1), and within-language semantic priming should occur in L1.

Figure S2. According to the concept mediation model, concepts can be directly accessed from input in the first and second language (L1 and L2, respectively). Therefore, within-language semantic priming effects should occur in L1 and L2 and cross-language semantic priming effects should occur from L1 to L2 and from L2 to L1.

Figure S3. According to the revised hierarchical model, there is a developmental shift from reliance on translation from the second language (L2) to the first language (L1) to having direct access to concepts in both L1 and L2. Therefore, translation priming effects should be stronger from L2 to L1 than from L1 to L2, within-language priming effects should be stronger for L1 than L2 and cross-language semantic priming effects should be stronger from L1 to L2 than from L2 to L1.

Figure S4. Mean RTs (in milliseconds) for each trial type of the looking-while-listening task when the prime word was in the more-proficient versus the less-proficient language.

Figure S5. Mean RTs for each condition of the looking-while-listening task by type of semantic relation between words.

Table S1. English translation equivalents of Spanish words presented in the lexical decision task.

Highlights.

• Language-minority children were more proficient in English than Spanish

• Translation priming occurred from Spanish to English

• Semantic priming occurred within English and from English to Spanish

• Evidence that L1 and L2 are simultaneously active for language-minority children