Distinct Neural Signatures of Novel Word–Referent Mapping in Monolingual and Bilingual Toddlers Authors: Maria M. Arredondo, Drew Weatherhead Published: 2026 ABSTRACT ABSTRACTEvidence suggests that bilingualism impacts children's lexical and brain development. Yet, the development of the brain's language network in early childhood–a period marked by rapid vocabulary growth–remains poorly understood, particularly with respect to differences between bilingual and monolingual children. This study used eye tracking and functional near‐infrared spectroscopy (fNIRS) to examine the cortical representations of lexical acquisition in frontal and temporal lobes. We relied on a fast mapping task that included three trial types: Familiar (familiar label–referent mapping; e.g., ball), Disambiguation (novel label–referent mapping; e.g., dax), and Retention (long term retention of two novel referents). Sixty English‐learning monolingual and bilingual children (27–32 months), from a majority English‐speaking society, participated in the study. Results revealed that monolinguals showed increased looking to the novel referent during disambiguation trials, with a similar trend during retention. In contrast, bilinguals did not show reliable novel label–referent mapping behavior. Brain activity differences between the language groups emerged in right temporal regions during disambiguation trials, with bilinguals showing increasing HbO activity in this region, than monolinguals. Bilinguals’ right hemisphere greater activity may reflect reduced automaticity in the use of disambiguation. These results offer foundational insight into neural markers supporting lexical acquisition, especially for those who may struggle in language acquisition such as late talkers. Introduction Language is supported by a brain network that is predominantly left‐lateralized within the frontal and temporal lobes. In the early stages of development, infants often show a left‐lateralized brain network when processing language (Peña et al. 2003; Dehaene‐Lambertz et al. 2006; Molavi et al. 2014; Petitto et al. 2012), and any inconsistent results showing a bilateral response to language are generally interpreted as a reflection of less linguistic experience or immature neural specialization (Garcia‐Sierra et al. 2011). Bilingualism is a common early life experience, which poses the question whether experience‐dependent brain development for language processing is based on experience with one or two languages. Bilingual children may receive less input in each of their languages relative to monolingual peers, though this depends on the total amount of language exposure, which can vary widely across families. When exposure across both languages is considered together, bilinguals encounter a wider range of linguistic features, including diverse phonemes, lexical items, morphological structures, and more. Furthermore, evidence suggests that bilingualism may change the brain both structurally (e.g., increased grey matter volume; Mechelli et al. 2004) and functionally (e.g., greater brain activity in left frontal regions during linguistic and non‐linguistic attention tasks; Arredondo et al. 2022; Arredondo et al. 2019a, Arredondo et al. 2017). Here, we focus on lexical acquisition given differences in bilinguals’ and monolinguals’ lexicon from an early age. Little is known about the brain functions that support lexical acquisition in early childhood. Studying bilingual and monolingual experiences provides a natural opportunity to understand the functional development of the brain's language network, how words are functionally represented during learning, and whether the processes supporting lexical acquisition differ across language groups. Summary This study investigates whether functional brain activity for word acquisition differs between bilingual and monolingual toddlers during referent selection and retention.We measured brain activity using fNIRS while children completed a fast mapping task, in which familiar and unfamiliar objects were presented and labeled.Monolingual children successfully mapped novel labels to referents, while bilinguals did not.Brain differences emerged in bilinguals’ right temporal regions, suggesting reduced automaticity with disambiguation. This study investigates whether functional brain activity for word acquisition differs between bilingual and monolingual toddlers during referent selection and retention. This study investigates whether functional brain activity for word acquisition differs between bilingual and monolingual toddlers during referent selection and retention. We measured brain activity using fNIRS while children completed a fast mapping task, in which familiar and unfamiliar objects were presented and labeled. We measured brain activity using fNIRS while children completed a fast mapping task, in which familiar and unfamiliar objects were presented and labeled. Monolingual children successfully mapped novel labels to referents, while bilinguals did not. Monolingual children successfully mapped novel labels to referents, while bilinguals did not. Brain differences emerged in bilinguals’ right temporal regions, suggesting reduced automaticity with disambiguation. Brain differences emerged in bilinguals’ right temporal regions, suggesting reduced automaticity with disambiguation. The Lateralization of the Language Network and Developmental Trajectories Theoretical models of language processing in the mature adult brain describe a dual‐stream network that supports speech perception, sound‐to‐meaning mapping, and language comprehension (Hickok and Poeppel 2004, 2007; Friederici 2011, 2012, 2015; Saur et al. 2008). The dual‐stream model outlines two primary pathways: dorsal and ventral. The dorsal stream is strongly left‐lateralized and links left temporal regions to the left inferior frontal gyrus. It is primarily involved in phonological processing, morpho‐syntactic processing, and speech production. The ventral stream is crucial for sound to meaning mapping, and therefore, is key for lexical acquisition. Unlike the dorsal stream, the ventral stream appears to be less left hemisphere dominant and perhaps more bilateral, connecting the posterior superior temporal regions to middle and inferior fronto‐temporal cortices (Friederici 2011, 2012; Skeide et al. 2016; Skeide and Friederici 2016). The core language network begins to emerge early in development. Infants engage left frontotemporal regions when processing and encoding speech sounds (Travis et al. 2011). For instance, infants repeatedly exposed to novel words over the span of three days exhibited increasing left‐hemispheric engagement when recognizing those words (Rossi et al. 2015; Obrig et al. 2017). During the first three years of life, children rely heavily on bottom‐up processes as they associate phonological word forms with semantic representations stored in memory (Skeide and Friederici 2016). Evidence suggests that top‐down mechanisms supporting access to and selecting of lexico‐semantic concepts begin to emerge around two years‐old and continue to mature through late childhood (Brauer and Friederici 2007; Skeide et al. 2014). This extended period in development likely reflects the protracted myelination of white matter tracts connecting temporal and frontal regions across both hemispheres (Dubois et al. 2015; Mohades et al. 2015; Skeide et al. 2014). With increased age and language experience, children's neural patterns for lexico‐semantic processing become increasingly adult‐like, shifting from reliance in both hemispheres toward a left‐lateralized brain network (Arredondo et al. 2019a, 2019b; Dehaene‐Lambertz et al. 2010; Hickok and Poeppel 2007; Skeide and Friederici 2016). Bilingualism in Lexico‐Semantic Processing Lexico‐semantic processing in bilingual and monolingual adults engages broadly similar brain regions. However, key differences emerge in frontal regions, particularly when bilinguals are less proficient in the language of testing (Abutalebi 2008; Arredondo et al. 2019a; Indefrey 2006; Leonard et al. 2010; Pang 2012). Bilinguals who are highly proficient in both languages show activation patterns that resemble those of monolinguals, primarily in left‐lateralized temporo‐parietal and frontal regions. When proficiency is lower, activation becomes more widespread, with greater recruitment on the left frontal cortex and, in some cases, bilateral frontal engagement, suggesting increased cognitive control demands and less automatic lexical access (Arredondo et al. 2019b; Indefrey 2006; Chee et al. 2001; Abutalebi 2008; Perani et al. 2003). Indeed, neural activation is typically more bilateral when lexical tasks are especially demanding. Studies with adult second‐language learners show that successful word learning is associated with left‐lateralized activation in frontal and temporal regions, whereas failed word learning elicits more bilateral or right‐hemisphere activation reflecting greater cognitive effort and reduced efficiency in processing (Yang et al. 2015). This pattern suggests that left‐lateralized engagement may serve as a neural marker of successful lexical learning, whereas broader or right‐hemisphere engagement may signal increased cognitive control and resource allocation. The present study investigates whether the functional network supporting word learning differs between bilingual and monolingual English‐learning toddlers. Specifically, we sought to identify the brain regions toddlers recruit during an indirect word learning task. To do so, we used a well‐established Fast Mapping paradigm based on the intermodal preferential looking procedure (Halberda 2003), in which toddlers first complete a referent selection phase that includes familiar word–referent mappings (Familiar trials) and novel word–referent mappings (Disambiguation trials), and are then tested on their memory of the new word–referent pairs during a retention phase (Retention trials). By examining the functional brain regions that toddlers rely on during fast mapping, we can characterize the developing language network that supports lexical acquisition in early childhood. Critically, we can also assess whether monolingual and bilingual toddlers recruit similar or distinct neural mechanisms during this process, and whether these differences are related to their language experience or performance. Theoretical Framework of Indirect Word Learning Much of early language acquisition relies on indirect word learning, in which the referent is not explicitly labeled. Young children are adept at mapping a new label onto an unfamiliar noun when it is paired with a familiar object; this is known as a disambiguation response (Clark 1990; Golinkoff et al. 1994; Halberda 2003; Baldwin and Markman 1989; Merriman et al. 1989). Children raised in monolingual households show this ability as early as 17 months, reliably inferring that a novel label belongs to the unfamiliar referent (Halberda 2003). Increasingly, researchers question whether disambiguation leads to subsequent retention of novel words. At 24 months, toddlers typically succeed in mapping novel labels to novel referents during disambiguation trials (Bion et al. 2013; Kalashnikova et al. 2018; Weatherhead et al. 2021). However, retention of these mappings is more precarious. Some studies suggest that 24‐month‐olds can retain novel mappings when they are reinforced through linguistic and/or ostensive signals. For example, when toddlers are familiarized with the novel object or label beforehand, or when speaker lip movements are visible throughout the task (Horst and Samuelson 2008; Kucker and Samuelson 2012; Spiegel and Halberda 2011; Weatherhead et al. 2021). Without such cues, most studies indicate that monolingual 24‐month‐olds fail to retain mappings acquired through fast mapping (Bion et al. 2013; Horst and Samuelson 2008; Rocha‐Hidalgo et al. 2021; Weatherhead et al. 2021). One notable exception is Kalashnikova et al. (2018), which found evidence of retention as early as 18‐months. Overall, stable retention of fast‐mapped words is generally not observed until around age three, improving steadily thereafter (Bion et al. 2013). Research on bilingual toddlers is even more limited and presents some conflicting results. Bilingual toddlers only sometimes show a disambiguation response at 24‐months, and they are less likely to retain novel words learned through fast mapping (Houston‐Price et al. 2010; Kalashnikova et al. 2018; Rocha‐Hidalgo et al. 2021) unless, like monolinguals, they receive extra‐linguistic support (Weatherhead et al. 2021). The differing experiences of labeling conventions in monolingual and bilingual children provide a compelling basis for comparison. Some studies suggest that bilingual children, due to their experience learning labeling conventions in two languages, are less likely to dismiss a new label for a known referent (Byers‐Heinlein and Werker 2009; Byers‐Heinlein 2013; Houston‐Price et al. 2010; Kalashnikova et al. 2014). Consistent with this claim, bilinguals exhibit weaker disambiguation effects, which become even less pronounced in children with a higher number of translation equivalents—words with labels in both languages (Byers‐Heinlein and Werker 2009, 2013; Byers‐Heinlein et al. 2020). A similar pattern has been observed in trilingual children, who demonstrate even lower reliance on disambiguation (Cote et al. 2022; Byers‐Heinlein and Werker 2009). This suggests that the more labels children learn for the same referent, the less they rely on disambiguation as a word‐learning strategy. However, findings remain mixed, as other studies have failed to establish a clear correlation between translation equivalents and disambiguation (Frank and Poulin‐Dubois 2002; Kalashnikova et al. 2018; Weatherhead et al. 2021). Additionally, both age and linguistic experience influence children in fast mapping tasks, but their roles and interactions are complex. The disambiguation response strengthens with age (at least in monolingual children; Bion et al. 2013; Halberda 2003), but when considered alongside children's vocabulary knowledge, age is not always a reliable predictor, likely due to its correlation with vocabulary growth (Lewis et al. 2020). One study examining 2‐ and 3‐year‐olds suggests that vocabulary size predicts stronger disambiguation, and manipulating children's experience with the familiar word‐object mapping used in the task enhances this effect, supporting the idea that language experience plays a causal role in disambiguation (Lewis et al. 2020). To this end, vocabulary size consistently predicts stronger disambiguation and better retention of new word‐object mappings in monolingual toddlers, though this is not the case with bilinguals (Bion et al. 2013; Kalashnikova et al. 2018; Weatherhead et al. 2021). This may reflect bilinguals’ divided exposure across two languages, resulting in smaller vocabularies in each language and more translation equivalents overall. Bilingual toddlers may also rely more heavily on social‐pragmatic cues or other learning strategies rather than disambiguation alone, reflecting the unique cognitive demands of managing two linguistic systems (e.g., Kandhadai et al. 2014; Weatherhead et al. 2021). The variability in findings across both monolingual and bilingual children highlights the complexity of early word learning and the multiple factors influencing successful acquisition. Rather than supporting a single, universal model of language development, the literature suggests that children's individual linguistic experiences (such as age, exposure, vocabulary knowledge) interact in complex ways to shape both behavior and neural responses. By measuring these variables alongside behavior and brain activity, we can begin to disentangle the complex interplay between language experience and lexical development. This approach will help clarify the mechanisms underlying early word learning and provide a more nuanced understanding of how diverse language experiences shape the developing brain. Present Study In the present study, we investigated whether the lateralization of the functional language network during lexical acquisition differs among English‐learning monolingual and bilingual children. We used a fast mapping task to determine the brain regions that support recognition of familiar words, referent selection via disambiguation, and subsequent retention of words learned through disambiguation. Given that language lateralization increases with age and proficiency, prior findings suggest that the development of a left‐lateralized lexical network may be linked to successful word recognition and word learning in early childhood. We therefore hypothesized that brain activity in left‐lateralized temporal and frontal regions would be associated with efficiency during disambiguated and retained novel word–referent mappings, similar to adults who successfully encode new words (Yang et al. 2015). In contrast, lower accuracy in Disambiguation or Retention trials might recruit more bilateral or widespread frontal and temporal regions, reflecting greater cognitive effort and less automated processing. Specifically, brain activity may differ among bilinguals and monolingual children as per their performance in the fast mapping task. We used functional near‐infrared spectroscopy (fNIRS) to measure brain activity in frontal and temporal cortical regions. Children between 27 and 32 months of age participated in the study. This age range was chosen given evidence suggesting that 24‐month‐olds regularly struggle to retain words in these studies, whereas 36‐month‐olds regularly succeed (in the absence of ostensive cues; Bion et al. 2013; Horst and Samuelson 2008; Weatherhead et al. 2021). Behaviorally, we hypothesized that monolingual toddlers would show evidence of both disambiguation and retention. Bilingual children, by contrast, might show a weaker disambiguation response (mapping a novel word to a novel object in the presence of a familiar one), and therefore may also show weaker retention of these mappings. In addition, we assessed how age, language exposure, vocabulary size, and the number of translation equivalents relate to performance, as these variables predict indirect word learning in prior work (e.g., Bion et al. 2013; Byers‐Heinlein and Werker 2009, 2013; Byers‐Heinlein et al. 2020; Lewis et al. 2020). Regarding brain activity, we predicted that differences between monolingual and bilingual may emerge in left frontal and temporal regions during successful word–referent selection, regardless of trial type (Familiar and Disambiguation). An alternative possibility is that bilingual and monolingual toddlers may show distinct patterns, potentially linked to their looking behavior in referent selection and retention. Because bilingual toddlers at this age often fail to show robust disambiguation or retention (Weatherhead et al. 2021), their neural responses might also be more diffuse or bilateral, reflecting greater cognitive effort. Method Participants Participants were growing up in Vancouver, British Columbia (Canada), a primarily English‐speaking society and were tested in English. Participant criteria included: (a) meeting either the monolingual criteria (i.e., exposure to a primary language for at least 90% of their lifetime) or the bilingual criteria (i.e., daily exposure from birth to another language other than English and to English for at least 30% of that time) as recommended by prior bilingual developmental research (Byers‐Heinlein et al. 2019), (b) no exposure to a third language, and (c) full term birth (gestation ≥37 weeks) with no history of significant hearing, vision or health problems (including amblyopia, strabismus, and hearing loss). From these criteria, a total of 87 toddlers participated. Of these, 27 children (21 bilinguals and 6 monolinguals) were tested but excluded from analysis due to fussiness and failure to complete task (no fNIRS and eye tracking data is available; 3 monolinguals, 12 bilinguals), refusal to wear the fNIRS cap (no fNIRS data is available; 3 monolinguals, 7 bilinguals) or the fNIRS cap moved positions throughout the task (unreliable fNIRS data; 1 bilingual), and experimenter error (no fNIRS and eye tracking data is available; 1 bilingual). Results include data from 60 children, of which there is fNIRS data available. Of these participants, 32 children were being raised monolingual (Mage = 30 months 20 days; range 28 months 10 days to 32 months 15 days; 19 females, 13 males) and 28 children were being raised bilingual (Mage = 30 months 10 days; range 27 months 21 days to 32 months 13 days; 15 females, 13 males) from birth. Based on previous behavioral work, we aimed for a final sample of approximately 30 children per language group (Bion et al. 2013; Kalashnikova et al. 2018; Weatherhead et al. 2021). Data collection was halted due to the COVID‐19 pandemic. Families were recruited from the Early Development Research Group (EDRG) database at the University of British Columbia (in Vancouver, B.C.) at the time of the child's birth, and resided near BLIND city at the time of testing. The study was approved by the University of British Columbia's ethical review boards and carried out in accordance with the Declaration of Helsinki and APA ethical standards; primary caretakers completed an informed consent form prior to families taking part in the study. Using a language background and exposure questionnaire (see below for details), the proportion of English exposure relative to exposure to other languages was calculated. The monolinguals were exposed to English 97.72% of the time (range: 90%‐100%). Only one child was reported to have any consistent exposure to another language (10% of the time; Mandarin). The bilinguals were on average exposed to 57.68% English (range: 30%–70%). The non‐English languages reported in the bilingual sample came from a diverse set of 12 languages: Bengali (1), Cantonese (6), Czech (1), French (3), Japanese (2), Korean (1), Mandarin (2), Punjabi (1), Russian (2), Spanish (6), Tagalog (2), and Tamil (1). Analyses did not reveal differences between the language groups in sex (χ 2(1, 60) = 0.20, p = 0.65), age (t(58) = 0.94, p = 0.35), parents’ education (χ 2(5, 120) = 8.89, p = 0.11), and total number of words produced in their languages (i.e., combined vocabulary in both languages for bilinguals when comparing to monolinguals; t(56) = 0.15, p = 0.88). In comparison to monolingual children, bilingual children were significantly less exposed to English (t(58) = 16.62, p < 0.001) and produced less number of words in English (t(58) = 3.73, p < 0.001). See Table 1 for descriptive statistics on demographic details and language background information. TABLE 1Demographic information of the sample (proportion) and descriptives of children's productive language assessments, average (standard deviation). Monolinguals (n = 32) Bilinguals (n = 28) Demographics: Parents’ education a High school education1.56%5.36%Some college or trade degree21.88%16.07%Bachelor's degree34.38%46.43%Master's degree9.38%17.86%Doctoral/professional degree26.56%12.50%No response6.25%1.79% Demographics: Parents’ ethnic/racial descent a Mixed37.93%46.43%Asian10.34%28.57%European51.72%21.43%Other (e.g., African, Latin American)0%3.58% English exposure (%) 97.71 (3.13)55.68 (13.23) Vocabulary production English vocabulary (# of words)75 (19)55 (23)Other language vocabulary022 (27)Total vocabulary (combined)75 (19)74 (35)Translation equivalents b 026 (26)Ability to combine words (range 1–3) c 3 (0)2.78 (0.51) aProportion of parents who reported demographics for each sample, including non‐responses. bNumber of words that conceptually mean the same in English and in bilinguals’ other language. cParents’ reporting of their child's ability to combine meaning using multiple words, such as “gimme water;” 1 = not yet, 2 = sometimes, 3 = often.John Wiley & Sons, Ltd. Demographic information of the sample (proportion) and descriptives of children's productive language assessments, average (standard deviation). Monolinguals (n = 32) Bilinguals (n = 28) aProportion of parents who reported demographics for each sample, including non‐responses. bNumber of words that conceptually mean the same in English and in bilinguals’ other language. cParents’ reporting of their child's ability to combine meaning using multiple words, such as “gimme water;” 1 = not yet, 2 = sometimes, 3 = often. Procedure Children were tested in a dimly lit, sound‐attenuated room. Children sat on their parent's lap and faced a computer monitor at an approximate distance of 60 cm. Prior to placing the fNIRS cap, measurements of the participant's head size (circumference, ear to ear, nasion to inion) were taken. Next, a fNIRS neuroimaging cap (selected based on the child's head circumference) with attached optodes was placed on the participant's head. The study started with a five point calibration routine to establish each child's eye characteristics and location in relation to the eye tracker. Following successful calibration, each toddler was presented with the experiment on the monitor screen. Neuroimaging and eye tracking data was collected simultaneously. During the experiment, parents wore opaque glasses to reduce bias and were instructed to refrain from interacting with their child. After the testing session, parents completed questionnaires regarding the child's language abilities and exposure, as well as the family's language and demographic backgrounds. As a thank you for their participation, children received an honorary degree certificate, a lab newsletter, and a gift of the parent's choosing (e.g., t‐shirt, tote bag, lab branded stuffed animal) at the end of the visit. Measures Demographics Questionnaire Parents reported on their child's health, birth weight, sex, and gestational age. Parents also reported their level of education, work status, and racial or ethnic background. Vocabulary Report Parents completed the MacArthur‐Bates Communicative Development Inventory (MCDI) Words and Sentences short form in English (Fenson et al. 1994). Parents whose child had exposure to more than one language were asked to mark items that the child produces in each language using the same English form. Language Background and Exposure Questionnaire Parents completed a one‐on‐one questionnaire with the experimenter (modified from Byers‐Heinlein et al. 2019; Orena et al. 2019), in which they reported on their child's typical daily and weekly language exposure in the months following birth until the testing date; this form included information about speakers who spent significant time with the child (e.g., at home, daycare, library, playgroups). From this assessment, a final proportion of English exposure relative to other languages was calculated. Indirect Word Learning (Fast Mapping) Task Children completed a fast mapping task, in which participants learned two new labels for two novel items (note: using two items is consistent with previous fast mapping tasks assessing for retention; Bion et al. 2013; Kalashnikova et al. 2018; Weatherhead et al. 2021). The task had two phases: the Referent Selection phase was the learning portion, in which participants were presented with the unfamiliar items and labels, and a Retention phase in which participants were tested on their acquisition of the new word–referent pairs. At the start and at the end of the task, an 8‐second video of blooming flowers appeared on the screen. Following the first video, the Referent Selection phase started and it consisted of Familiar and Disambiguation trials. During both types of trials, participants viewed images of two objects (one on the left and one on the right side of the screen), one of a familiar item (e.g., a car, cup, book) and the other of a novel object (e.g., a coffee grinder‐like object); see Figure 1 for example trial. During Familiar trials, the familiar object was labeled following a carrier phrase. Familiar labels and objects included book, ball, cookie, car, cup, shoe; of which each label‐object pairing was presented twice, for a total of 12 Familiar trials. During Disambiguation trials, the novel object was labeled following a carrier phrase. Novel labels included “dofa” and “modi”; of which each label‐object pairing was presented six times, for a total of 12 Disambiguation trials. In total, children were presented with 24 trials (12 Familiar trials and 12 Disambiguation trials), all of which were randomized and counterbalanced for left/right target location, and the same label was not presented consecutively more than twice. Carrier phrases included “Look at the [label]!”, “Where is the [label]?”, and “Find the [label].” On average, the total time of the labeling event was 3 seconds. Sample trials for the Referent Selection and Retention phases of the experiment. The total length of each Referent Selection phase trial was 8 seconds. At the start of the trial, the participant viewed the objects in silence (2.5 s), which served as the pre‐labeling window. Then, the carrier sentence along with the label (familiar or novel) played (3 s). A unique NIRS trigger was sent at the start of the labeling event, specifically at the start of the carrier sentence, to mark the trial type (Familiar or Disambiguation). Silence followed the labeling event (2.5 s). The post‐labeling window began at the onset of the target label, rather than at the onset of the carrier phrase, and continued until the window reached 2.5 s. Therefore, both the pre‐labeling window and the post‐labeling window were each a total of 2500 ms (2.5 s). A proportion score was calculated for the pre‐ and post‐labeling windows (i.e., looking to the target divided by the total time spent looking at either object), then a difference score was calculated (looking to target post‐labeling minus looking to the target pre‐labeling; see Weatherhead et al. 2021), and finally, an average score. The difference score gives a measure of how toddlers’ looking time increased to the target after the label was uttered, eliminating any potential pre‐existing preferences for the objects. Recognition was inferred if the child's attention increased to the target object post‐labelling. Following the Referent Selection phase, children watched a reward video for 10 s. Then, children were presented with the Retention phase. During Retention trials, the two novel objects shown during the Referent Selection phase were presented in tandem and only one was labeled. A total of eight randomized Retention trials were presented (4 trials per word). Similar to Referent Selection trials, the label was presented alongside a carrier phrase during Retention trials. Each Retention trial was also approximately 8 s. At the start of the trial, participants viewed the two unfamiliar objects in silence (2.5 s), then the carrier sentence along with the label played (3 s; examples: “Find the Dofa!” or “Look at the Modi!”). A unique NIRS trigger was sent at the start of the labeling event to mark the Retention trial type. Lastly, silence followed the labeling event (2.5 s). Target objects’ position (i.e., left or right) were counterbalanced across trials. Similar to the Referent Selection phase, the pre‐labeling window included the 2.5 s of silence at the start of the trial and the post‐labeling window included the 2.5 s period after the onset of the label. As in the Referent Selection Phase, a difference score was calculated. The design of the Retention phase is consistent with previous fast mapping studies such as Bion et al. (2013), Kalashnikova et al. (2018), and Weatherhead et al. (2021). Retention of the new word was inferred if the child's looking increased to the correct object following labeling. Neuroimaging Design Because the study involved neuroimaging with young children (who have difficulty staying still and sustaining attention only briefly), the task incorporated jittered inter‐stimulus intervals (ISIs) to improve the estimation of the hemodynamic response for the different trial types; this method has been used by others in language research (Arredondo et al. 2019a, 2019b). We used OptSeq2 (Dale 1999), a tool for optimizing the order and timing of events using jitters to help separate overlapping hemodynamic responses across closely spaced trials. This approach reduces habituation and expectation effects because participants cannot predict which stimulus type will appear next (Dale 1999). Specifically, we entered the duration of each trial and number of conditions in the Referent Selection phase (Familiar and Disambiguation) in the program. The program set the total duration of ISIs to 1/3 of the phase length (60 s during the Referent Selection phase). Of those 60 s, the jittered ISIs were inserted between trials and ranged between 1 to 8 s. We note that jitters of 1 s length were those that separated trials of the same condition. See Supplementary Materials for the trial order and ISI timing that we followed using the OptSeq2 output. Similarly, for the Retention phase, the program set the total duration of ISIs to ⅓ of the phase length (23 s) and jittered ISIs ranged between 1 to 8 s; see Supplementary Materials for OptSeq2 output. Participants were assigned to one of two pseudo‐randomized orders. The Referent Selection phase lasted approximately 4.5–5 min; the Retention phase lasted approximately 1.5–2 min; and the full task lasted 6–7 min. Stimuli The two images of novel objects were extracted from the TarrLab Object DataBank (Tarr, n.d.), with the criteria that they did not resemble any familiar objects in our stimuli or any other likely known objects at this age. All objects were controlled for size and perceptual similarity. Arrays of two objects were presented on a grey background. Arrays consisted of one familiar object and one novel object (familiar trials and disambiguation trials), and the two novel objects from the Disambiguation trials (retention trials only); see Figure 1. Arrays were counterbalanced such that each object, and each object type (familiar vs. novel), appeared on each side of the screen in equal proportions. Audio recordings of the labeling phrases were recorded in infant‐directed speech by a female native speaker of Canadian English producing the eight target words, two novel (dofa, modi) and six familiar (cup, ball, shoe, car, book, cookie) words in three carrier phrases (“where is the [target word]?”, “look at the [target word]” and “find the [target word]”). The audio stimuli were equated in amplitude and volume. Familiar words were selected based on the WordBank database (Words & Sentences Form; Frank et al. 2016), all words were reported to be spoken by 94% or more of children at 30 months old. In addition, all participants were reported as producing the familiar words. Children saw all six familiar objects during familiar and disambiguation trials. Stimuli were recorded in a sound‐treated room and were later equated for amplitude in Praat (Paul and David 2020). Apparatus Eye Tracking Eye gaze data was collected using a SMI Red250Mobile eye tracking system that sampled at a rate of 60 Hz. The task and areas of interest were programmed using SMI Experiment Suite 360 software on a Dell Precision laptop. The task was presented to the participants on a 22″ LCD color display monitor. Auditory stimuli were played at approximately 65 dB via a computer speaker located behind the monitor screen. A Sony Handycam camcorder was placed on top of the monitor screen and recorded the child's behavior in real time and monitor behavior to the screen. fNIRS Neuroimaging Method and Procedure We used the LED‐based NIRScout 16×16 functional Near‐Infrared Spectroscopy (fNIRS) system by NIRx Medical Technologies LLC with 760 and 850 nm wavelengths. The data were sampled at 3.91 Hz. Three black polyester‐cloth EasyCap head caps (sizes: 48, 50, and 52 cm) were available for data collection and one was chosen based on each individual child's head circumference. The head cap held the fNIRS probes during data collection. The probes included eight near‐infrared lights and eight detectors per hemisphere yielding a total of 40 data channels (20 per hemisphere). The source and detector channels were spaced 2.5 cm apart. Sources and detectors were mounted onto slit openings in the cap prior to placing them on the child's head for data collection. The cap was applied consistently on each child using the international 10‐10 transcranial system positioning (Jurcak et al. 2007) for the following points: Fz, FpZ, Cz, Oz, and T7/8. The probe configuration was designed to measure hemoglobin change (a correlated measure of brain activity) in bilateral frontal and temporal regions (see Figure 2 and Table 2). Specifically, this montage was designed to measure cortical activity in “classic language regions” for semantic processing, including superior and middle temporal gyrus, inferior frontal gyrus, and sensory/motor regions (Skeide and Friederici 2016; Weiss‐Croft and Baldeweg 2015). In addition, we measured both hemispheres due to evidence suggesting that lateralization of the language network develops with increasing age and earlier age of acquisition (Holland et al. 2007; Leung et al. 2026). In order to visualize and estimate the brain regions maximally covered by the channels, we estimated approximate MNI brain coordinates using AtlasViewer (Aasted et al. 2015), a MATLAB‐based graphical user interface. We employed digitized points to register the probe's geometric structure to a standardized brain template and obtain standardized anatomical locations for each optode (see Figure 2 and Table 2). NIRS probe configurations and their placement on the left hemisphere. Red circles = sources/emitters of near‐infrared light, blue circles = detectors, yellow lines = channels connecting sources and detectors. TABLE 2Approximate brain regions for the fNIRS probe configuration, for each mirrored channel across the left and right hemisphere.Source, DetectorChannelLeft Hem.Right Hem.Approximate Brain Region11, 19, 9Superior frontal gyrus22, 110, 9Superior frontal gyrus33, 111, 9Middle frontal gyrus43, 211, 10Inferior frontal gyrus53, 311, 11Middle frontal gyrus64, 112, 9Superior frontal and middle frontal gyri74, 312, 11Middle frontal gyrus84, 512, 13Precentral gyrus95, 213, 10Inferior frontal and (anterior) superior temporal gyri105, 313, 11Inferior frontal and (anterior) superior temporal gyri115, 413, 12(anterior) superior temporal and middle temporal gyri126, 314, 11Middle frontal gyrus, precentral region136, 414, 12Superior temporal and middle temporal gyri146, 514, 13Postcentral156, 614, 14Postcentral, inferior parietal region167, 415, 12Middle temporal gyrus177, 615, 14Superior Temporal gyrus, Inferior Parietal region187, 715, 15(posterior) Superior temporal gyrus198, 716, 15(posterior) Middle temporal gyrus208, 816, 16(posterior) Inferior temporal gyrusJohn Wiley & Sons, Ltd. Approximate brain regions for the fNIRS probe configuration, for each mirrored channel across the left and right hemisphere. Analysis Looking Behavior Children's fixations to the objects (center left, center right) were recorded during the task. Eye tracking recording began at the onset of each trial. All eye tracking data were processed using the SMI BeGaze software. Trials in which there were no eye tracking data during either the pre‐labeling or post‐labeling phase were excluded from gaze coding analyses. This resulted in the exclusion of 5.40% of Familiar trials, 5.71% of Disambiguation trials, and 12.50% of Retention trials. Functional Near‐Infrared Spectroscopy (fNIRS) We used the NIRS Brain AnalyzIR Toolbox (Santosa et al. 2018), an open source MATLAB‐based package for fNIRS data analysis, and Homer3 (Huppert et al., 2009) for visualization of hemodynamic responses. First, scripts were edited so that the duration of trials was set at 5.5 s, this timing included both the labeling period (3 s) and the post‐labeling period (2.5 s). Next, the raw time course data was converted into units of optical density change (OD). The OD data were then converted into hemoglobin concentration data using the modified Beer‐Lambert law, yielding oxygenated (HbO) and deoxygenated hemoglobin (HbR) values. In order to correct for motion artifacts and serial correlations, we carried out a pre‐whitening filter on participants’ hemoglobin data time series. At the first‐level, each participant's hemoglobin concentration data was analyzed using an autoregressive iterative least squares (AR‐IRLS) algorithm, which has shown to reduce false positive rates (i.e., type I errors) in fNIRS data from 3–5 years‐old children (Barker et al. 2013). The first‐level general linear model (GLM) analysis estimated beta values, which are indices of signal change in the brain response, for each channel, each participant, and each trial type (Familiar, Disambiguation, Retention) relative to baseline. Second‐level group analyses were conducted using a linear mixed‐effects model (LMEM). We specified a model that included Condition (Familiar, Disambiguation, Retention), Group (monolingual, bilingual), and their interaction as fixed effects, with Age included in expanded candidate models. Participants were treated as a random intercept, and hemoglobin beta values (HbO and HbR) served as the dependent variable. We then used Akaike Information Criterion (AIC) to compare nested models and select the best‐fitting LMEM. The candidate model set included: (i) Group x Condition, and (ii) Group x Condition + Age. All statistical analyses were evaluated at a False Discovery Rate (FDR) threshold correction of p = 0.05 (Benjamini and Hochberg 1995). The FDR correction was applied across channels, and in accordance with the NIRS Brain AnalyzIR Toolbox. Results Looking Behavior Performance See Figure 3. During Familiar trials , bilingual and monolingual children's amount of looking time increased to the correct referent following labeling (relative to pre‐labeling) (bilingual: t(27) = 5.00, p < 0.001, d = 0.95; and monolingual: t(27) = 4.85, p < 0.001, d = 0.92). However, during Disambiguation trials , amount of looking time increased to the novel object following labeling (relative to pre‐labeling) only for monolingual toddlers (t(27) = 2.58, p = 0.016, d = 0.49); bilinguals’ looking time did not increase (t(27) = 1.06, p = 0.30, d = 0.20). During Retention trials , looking behavior to the referent did not increase following labeling (relative to pre‐labeling) for either monolingual or bilingual toddlers’ (t(25) = 1.11, p = 0.28 and t(27) = −0.08, p = 0.94 respectively). Looking time is presented as a difference score, calculated as the proportion of looking to the target post‐labeling minus pre‐labeling. Performance in the Fast Mapping task is shown in grey bars for monolinguals and striped bars for bilinguals. Error bars represent standard error. *p < 0.05 **p < 0.01, ***p < 0.001 indicates the result of a one‐sample two‐tailed t‐test against chance (set at 0). Pearson correlations between looking behaviors for each condition, vocabulary and age were not significant for the monolingual group (see Table 3). Within the bilingual data, children's English vocabulary increased with greater English exposure; r(26) = 0.46, p = 0.013. In addition, bilinguals’ looking behavior towards the novel object during disambiguation trials increased post‐labeling with increasing age; r(26) = 0.49, p = 0.009 (Figure 4). Aside from significant correlations between the vocabulary variables, there were no other significant correlations; see Table 4. TABLE 3Pearson correlations among age, English vocabulary, and looking behavior across experimental conditions in the monolingual sample.Variable12341. Age in months—2. English vocabulary0.09—3. Familiar trials a 0.12−0.09—4. Disambiguation trials a −0.080.05−0.13—5. Retention trials a 0.13−0.04−0.090.20 aValues for familiar, disambiguation, and retention trials represent the proportion of total looking time directed toward the target referent. No correlations were statistically significant.John Wiley & Sons, Ltd. Pearson correlations among age, English vocabulary, and looking behavior across experimental conditions in the monolingual sample. aValues for familiar, disambiguation, and retention trials represent the proportion of total looking time directed toward the target referent. No correlations were statistically significant. Scatterplot showing bilinguals’ looking time increases post‐labeling towards the novel object (during disambiguation trials) with increasing age. TABLE 4Pearson correlations among age, vocabulary measures, English exposure, and looking behavior across experimental conditions in the bilingual sample.Variable123456781. Age in months—2. English vocabulary0.11—3. Combined vocabulary b 0.250.79 *** —4. Translation equivalents0.030.64 *** 0.61 ** —5. English exposure−0.090.46* 0.41* −0.03—6. Other language vocabulary0.240.070.85 *** 0.250.19—7. Familiar trials a −0.32−0.22−0.08−0.14−0.030.14—8. Disambiguation trials a 0.49 ** 0.110.22−0.020.070.250.13—9. Retention trials a −0.00−0.11−0.00−0.120.18−0.02−0.070.27 aFamiliar, disambiguation, and retention trial scores represent difference scores calculated as the proportion of looking time directed toward the target object after labeling minus the proportion before labeling. bCombined vocabulary represents the sum of words produced in English and all other languages.*p < .05, **p < .01, ***p < .001.John Wiley & Sons, Ltd. Pearson correlations among age, vocabulary measures, English exposure, and looking behavior across experimental conditions in the bilingual sample. aFamiliar, disambiguation, and retention trial scores represent difference scores calculated as the proportion of looking time directed toward the target object after labeling minus the proportion before labeling. bCombined vocabulary represents the sum of words produced in English and all other languages. *p < .05, **p < .01, ***p < .001. Functional NIRS Data Results The best‐fitting omnibus LMEM was selected based on the lowest AIC, which in this case included the Group x Condition interaction (average AIC = 66,929; average AIC for Group x Condition + Age = 67,090). Group X Condition Interaction Oxygenated Hemoglobin (HbO) The analysis identified significant group differences in one right‐hemisphere channel (Channels 13, Source 14, Detector 12) during Disambiguation trials; which approximately covered right middle temporal and superior temporal gyri. Contrasts showed increasing oxygenated hemoglobin during Disambiguation trials for bilinguals (b = 5.99, SE = 1.36, t = 4.40, df = 174, p < 0.001, FDR‐corrected < .001) than monolinguals (b = 0.31, SE = 1.16, t = 0.27, df = 174, p = 0.79, FDR‐corrected = 0.89); b = 5.68, SE = 1.79, t = 3.17, df = 174, p = 0.001, FDR‐corrected = 0.014. No other Group x Condition contrasts were significant for HbO. These results suggest that differences between the groups (monolinguals and bilinguals) during a word learning task may emerge in the right hemisphere and on the conditions for mapping novel label‐object pairings. Indeed, Disambiguation is the condition that showed differences in looking behavior between bilinguals and monolingual children. See Figure 5 for visualizations of bilinguals’ and monolinguals’ hemodynamic response with minimal preprocessing as well as with prewhitening filtering. Because preprocessing influenced the shape and variability of the visualized responses, minimally processed and prewhitened time courses are presented together for transparency; statistical inferences are based on the prespecified AR‐IRLS models. Hemodynamic response functions in significant HbO right‐hemisphere channel among bilingual and monolingual toddlers. Group‐averaged changes in oxygenated (HbO; red) and deoxygenated hemoglobin (HbR; blue) are shown for Channel 13 (Source 14–Detector 12) for bilinguals (A–B) and monolinguals (C–D) during Disambiguation of novel word–referent trials. Responses are displayed following prewhitening (A, C) and minimal preprocessing (B, D) to illustrate how filtering affects the estimated time course. Minimal processing was carried out as follows: Using Homer3, channels were pruned (enPruneChannels, range 1e‐3 to 1e7, SD range 0–45), converted to optical density (hmrIntensity2OD), motion artifacts were identified (hmrMotionArtifactByChannel, tmotion 0.5, tMask 1, STDEVthresh 50, AMPthresh 1), applied band‐pass filter (hmrBandpassFilt, hpf 0.01, lpf 0.50), converted to hemoglobin concentration (hmrOD2Conc) and averaged (hmrBlockAvg, 0 to 16). Time zero indicates the onset of the auditory prompt (e.g., “Look at the dofa!”). This channel was identified as significant based on the prewhitened HbO analyses. Deoxygenated Hemoglobin (HbR) The analysis revealed significant group differences in channel 9 (Source 5, Detector 2), which approximately covered the left inferior frontal region and left anterior superior temporal gyrus. Contrasts showed decreasing deoxygenated hemoglobin during Familiar and Disambiguation trials for bilinguals (Familiar: b = −2.52, SE = 0.93, t = −2.70, df = 174, p = 0.007, FDR‐corrected = .04; Disambiguation: b = ‐3.18, SE = 0.93, t = −3.40, df = 174, p < 0.001, FDR‐corrected = 0.007) than monolinguals (Familiar: b = 1.39, SE = 0.83, t = 1.66, df = 174, p = 0.10, FDR‐corrected = 0.24; Disambiguation: b = 0.17, SE = 0.84, t = 0.20, df = 174, p = 0.84, FDR‐corrected = 0.92); Familiar: b = −3.91, SE = 1.25, t = −3.12, df = 174, p = 0.002, FDR‐corrected = .01; Disambiguation: b = −3.35, SE = 1.26, t = −2.66, df = 174, p = 0.008, FDR‐corrected = .04. There were no other significant differences in HbR between the groups across conditions. These results suggest that differences between the groups (monolinguals and bilinguals) also emerge in the left hemisphere during a word learning task, and on the conditions for mapping familiar and novel label‐object pairing. See Figure 6 for visualizations of bilinguals’ and monolinguals’ hemodynamic response with minimal preprocessing as well as with prewhitening filtering for significant HbR channels. Hemodynamic response functions in significant HbR left‐hemisphere channels among bilingual and monolingual toddlers. Group‐averaged changes in oxygenated (HbO; red) and deoxygenated hemoglobin (HbR; blue) are shown for Channel 9 (Source 5–Detector 2) during familiar word–referent trials (A–B, E–F) and disambiguation trials (C–D, G–H). Responses are displayed following prewhitening (A, C, E, G) and minimal preprocessing (B, D, F, H) to illustrate how filtering affects the estimated time course. Minimal processing was carried out as follows: Using Homer3, channels were pruned (enPruneChannels, range 1e‐3–1e7, SD range 0–45), converted to optical density (hmrIntensity2OD), motion artifacts were identified (hmrMotionArtifactByChannel, tmotion 0.5, tMask 1, STDEVthresh 50, AMPthresh 1), applied band‐pass filter (hmrBandpassFilt, hpf 0.01, lpf 0.50), converted to hemoglobin concentration (hmrOD2Conc) and averaged (hmrBlockAvg, 0–16). Time zero indicates the onset of the auditory prompt (e.g., “Look at the dofa!”). This channel was identified as significant based on the prewhitened HbR analyses. Within‐Group Comparisons Novel‐Referent Mapping (Disambiguation > Familiar) Monolingual children did not show significant differences in brain activity for this contrast (see Table 6). In contrast, bilingual children showed greater brain activity for disambiguation trials than familiar trials in two channels in the right hemisphere covering near superior and middle temporal regions (see Table 5). HbR results are generally similar to HbO; see Supplementary Materials for HbR tables. TABLE 5Significant pairwise condition differences in oxygenated hemoglobin (HbO) among bilingual toddlers.SourceDetectorHem.BetaSET‐value p FDR Relative power Disambiguation > Familiar 1412R5.361.533.510.0010.0050.241512R7.141.803.96<0.0010.0010.20 Familiar > Disambiguation None Retention > Familiar 45L8.331.575.29<0.001<0.0010.2352L7.842.223.530.0010.0050.1653L3.481.292.710.0070.0370.2863L12.223.323.68<0.0010.0030.1187L9.161.416.49<0.001<0.0010.2699R4.711.253.78<0.0010.0030.29109R6.741.644.12<0.0010.0010.22119R4.801.084.43<0.001<0.0010.341111R8.382.064.07<0.0010.0010.18129R5.001.673.000.0030.0190.221213R9.301.825.10<0.001<0.0010.201312R8.332.792.990.0030.0190.131411R11.153.603.100.0020.0140.10 Familiar > Retention None Retention > Disambiguation 45L6.101.583.86<0.0010.0020.2352L9.242.244.12<0.0010.0010.1687L4.681.423.300.0010.0090.26119R2.991.092.750.0070.0340.341213R5.631.833.080.0020.0150.20 Disambiguation > Retention None Note: Positive beta values indicate greater HbO activity in the first condition relative to the second. Only channels that remained significant after false discovery rate (FDR) correction are reported.Abbreviations: Hem, hemisphere; L, left; R, right; SE, standard error.John Wiley & Sons, Ltd. Significant pairwise condition differences in oxygenated hemoglobin (HbO) among bilingual toddlers. Familiar > Disambiguation Retention > Familiar Familiar > Retention Retention > Disambiguation Disambiguation > Retention Note: Positive beta values indicate greater HbO activity in the first condition relative to the second. Only channels that remained significant after false discovery rate (FDR) correction are reported. Abbreviations: Hem, hemisphere; L, left; R, right; SE, standard error. Novel‐Referent Retention (Retention > Familiar, Retention > Disambiguation) Monolingual children showed greater activity for retention trials than familiar trials in four channels in the left hemisphere, approximately covering the superior and middle frontal gyri, as well as superior and middle temporal regions, and two in the right hemisphere covering the superior frontal gyrus (see Table 6). In contrast, bilinguals showed broad activity across the left and right hemisphere covering multiple regions in frontal and temporal lobes (see Table 5). TABLE 6Significant pairwise condition differences in oxygenated hemoglobin (HbO) among monolingual toddlers.SourceDetectorHemBetaSET‐value p FDR Relative Power Disambiguation > Familiar None Familiar > Disambiguation None Retention > Familiar 41L4.621.572.940.0040.0210.2343L4.201.542.720.0070.0360.2454L10.052.374.24<0.0010.0010.1564L4.101.203.410.0010.0060.3099R5.001.423.520.0010.0050.26109R4.501.502.990.0030.0190.24 Familiar > Retention None Retention > Disambiguation 54L7.552.383.170.0020.0130.15 Disambiguation > Retention None Note: Positive beta values indicate greater HbO activity in the first condition relative to the second. Only channels that remained significant after false discovery rate (FDR) correction are reported.Abbreviations: Hem, hemisphere; L, left; R, right; SE, standard error.John Wiley & Sons, Ltd. Significant pairwise condition differences in oxygenated hemoglobin (HbO) among monolingual toddlers. Disambiguation > Familiar Familiar > Disambiguation Retention > Familiar Familiar > Retention Retention > Disambiguation Disambiguation > Retention Note: Positive beta values indicate greater HbO activity in the first condition relative to the second. Only channels that remained significant after false discovery rate (FDR) correction are reported. Abbreviations: Hem, hemisphere; L, left; R, right; SE, standard error. Known‐Referent Mapping (Familiar > Retention and Familiar>Disambiguation) Monolingual and bilingual children did not show significant differences in brain activity for these contrasts (see Table 6). HbR results are generally similar to HbO; see Supplementary Materials for HbR tables. Discussion The present study examined whether English‐learning monolingual and bilingual toddlers differ in the functional development of the language network during lexical acquisition. Guided by dual‐stream models of language processing (Hickok and Poeppel 2004, 2007; Friederici 2011, 2012, 2015; Saur et al. 2008), we investigated whether differences emerge in language regions—specifically ventral regions which are associated with efficient, adult‐like sound‐to‐meaning mapping—during successful word learning. Using an intermodal preferential looking paradigm, children identified familiar referents, mapped novel labels onto unfamiliar objects, and later retrieved those novel label‐referent pairings during a retention phase. We predicted differences may emerge in left frontal and temporal regions during successful word–referent mapping. Alternatively, bilingual and monolingual toddlers may show distinct patterns related to their looking behavior during referent selection and retention. Because bilingual toddlers often show weaker disambiguation of novel labels (Byers‐Heinlein and Werker 2009; Byers‐Heinlein 2013), we further hypothesized that bilingual toddlers might recruit broader, potentially bilateral regions reflecting increased cognitive effort or less efficient lexical access, than monolinguals. Three main findings emerged. First, both monolingual and bilingual toddlers robustly recognized familiar words; however, monolingual toddlers successfully mapped novel labels onto novel referents, whereas bilingual toddlers did not. Second, neither bilingual nor monolingual toddlers showed reliable retention of the newly learned word–referent mappings. Third, group differences emerged during novel word–referent mapping: compared with monolingual toddlers, bilingual toddlers showed increases of HbO in channels associated with right temporal regions during Disambiguation trials and greater decreases of HbR in channels associated with left inferior frontal and anterior temporal regions during Familiar and Disambiguation trials. This pattern suggests more widespread recruitment during word mapping in bilingual toddlers. Together, these findings partially support our predictions: less successful novel word mapping is associated with more diffuse or broader bilateral engagement, possibly reflecting increased cognitive effort or less efficient lexical access. Children's Word‐Referent Selection Abilities and Associated Brain Responses Children's looking patterns were largely consistent with prior literature. Monolingual toddlers displayed the expected disambiguation response and a trend toward retention, whereas bilingual toddlers did not reliably map novel labels to novel referents. These findings are consistent with studies showing that word–referent mapping is robust by 24–30 months for bilingual children, whereas retention remains fragile in the absence of ostensive and extra‐linguistic cues (Bion et al. 2013; Byers‐Heinlein and Werker 2009; Byers‐Heinlein 2013; Kalashnikova et al. 2015, 2018; Rocha‐Hidalgo et al. 2021; Weatherhead et al. 2021). One notable difference from prior studies (Davidson et al. 1997; Davidson and Tell 2005; Bialystok et al. 2010) is that bilinguals in our study were more likely to show a disambiguation response with increasing age. This pattern is consistent with recent findings by Rochanavibhata et al. (2022), who reported age‐related increases in bilingual toddlers’ reliance on disambiguation. Together, these results suggest that by the end of the third year of life, some bilingual toddlers may begin to use disambiguation, although this developmental shift may occur later or more gradually than in monolingual peers. Neural Signatures of Word‐Referent Selection Although both groups successfully recognized familiar word–referent pairings, interaction analyses revealed that the groups differed in their hemodynamic responses during Familiar trials, with bilingual toddlers showing a greater decrease in HbR than monolingual toddlers in a channel approximating the left inferior frontal and anterior temporal regions. This pattern supports the hypothesis that left‐lateralized regions are engaged during successful lexical access and semantic integration (Brauer and Friederici 2007; Skeide et al. 2014), and that a left‐lateralized language network may emerge with increasing age and as children gain experience with language and its lexicon. In addition, interaction analyses did not identify any channels in which monolingual toddlers showed greater condition‐related hemodynamic responses than bilingual toddlers. One possibility is that neural differences in monolingual toddlers, if present, may involve regions not captured by the current fNIRS montage or by fNIRS more generally, including deeper structures such as the hippocampus, which lie beyond the penetration depth of near‐infrared light. The clearest group difference emerged during Disambiguation trials. We hypothesized that bilingual and monolingual toddlers may show distinct activation patterns related to their looking behavior during referent selection and retention. Because bilingual toddlers often show weaker disambiguation of novel labels (Byers‐Heinlein and Werker 2009; Byers‐Heinlein 2013), we further predicted that bilingual toddlers might recruit broader, potentially bilateral cortical regions, reflecting increased cognitive effort or less efficient lexical access. The hypothesis was supported. During interaction and within‐group analyses for novel word–referent mapping, bilingual toddlers showed increases in HbO across right‐hemisphere frontal and temporal cortices than monolinguals. Thus, we speculate that this right‐hemisphere pattern may reflect more effortful processing. This interpretation is consistent with research on adult L2 learners showing that less successful word learning is often accompanied by broader, less efficient bilateral or right‐hemisphere engagement (Yang et al. 2015). The HbR interaction findings in left‐hemisphere regions further supports the possibility of bilateral engagement. Overall, hemoglobin patterns parallel the behavioral disambiguation results and suggest that the difficulty accessing lexico‐semantic representations is associated with bilateral and especially right‐hemisphere, cortical engagement, which may reflect increased cognitive demands or less efficient processing. Lastly, both groups struggled to retain previously learned novel label‐referent pairings and both showed greater bilateral activity during Retention trials (Retention > Familiar, Retention > Disambiguation). This finding further supports the interpretation that right‐hemisphere activity may reflect more effortful processing. In summary, the present study provides converging behavioral and neuroimaging evidence that right‐lateralized patterns are consistent with children's difficulty mapping and right‐lateralized engagement may serve as an early neural marker of difficulty mapping novel words. Vocabulary knowledge during the first three years of life is one of the strongest predictors of academic achievement and later life success (e.g., Bleses et al. 2016). Thus, identifying early neural markers that differentiate efficient and less efficient word‐learning pathways may have important implications for understanding typical and atypical language development. These results lay the groundwork for examining how young children with language delays, such as late talkers or children with developmental language disorders, engage the brain's language network during word learning, as well as how this engagement changes in response to intervention or targeted language support. Strengths, Limitations and Future Directions Strengths of this study include the integration of behavioral and neuroimaging measures within the same fast mapping paradigm, offering a mechanistic view of how toddlers link novel words to referents. Our use of toddler‐optimized fNIRS methods, channel‐wise mixed‐effects modeling, and a diverse bilingual sample provides robust insights into early lexical acquisition. Importantly, this is the first study to map the neural correlates of word mapping in bilingual toddlers. The use of fNIRS with toddlers includes several limitations for the present study. One major constraint is the limited spatial resolution (∼3 cm), as well as variability in cap positioning, both of which make it difficult to precisely localize neural activity. A related limitation is that our measurements were restricted to frontal and temporal cortical areas and therefore did not capture activity in deeper or more posterior brain areas. The study also included a relatively small number of trials, which is common in toddler research due to children's limited attention spans and low tolerance for longer testing sessions. The trial conditions also differed in the number and familiarity of the labels presented. Familiar trials included six established labels, whereas Disambiguation and Retention trials repeatedly presented two novel labels. Although these differences are inherent to the fast mapping paradigm and reflect distinct word‐learning demands, they may have influenced both looking behavior and hemodynamic responses. Thus, condition‐related differences may also reflect differences in stimulus familiarity and repetition. Future studies could more closely match the number, familiarity, and frequency of labels across conditions to disentangle these effects. Finally, the short inter‐trial intervals may have resulted in temporal overlap across hemodynamic responses. Although we used OptSeq2 (Dale 1999) to design a paradigm with jittered rest intervals, an approach previously used in language studies with school‐age children (Arredondo et al. 2019a, 2019b), it is still possible that hemodynamic responses overlapped across trials and influenced the results. To address these constraints, we analyzed the data using GLM and LMEM approaches and applied FDR corrections, which are conservative and widely accepted analytic strategies in fNIRS research. The visualized hemodynamic time courses also reveal differences: minimally processed data retained substantial variability that were attenuated following AR‐IRLS prewhitening. The minimally processed data may also reflect slow drift, motion‐related variance, serial autocorrelation, and/or overlapping responses across trials. Although prewhitening reduces these sources of structured noise and the risk of false‐positive findings, it may also attenuate genuine delayed signal components. Accordingly, the apparent return toward baseline in the prewhitened time courses should not be interpreted as a direct representation of the underlying physiological response. The precise shape and timing of the hemodynamic responses should therefore be interpreted cautiously, and the minimally processed and prewhitened plots are presented together to demonstrate the effect of preprocessing on the signal. This work is limited in its scope by variability in bilingual children's heritage language exposure and by our decision to test only the majority societal language (English). Future research should examine whether differences emerge as a function of proficiency across languages. Other factors such as acculturation, executive functions, and contexts in which children often learn new words may also influence fast mapping performance and brain activity. Despite these limitations, this study is one of the few to examine how word learning heuristics are represented in the brain during early childhood, providing foundational evidence for future work. Finally, the contexts of laboratory‐based indirect word learning tasks may be particularly demanding for young children, especially bilinguals. Toddlers often require multiple exposures to solidify new words, and the requirement to map labels across different contexts (e.g., books, toys, real‐life interactions) may not be fully captured in brief computer‐based tasks. In our study, the number of presentations per novel word (six trials for each word, 12 trials in total) may not have been sufficient for reliable disambiguation and retention. This constraint reflects an inherent trade‐off: designing a task short enough to maintain toddler engagement while still capturing reliable neural responses. Future studies may benefit from designs that allow more exposures or incorporate richer contextual cues, while maintaining feasibility for neuroimaging with young children. Conclusion The ability to learn new words quickly and efficiently is a task that young children often do seamlessly. The combined behavioral and neuroimaging results from this study indicate that the use of disambiguation as a word learning strategy is still developing in 27‐ to 32‐month‐old toddlers. Monolingual children showed evidence of mapping novel words to referents and a trend toward retaining these new label‐object pairings, whereas bilingual children did not show reliable disambiguation or retention. Differences in brain activity between the two language groups offer foundational insight into how toddlers recruit the brain's language network when learning new words, and highlight neural markers that differentiate more and less efficient lexical processing. Such signatures may ultimately inform research on developmental language disorders, late talkers, and children receiving speech‐language intervention, providing a window into both typical and atypical developmental trajectories. Author Contributions Drew Weatherhead: conceptualization, investigation, writing – original draft, methodology, validation, visualization, writing – review and editing, software, formal analysis, project administration, data curation, supervision. Maria M. Arredondo: conceptualization, investigation, writing – original draft, methodology, validation, visualization, writing – review and editing, software, formal analysis, project administration, data curation, supervision. Conflicts of Interest The authors certify no affiliations with or involvement in any organization or entity with any financial interest in the subject matter or materials discussed in this manuscript. Supporting information Supporting Information: desc70272‐supp‐0001‐SuppMat.docx