Abstract
Mutual exclusivity (ME) refers to the assumption that there are one-to-one relations between linguistic forms and their meanings. It is used as a word-learning strategy whereby children tend to map novel labels to unfamiliar rather than familiar referents. Previous research has indicated a relation between ME and vocabulary development, which could either be due to children’s developing knowledge of the labels for familiar objects, or to enhanced general word-learning skills. In this study, ME was related to receptive vocabulary for 17- to 19-month-old children in a novel paradigm where children’s familiarity with the objects and labels was controlled. It was found that infants with larger receptive vocabularies employed ME to a greater extent than infants with a smaller vocabulary size. The results indicate that ME use is more reliable in infants with larger receptive vocabulary size, and, critically, that ME gradually consolidates as an abstract word-learning strategy as infants’ linguistic experience increases.
Infants and children rely on a number of default assumptions or strategies that allow them to successfully identify the referents of novel words in ambiguous or non-ostensive naming situations. One of these assumptions is mutual exclusivity (ME), which constrains novel labels to map onto unfamiliar rather than familiar referents (Markman & Wachtel, 1988). It is a robust finding that infants (e.g. Bion, Borovsky, & Fernald, 2013; Halberda, 2003; Markman, Wasow, & Hansen, 2003), children (Halberda, 2006) and adults (Halberda, 2006; Kalashnikova, Mattock, & Monaghan, 2014; Malone, Kalashnikova, & Davis, 2015) rely on this assumption when presented with ambiguous labels.
There is, however, debate over when ME emerges during the early stages of lexical acquisition, and the relation between ME use and infants’ early lexical competence. The earliest manifestations of ME have been reported at 10 months of age (Mather & Plunkett, 2012; Pruden, Hirsh-Pasek, Golinkoff, & Hennon, 2006), but other studies have not been successful in eliciting reliable ME responses in infants even at the age of 18 months (Bion et al., 2013; Mather & Plunkett, 2010). With respect to language development, there is also variation in the extent to which ME has been shown to relate to lexical development. Markman and colleagues provided evidence for early use of ME among 16- to 24-month-old infants (Liittschwager & Markman, 1994; Markman et al., 2003), at a point where vocabulary knowledge is limited. Markman et al. (2003), for instance, showed that infants with productive vocabularies below 50 words (so prior to the vocabulary explosion) were able to use ME, suggesting that it may be available at the onset of lexical acquisition as a constraint dedicated to facilitating the process of word learning (Markman, 1990; Markman & Wachtel, 1988; Merriman & Bowman, 1989), and therefore independent from lexical knowledge or experience.
However, there is converging evidence that infants’ tendency to rely on ME in referent selection tasks is related to language development, as measured by vocabulary size (Graham, Poulin-Dubois, & Baker, 1998; Mervis & Bertrand, 1994). These studies showed individual differences among 16- to 22-month-old infants’ performance in ME tasks whereby only infants with larger vocabularies demonstrated reliable use of the assumption. In a more recent study, Bion et al. (2013) assessed ME in 18-, 24- and 30-month-old infants and showed that infants’ ME use was significantly correlated with productive vocabulary scores in the older groups. However, in their study, for the 18-month-olds, no systematic use of ME was found and vocabulary size did not correlate to ME scores. Taken together, these findings show that it may not be the case that ME use is absent or present at a certain age, but that it emerges gradually as the infant’s linguistic experience increases.
There are several possibilities for the way in which ME and vocabulary development are related. First, as reflected in a computational model of lexical acquisition (McMurray, Horst, & Samuelson, 2012), novel word recognition may benefit from experience with particular words that are familiar and that also appear in the word-learning situation. As the consolidated links between familiar labels and their referents become stronger, their links to novel referents are weakened leading to a stronger ME effect. This theoretical view of the relation between vocabulary and ME use is thus that infants’ early lexical knowledge may have a direct influence on use of ME (Kucker, McMurray, & Samuelson, 2015; McMurray et al., 2012). That is, level of familiarity with the names of objects used as distracters in the ME task affects application of ME as a strategy.
A second alternative is that vocabulary development influences learning of the word-referent naming process. Thus, the increasing amount of exposure to one-to-one correspondences in linguistic input may also play a role in the emergence of ME as a word-learning strategy. Infants acquiring more than one language show a significantly weaker ME effect compared to their monolingual counterparts (Byers-Heinlein & Werker, 2009; Houston-Price, Caloghiris, & Raviglione, 2010), and the strength of their reliance on ME is related to the number of translation equivalents in their vocabularies (Byers-Heinlein & Werker, 2013). That is, monolingual infants’ increasing reliance on ME may result from their increasing insight that a linguistic form is assigned to each referent in the environment (Mervis, Golinkoff, & Bertrand, 1994) and increasing understanding about the different word classes and their meanings (e.g. that nouns tend to refer to object categories rather than properties of objects; Graham et al., 1998; Mervis & Bertrand, 1994). This indicates that ME use may be influenced by developing abstract knowledge of the relation between words and their referents.
A third alternative is that vocabulary size and ME are not directly related, but both have a separate independent cause. Such a view would be consistent with the domain-general perspective on ME, such that ME relates to general learning of the communicative process, which has consequences both for vocabulary knowledge and ME use (Baldwin & Moses, 2001). Furthermore, general attentional biases may account for infants’ tendency to reason by exclusivity in referent selection tasks at the early stages of lexical development (Hollich et al., 2000; Horst, Samuelson, Kucker, & McMurray, 2011), and so ME may not be entirely consequential upon language learning. For instance, 10-month-old infants have demonstrated selection of novel objects as referents for novel labels as a function of a general bias towards attentionally-salient (Pruden et al., 2006) or novel objects (Mather & Plunkett, 2010). Similarly, Mather and Plunkett (2012) demonstrated that ME-like responses could be elicited based on a novelty bias even in situations when infants have not been presented with a competitor for which the label is familiar. In their experiments, 22-month-old infants were presented with a familiarised but not labelled novel object and a completely novel object (not familiarised). Upon hearing a novel label, infants selected the non-familiarised novel object (see Horst et al., 2011 for a similar finding with 24-month-old infants). Thus, honing these endogenous attentional biases may benefit early referent selection processes (Hollich et al., 2000; Houston-Price, Plunkett, & Duffy, 2006).
In the most commonly used ME paradigm, distinguishing between these accounts of the role of vocabulary in ME has not been possible because particular vocabulary knowledge and general use of ME as a strategy are conflated. In the standard ME paradigm, infants are presented with two objects, one familiar (e.g. a spoon) and one unfamiliar (e.g. a whisk) and are requested to find the referent of a novel label (e.g. where is the whisk?). This paradigm requires that the child must first retrieve the meaning of the familiar label, identify the familiar object as its referent, and then exclude this object as a potential referent for the novel label. If the complexity of these processes increases due to low familiarity with that label, the child will be less likely to apply exclusion and avoid lexical overlap (Grassmann, Schulze, & Tomasello, 2015; Merriman & Marazita, 1995).
Grassmann et al. (2015) tested 2-, 3- and 4-year-old children’s reliance on ME in a task where children’s level of familiarity with the distracter labels was manipulated, such that children were either able to produce the familiar labels spontaneously, produce the labels upon request, or children were able to comprehend but not produce the labels. They showed that, above the contribution of age, label familiarity was a significant predictor of the extent to which children relied on the ME assumption: children were most likely to exhibit ME in cases where they were highly familiar with the label of the distracter object. Therefore, it is possible that the relation between children’s size of vocabulary and the emergence of the ME assumption is mediated by the level of familiarity with the objects and their labels used as distracters in the ME experimental paradigm. Using the standard ME paradigm (i.e. familiar distracter and an unfamiliar target in the presence of a novel label), it is thus not clear whether development of the ME assumption associated with vocabulary development is facilitated by particular knowledge of the specific words used as familiar labels in the study, or whether ME develops as an abstract assumption in tandem with vocabulary development. In this case, if familiarity is controlled, it may be the case that the use of ME will not be observed among young infants.
The present study investigated whether young infants were able to rely on ME, and whether the extent of ME use would relate to individual vocabulary size in a referent selection task that does not include familiar label competitors. We investigated ME use in a group of 17- to 19-month-old infants, at the age when vocabulary tends to begin to undergo significant growth (Nazzi & Bertoncini, 2003; Regier, 2003). A novel-novel ME (Diesendruck & Markson, 2001) preferential looking paradigm was employed where the infant’s ability to use ME was assessed based on a recently established mapping. In contrast to the standard novel-familiar paradigms, infants were not presented with an object that was very familiar to them paired with a novel object. Instead, they were presented with two novel objects, one of which was previously named and the other was not, and were then asked to find a referent for a different novel label that they have not heard before. This paradigm enables us to control for the effects of familiarity of the competing label on young infants’ ability to reason by exclusion in a referent selection task.
This study aimed to determine whether ME was observable in children prior to extensive language development, or whether it emerged gradually as vocabulary developed. The ME paradigm we used enabled us to extract the effect of individual vocabulary knowledge from observations of ME, which may have obscured previous studies’ ability to detect ME prior to vocabulary development. Two predictions were constructed based on previous research. First, if ME depends on particular knowledge about individual words (Grassmann et al., 2015; McMurray et al., 2012) or general learning and attentional mechanisms (Horst et al., 2011; Mather & Plunkett, 2012), then it is predicted that the relation between vocabulary size and ME would not be observed. If, however, ME develops as an abstract principle associated with vocabulary development in accordance with abstract knowledge about word-referent mappings (Graham et al., 1998; Mervis & Bertrand, 1994), then infants should manifest reasoning by exclusion and increase their looking duration to the novel-unnamed object in response to the novel label in correspondence with their vocabulary size. That is, a significant relationship is expected between infants’ vocabulary size and their reliance on ME reasoning in the present referent selection task.
Method
Participants
Twenty-seven 17- to 19-month-old infants (20 female) participated. Their ages ranged from 533 to 600 days (M = 563.74, SD = 19.03). Two additional infants participated but were excluded from final analyses because of equipment failure. All infants were typically developing and came from monolingual English-speaking families.
Infants’ receptive vocabulary was assessed through the Oxford Communicative Development Inventory (CDI; Hamilton, Plunkett, & Schafer, 2000), an adaptation of the MacArthur–Bates CDI (Fenson et al., 1994) for use with infants raised in Britain. The mean receptive vocabulary score for the sample was 248.15 words (SD = 97.09).
Materials and apparatus
Four three-dimensional object images (approximately 6 × 6 cm) were selected from the TarrLab Object Data Bank (TarrLab, Brown University, 1996). Two objects were familiar (familiarisation trial) and two were novel (test trials). The objects were embedded in a video sequence that also included the video and audio recording of a female speaker. This was a native English speaker who presented the stimuli in infant directed speech. The video image of the speaker was included to increase infants’ attention to the presentation of the novel labels by providing them with additional social cues that could facilitate learning (Roseberry, Hirsh-Pasek, & Golinkoff, 2014; Scofield & Williams, 2009). The labels banana, cup, toma and modi were used to refer to the familiar and unfamiliar objects, respectively. An audio recording of the speaker exclaiming, ‘Look! They are nice! Wow! They are pretty!’ was recorded for the baseline phase. Each video sequence consisted of three phases: naming, baseline and test (Figure 1).

Graphical representation of the ME task.
Data were collected using a Tobii X120 (33 Hz sampling rate) eye-tracking system. Stimuli were presented through Tobii Clearview software on a 32-inch TV monitor. The infant sat on their caregiver’s lap in a quiet room free from distraction, approximately 60 cm away from the monitor, and the eye-tracker was placed under the monitor out of the child’s view. The eye-tracker was located 33 cm below the centre of the screen. When displayed on the screen, the size of the objects was 15 cm, and the distance between the objects was 30 cm. Caregivers were instructed to look away from the screen during the task. The experimenter controlled the study from a computer located out of sight of the infant. Before commencing the referent selection task, the experimenter completed a five-point calibration routine for the eye-tracker (provided by Tobii Clearview software).
Procedure
The referent selection task consisted of four trials. The first trial was the familiarisation trial where objects and labels known to the child were used in order to familiarise the infants with the computerised task and assess their ability to engage and attend to the task. This was followed by three test referent selection trials where the two objects were unfamiliar and novel labels were used to name and request the objects. All infants were presented with only one target and one distracter label, which were repeated across the three trials giving them more opportunity to establish the mapping.
All trials comprised the three phases described below. The labels banana (distracter) and cup (target) were used in the familiarisation trial, and the labels toma (distracter) and modi (target) were used in the three test trials.
Analyses of eye-tracking data
The two phases of interest for the analyses were baseline and test. Each phase included two areas of interest: the distracter object and the target object. The side of presentation of the target object (right and left side of the screen) was alternated across trials. Fixation duration at each object was recorded at 250 ms after the onset of the test label for 2000 ms. Given that infants were exposed to two presentations of the target label (once before seeing the objects and one repetition after the objects appeared on the screen), their fixations to the target object were captured at 250 ms after the second presentation of the label. Only fixations captured in the time window of 250 ms to 2000 ms after the onset of the target label were analysed following previous research using preferential looking paradigms (e.g. Swingley & Aslin, 2000; Swingley & Fernald, 2002) since fixations after the 2000 ms cut-off point cannot be reliably interpreted as a response to the auditory stimulus. These measures were converted into proportions of time that the infant spent fixating at each object out of the two possible objects (distracter or target), and then averaged across the three test trials. Fixations at target in the test phase were compared to fixations at target in the baseline phase to ensure that an increase in fixation compared to the distracter occurred in response to the target label as opposed to a visual preference for the target.
Results
In order to assess whether infants attended to the target object as a function of the target label, fixation durations at the target in baseline and test phases were compared in the familiarisation and novel label trials. In the familiarisation trial, a significantly higher proportion of fixation duration was directed to the target object in the test (M = .64,SD = .39) compared to the baseline phase (M = .34, SD = .25), t(26) = −3.421, p = .002, d = 1.34. That is, infants were engaging with the task and were capable of attending to the request from the speaker.
Then, infants’ performance in the test referent selection trials was assessed. Infants’ vocabulary scores were included to account for the effect of individual linguistic proficiency. A repeated measures ANCOVA with phase (baseline, test) as factor and CDI receptive vocabulary score as covariate showed a significant effect of phase indicating an overall ME effect, F(1, 23) = 6.749, p = .016, η2 = .227, no main effect of CDI,F(1, 23) = 2.239, p = .148, η2 = .089, but, critically, a significant phase by CDI interaction, F(1, 23) = 9.772, p = .005, η2 = .298. Infants significantly increased the duration of their fixations at the target object in response to the novel label in the test phase (M = .58, SD = .27) compared to the baseline phase (M = .54, SD = .08), in proportion to their receptive vocabulary size.
To investigate individual differences in ME use, a ME score was calculated for each infant as the difference between the proportion of fixation duration to the target in test compared to the baseline condition where a larger difference score denoted greater fixation duration at target in response to the novel label. A partial correlation analysis controlling for infants’ age showed a significant relation between infants’ ME difference scores and their receptive vocabulary size, r(22) = .540, p = .006. To further test this relation, a larger vocabulary (n = 14; Mean CDI = 322.57, SD = 65.86) and a smaller vocabulary (n = 13; Mean CDI = 168, SD = 48.02) sub-group were created based on the median split of vocabulary size scores (Median = 225). An independent-samples t-test with ME difference scores as the dependent variable demonstrated that indeed infants with larger vocabulary size (M = .14, SD = .31) showed significantly greater ME use than infants with smaller vocabulary size (M = −.07, SD = .12), t(23) = −2.140, p = .043,d = .89 (Figure 2).

ME difference scores for the smaller (n = 13) and larger vocabulary (n = 14) sub-groups (error bars represent standard error of the means).
In order to further examine differences in performance between the smaller and larger receptive vocabulary groups, infants’ fixation patterns were analysed across the time course of the analysis window in 100 ms time bins between 250 ms and 1950 ms during the test phase of the referent selection trials (Figure 3). One-sample t-test analyses were conducted to assess whether infants fixated to the target object more often than predicted by chance at each time slot (chance = .5). This was the case for infants in the larger vocabulary group at 750 ms, t(12) = 2.388, p = .034, marginally at 850 ms, t(12) = 2.062, p = .062, at 950 ms, t(13) = 2.956, p = .011, 1050 ms, t(13) = 2.188, p = .048, and 1250 ms, t(13) = 2.188, p = .048 (all other p > .05). On the other hand, infants in the smaller vocabulary group fixated to the target above levels predicted by chance only at 650 ms, t(10) = 2.834, p = .018, and 750 ms, t(10) = 3.318, p = .008 (all other p > .05). It should be noted that the values reported here have not been adjusted for multiple comparisons to avoid Type II errors consequent upon the increased variability resulting from sampling the data at 100 ms time bins for the two sub-samples in this study. Finally, independent t-test analyses were conducted to compare the proportion of fixations to the target at each time-slot for the smaller and larger vocabulary size groups, which showed that infants in the larger vocabulary group directed more fixations to the target at the 1250 ms time slot, t(23) = 2.408, p = .024 (all other p > .05). These fixation patterns further illustrate that all infants in this study were employing ME reasoning in this referent selection task, but infants with a larger receptive vocabulary size were able to use it more reliably and consistently to select the referent for the novel label.

Time course fixation patterns to the target and distracter objects by the smaller (n = 13) and larger (n = 14) vocabulary sub-groups (one-sample t-test analyses against chance performance; chance = .05; *p < .05; ^p = .062). Error bars represent standard error of the means.
Discussion
The present study revealed that ME use by 17- to 19-month-old infants was related to their receptive vocabulary size, even when knowledge of the labels within the task was controlled. Infants with larger vocabularies were more consistent in directing their looks to an unfamiliar object in response to a novel label. These results align with the view that ME is a gradually developing strategy in relation to infants’ growing vocabulary proficiency.
Our findings, therefore, fail to confirm that ME operates as an assumption that facilitates word learning at the early stages of language acquisition (Markman et al., 2003). A closer inspection of our data shows that infants in the smaller vocabulary group who were not consistently employing ME had a receptive vocabulary score of approximately 160 on the CDI, which can correspond to a receptive vocabulary size of over 200 words (Mayor & Plunkett, 2011). Thus, while these infants relied on ME to a lesser extent than their peers with larger receptive vocabulary size, they did seem to be capable of acquiring this number of words successfully, consistent with the view that ME is not likely to be a necessary precursor of early vocabulary acquisition. Instead, our findings indicate that ME use becomes reliable when infants have acquired more extensive lexical competence.
Previous studies that have demonstrated a relationship between children’s early ME use and their vocabulary size have used highly familiar distracters (e.g. Bion et al., 2013; Graham et al., 1998; Houston-Price et al., 2010), which meant that children’s tendency to reason by exclusion may have been mediated by their level of familiarity with the distracter and its label (Grassmann et al., 2015). However, the use of two unfamiliar objects in this task demonstrates that this relationship is significant even when children’s familiarity with the distracter and its label are controlled for. Exposure to objects (Fennell, 2012; Kucker & Samuelson, 2012) and labels (Swingley, 2007) used in the paradigm can facilitate encoding of novel labels in experimental tasks. Especially in the case of ME, higher familiarity with the distracter’s label can lead to higher reliance on ME due to increased competition between the labels in the vocabulary and consequently a decrease in the complexity of the task (Grassmann et al., 2015). Here we demonstrate that, in addition, higher linguistic competence manifested in larger vocabulary size also facilitates performance in this referent selection task.
In order to control for children’s familiarity with the object and label used as distracters, our paradigm presented infants with an initial novel object-novel label mapping before proceeding to the disambiguation task. That is, infants’ performance in the test phase of the task relied on their ability to first encode the distracter object-label mapping presented in the naming phase. Thus, it could be the case that infants’ overall ability to map novel labels to novel objects was also mediating their performance in this task. It is well known that infants at 18 months are successful at determining word-object associations after a limited exposure time in computerised tasks (e.g. Bion et al., 2013; Byers-Heinlein, Fennell, & Werker, 2013; Stager & Werker, 1997; Yoshida, Fennell, Swingley, & Werker, 2009), but it is still possible that some infants were more successful at establishing and retaining these initial mappings than others. Therefore, good word learners overall may have also been good at using ME. That is, the present paradigm can reveal infants’ more advanced word-learning capacities whereby instead of mapping the novel label to the most unfamiliar object in each trial, they were also relying on the ability to maintain the other novel label encoded in the naming phase (Wilkinson, Ross, & Diamond, 2003). This possibility, however, is not contradictory to our conclusions.If ME emerges as a product of more extensive word-learning experience, it is natural for it to be related to infants’ general referent selection and familiar word processing skills (McMurray et al., 2012).
The present study does not allow us to exclude the possibility that infants were relying on more general non-linguistic or pragmatic information to disambiguate the meaning of the novel labels in this task (Horst et al., 2011; Mather & Plunkett, 2012; Pruden et al., 2006). Even though both objects presented in the task were not previously familiar to the child, it is possible that infants were mapping the unfamiliar label to the most novel object presented in the task. An effort was made to equate the salience of the two novel objects, but it must be noted that infants did receive more exposure to the distracter than to the target object (i.e. infants saw the distracter in the naming, baseline and test phases, but they saw the target only in the baseline and test phases). In addition, unlike the commonly used preferential looking ME paradigm (e.g. Bion et al., 2013; Byers-Heinlein & Werker, 2009; Halberda, 2003; Houston-Price et al., 2010), the present task included a speaker who provided referential cues in the naming phase (i.e. gaze and pointing), which could have been interpreted as an intention to single out the distracter object. However, this interpretation cannot entirely account for our findings. First, the present analyses compared infants’ fixation duration at target in the baseline and test phases, ensuring that an increase in fixation duration at test would indicate a response to the novel label above any pre-existing individual preferences for the object. Second, the novelty account does not explain the significant effect of vocabulary size found in this study since this type of attentional bias can guide referent selection processes even before infants’ first birthday when their vocabulary size is very limited (Pruden et al., 2006). Therefore, our findings suggest that 17- to 19-month-old infants in this study were using ME as an abstract word-learning strategy instead, and its use was shaped by infants’ own lexical experience.
Our results add to the growing evidence for a developmental and a dynamic view of ME rather than postulating it as a lexical constraint that becomes available at a certain point in development. While very young infants may be able to resolve exclusivity-based problems by relying on endogenous learning and attentional biases, as they grow older, they recruit other sources of information such as their individual linguistic experience and understanding of the linguistic and non-linguistic input to solve the task of fast mapping (Hollich et al., 2000). This also emphasises the influence of the experimental paradigm and stimuli on infants’ ability to reliably use ME: factors such as saliency and referential cues (Hollich et al., 2000), familiarity of the competing label (Grassmann et al., 2015), phonological form of the novel label (Mather & Plunkett, 2011; Merriman & Schuster, 1991), similarity of the novel exemplar to other familiar objects (Merriman & Marazita, 1995) and number of familiar competitors (Horst, Scott, & Pollard, 2010) can all affect whether the assumption is manifested or not at a certain age.
It is well known that young infants and children reason by exclusivity to identify the referents of novel labels presented in non-ostensive communicative situations. From early on, infants can manifest this reasoning by relying on a number of endogenous attentional biases that guide them to map unfamiliar labels to the more novel or salient objects in their environment. This mapping process is also aided by lexical competition, which arises when the child is highly familiar with the distracter or distracters presented in the task. However, this study demonstrates that general learning mechanisms and familiarity with competing labels cannot account for all the early manifestations of ME. Here, infants who were more experienced language users were capable of employing ME even in a more complex situation where referent selection relied on a recently established mapping and thus competition between two novel labels and novel objects. Therefore, with increasing linguistic experience and maturation of social and communicative skills, basic learning mechanisms can be transformed into more sophisticated strategies such as ME. This can lead to a more systematic use of this strategy, which is interlinked with infants’ emerging communicative skills and increasing vocabulary knowledge.
Footnotes
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: PM was supported by the International Centre for Language and Communicative Development (LuCiD) at Lancaster University, funded by the Economic and Social Research Council (UK) (ES/L008955/1).
