Knowing a lot for one’s age: Vocabulary skill and not age is associated with anticipatory incremental sentence interpretation in children and adults

https://doi.org/10.1016/j.jecp.2012.01.005Get rights and content

Abstract

Adults can incrementally combine information from speech with astonishing speed to anticipate future words. Concurrently, a growing body of work suggests that vocabulary ability is crucially related to lexical processing skills in children. However, little is known about this relationship with predictive sentence processing in children or adults. We explore this question by comparing the degree to which an upcoming sentential theme is anticipated by combining information from a prior agent and action. 48 children, aged of 3 to 10, and 48 college-aged adults’ eye-movements were recorded as they heard a sentence (e.g., The pirate hides the treasure) in which the object referred to one of four images that included an agent-related, action-related and unrelated distractor image. Pictures were rotated so that, across all versions of the study, each picture appeared in all conditions, yielding a completely balanced within-subjects design. Adults and children quickly made use of combinatory information available at the action to generate anticipatory looks to the target object. Speed of anticipatory fixations did not vary with age. When controlling for age, individuals with higher vocabularies were faster to look to the target than those with lower vocabulary scores. Together, these results support and extend current views of incremental processing in which adults and children make use of linguistic information to continuously update their mental representation of ongoing language.

Highlights

► Anticipatory sentence interpretation is examined in children (3–10 years) and adults. ► Both groups rapidly combined agent and action cues to predict sentence final objects. ► Vocabulary but not age was associated with anticipatory fixations in both groups. ► Adults and children used similar sentence interpretation strategies. ► Results suggest important relationships between vocabulary and predictive processing.

Introduction

One of the challenging aspects of real-time spoken language comprehension is that language must be processed incrementally and at a relatively fast speed. The meaning of a sentence evolves as it unfolds, and sentential meaning cannot generally be inferred from any single word alone. In light of these challenges, it has been hypothesized that listeners use a strategy of continually generating expectancies about upcoming referents. Although this hypothesis remains controversial, over the past two decades a growing body of computational evidence suggests that prediction is a powerful mechanism for learning (Elman, 1990, Misyak et al., 2010, Rodriguez et al., 1999), and a large number of empirical studies using behavioral and neurophysiological techniques have provided support for expectancy generation in sentence comprehension in adults.

Much less is known about the role of predictive processing during childhood. There are strong suggestions that prediction may underlie children’s ability to segment continuous streams of auditory input into word-sized chunks (Estes et al., 2007, Saffran et al., 1996). Nor is it known what factors might influence individual children’s abilities to engage in expectancy generation, although one likely candidate is vocabulary size. For the purposes of this study, we were particularly interested in the relationship between vocabulary and predictive processing in sentences. Vocabulary size is highly associated with speed of comprehension in looking tasks (Fernald et al., 2006, Marchman and Fernald, 2008). Moreover, there are a number of proposals in the language development literature that highlight the relationship between lexical and grammatical development. These suggest that vocabulary knowledge may serve an important role in language tasks that require meaning and structure to be parsed and interpreted across multiple words, as in online sentence comprehension. For example, early growth of the lexicon has been argued to be the critical foundation from which basic grammar emerges (Bates and Goodman, 1997, Marchman and Bates, 1994). This proposal is supported by many observations that development of the lexicon and early morphosyntactic competence is strongly correlated, with lexical development preceding grammatical development. This relationship has been noted in cross-sectional and longitudinal studies (Dale et al., 2000, Fenson et al., 1994) and across many languages, including Icelandic (Thordardottir, Ellis Weismer, & Evans, 2002), Italian (Caselli, Casadio, & Bates, 1999), and Hebrew (Maital, Dromi, Abraham, & Bornstein, 2000), in bilingual language acquisition (Marchman, Martínez-Sussmann, & Dale, 2004), and in atypical language development (Moyle, Ellis Weismer, Evans, & Lindstrom, 2007). The idea that sentence processing may be linked to lexical knowledge is also supported by a number of grammatical theories that firmly root syntactic competence in the lexicon, including Combinatory Categorical Grammar (Steedman & Baldridge, 2006), Head-driven Phrase Structure Grammar (Pollard & Sag, 1994), and Lexical-Functional Grammar (Bresnan, 2001). Therefore, a central goal of this study was to extend our understanding of the development of anticipatory sentence processing mechanisms by directly comparing the online processing of transitive sentence comprehension in 3- to 10-year-old children with that in adults and by testing the degree to which sentence interpretation and prediction skills are associated with differences in vocabulary level.

Much of the evidence for predictive processing in sentential comprehension in adults involves inferences made from measurement of eye movements in response to language while viewing a visual scene. As objects are mentioned, visual attention is directed toward the spatial location of the object in the scene (Cooper, 1974, Tanenhaus et al., 1995), often before the word is even complete (Allopenna et al., 1998, Dahan et al., 2001). More dramatically, eye gaze may be directed toward objects even before they are mentioned (Altmann and Kamide, 1999, Altmann and Mirkovic, 2009, Kamide et al., 2003). Thus, eye gaze appears to capture moment-to-moment changes in the comprehension of language. Results from numerous studies using this technique (known as the visual world paradigm [VWP] in the adult literature), and others suggest that adults continuously update their mental representation of ongoing events in the sentence and use cues from many different sources. These include semantic features (Federmeier & Kutas, 1999), event-level expectations (Bicknell et al., 2010, Hald et al., 2007, Kamide et al., 2003, Metusalem et al., 2010), prosodic cues (Salverda, Dahan, & McQueen, 2003), phonological information (DeLong et al., 2005, VanBerkum et al., 2005), verb tense markers (Altmann & Kamide, 2007), grammatical and biological gender (Arnold et al., 2000, Lew-Williams and Fernald, 2010, Tanenhaus et al., 2000, Wicha et al., 2004), verb selectional restrictions (Altmann & Kamide, 1999), pronominal adjectives (Sedivy, Tanenhaus, Chambers, & Carlson, 1999), verb structural biases (Trueswell, Tanenhaus, & Kello, 1993), and referential restrictions (Tanenhaus et al., 1995). Crucially, this updating mechanism is proactive. It not only “reacts” and “integrates” to information it receives but also actively generates expectations for plausible sentence continuations.

A concrete example of this process is illustrated in a series of studies reported by Kamide et al. (2003), which yielded some of the earliest and most convincing support for active prediction during adult sentential comprehension. In their Study 2, participants viewed simple visual scenes containing multiple agents and objects with similar affordances, including drinkable objects such as BEER and MILK. Participants then heard sentences such as “The man will drink the beer” while their eye movements to this scene were simultaneously recorded. After hearing the verb drink but before beer, participants fixated on the image of the BEER more than the image of the MILK, indicating that their eye movements reflected more than a simple lexical association between the verb drink and the upcoming object; rather, it reflected predictions motivated by a combination of the prior agent and verb. Appropriate controls also established that the effect was not the result of lexical associations between man and beer or due to visual saliency of BEER over MILK.

This result demonstrates that expectancy generation may reflect integration from multiple cues. In this case, those cues were the agent and the verb in combination. Moreover, the integration draws on real-world knowledge regarding which objects are likely patients of a verb given the agent carrying the action denoted by the verb (for related findings, see also Bicknell et al., 2010, Hare et al., 2009, Matsuki et al., 2011).

Children also have been found to be able to integrate information from multiple sources to comprehend sentences in real time. However, children are not always sensitive to the same cues as adults. For example, children under 5 or 6 years of age seem to show difficulty in interpreting referential or extralinguistic cues to meaning in a visual scene (Kidd and Bavin, 2005, Snedeker and Trueswell, 2004, Trueswell et al., 1999). Despite these limitations, children still do use many sources of information with great skill and speed. Both young children and adults can call on their understanding of the syntactic and semantic meanings of verbs (Altmann and Kamide, 1999, Fernald et al., 2008, Nation et al., 2003), adjectives (Fernald, Thorpe, & Marchman, 2010), and article grammatical gender (Lew-Williams & Fernald, 2010) to correctly predict an upcoming referent before it is spoken.

For example, Fernald and colleagues (2008) have shown that children as young as 26 months are able to generate anticipatory looks to an object (e.g., cookie) on hearing a verb “Eat the …,” much like adults (Altmann & Kamide, 1999). Moreover, electrophysiological and neuroimaging data suggest that the neural mechanisms underlying predictive skills such as noting incongruities in label–object pairings are present by the second year of life (Friedrich and Friederici, 2004, Friedrich and Friederici, 2005, Mills et al., 1997, Travis et al., 2011). Taken together, these discoveries suggest that the skills adults use to swiftly and efficiently understand rapidly spoken sentences are in place from the earliest moments of language acquisition.

As was noted above, it seems that such integration and prediction skills are both a marker and predictor of language skill and growth, especially in relation to vocabulary. Vocabulary skill has an important relation to processing speed, as children with larger vocabularies are also quicker to comprehend spoken words. Accuracy and speed in online lexical comprehension at 25 months of age correlates with earlier measures of vocabulary growth between 12 and 24 months (Fernald et al., 2006). Even more striking, these early differences in the speed of lexical processing at 25 months have long-term consequences for language acquisition and are correlated with language outcomes 6 years later (Marchman & Fernald, 2008). Differences in language ability are also associated with more complex measures of predictive ability. For example, the speed with which 26-month-olds correctly predict the upcoming object of a verb in a sentence such as “Eat the cookie” is also associated with concurrent vocabulary size (Fernald et al., 2008).

It is well established that dramatic differences in vocabulary size between individuals can be observed from across the earliest stages of learning during infancy through adulthood (Fenson et al., 1994, Verhaeghen, 2003). Children typically begin to produce their first words at around the end of their first year, and vocabulary expands rapidly throughout childhood. However, the timeline and trajectory of this growth varies considerably (Fenson et al., 1994). Whereas some children may speak their first words at 12 months and may know as many as 550 words by 24 months, others might not begin to speak for another half year and might know only 50 words by 2 years (Fenson et al., 1994). It should be emphasized that children at both ends of this lexical learning spectrum are still considered within the “normal” range despite this tremendous initial variability.

These differences become even more sizable by the time children begin school. For instance, some preliterate 5- or 6-year-olds in first grade will have an expressive vocabulary of as many as 5000 words, whereas others may produce only half as many words (Beck & McKeown, 1991). Across the school years, the average student will learn thousands of new words per year, amounting to several new words a day (Anglin, 1993, Graves, 1986). However, vocabulary growth in school is considerably slower for children who begin with smaller vocabularies (White, Graves, & Slater, 1990).

Finally, vocabulary growth does not stop after childhood, and vocabulary size differences continue throughout the lifespan (Verhaeghen, 2003). For instance, differences in vocabulary between individuals who have attended college and those who have not can be significant. One study estimated the vocabulary knowledge between adults with a high school education and those with a college degree to differ by 5000 word families, which is a measurement of word knowledge that counts a root word plus its inflected forms and derivations as a single family (Zechmeister, Chronis, Cull, D’Anna, & Healy, 1995). Variability in vocabulary size is notable even in college students—a population that has been at least partially selected by performance on tests of linguistic proficiency in entrance exams (Martino & Hoffman, 2002).

Differences in early vocabulary size have important impacts on later language outcomes and school achievement. Early vocabulary growth not only is a building block for acquiring grammatical skills (Bates and Goodman, 1999, Marchman and Bates, 1994) but also is at least partially related to language and reaching achievement outcomes many years later (Cunningham and Stanovich, 1997, Marchman and Fernald, 2008). Vocabulary size has also been shown to associate with a number of other linguistic and cognitive abilities in school-age children, including phonological working memory (Gathercole & Baddeley, 1989), phonological awareness (Metsala, 1999), and reading comprehension (Cunningham and Stanovich, 1997, Stahl and Nagy, 2006). A particularly strong relationship between vocabulary level and reading comprehension has been noted since the 1920s (Whipple, 1925; see Nation, 2009, for a review). However, early vocabulary size alone is not enough to reliably detect which individuals will be most at risk for later language and learning disabilities, as noted by numerous investigators who have tried, and largely failed, to accurately identify children who will receive later diagnoses of language impairments solely from infant measures of vocabulary (Ellis Weismer, 2007, Paul, 1996, Rescorla, 2000, Whitehurst and Fischel, 1994). In sum, vocabulary and linguistic processing speed seem to have an important relationship during early childhood, although it is likely that the relationship between the two and later outcomes may be more complex.

Taken together, these findings motivate a need for a deeper and more detailed understanding of children’s predictive abilities in language comprehension and for a better understanding of the relationship of this ability to other markers of language skill such as vocabulary. The goal of the current study was to address these gaps by asking three specific questions. First, we asked whether the results of prior studies that find evidence for linguistic prediction across one or two words scale up to more complex language processing tasks that use more challenging stimuli. Advancing our understanding of how words are understood in sentences that require children to calculate more complex multiword contingencies is necessary if we are to fully understand how humans can swiftly and efficiently understand complex and novel multiword utterances, which is arguably one of the defining characteristics of language (Hockett, 1960). Second, to better understand the developmental trajectory of this skill, we examined the development of this skill across a wide range of ages. In the current study, we included participants from 3–10 years of age as well as adults. Finally, because vocabulary appears to be an important marker of processing speed in young children, we asked whether this relationship also exists in performance on this experimental task.

To answer these questions, we built on the task used by Kamide and colleagues (2003) described above. Several modifications were made to this task, and extensive norming was conducted to ensure that both adults and children could comprehend the images and sentences. One important adaptation involved the layout of the visual display. Rather than provide visual scenes that included the sentential agents as was done in the original study by Kamide and colleagues, we used a four-alternative forced-choice display in which participants were asked to indicate which image corresponded to the sentence-final target. The purpose of this modification was twofold. First, it served to reduce the visual complexity of the visual scenes, thereby increasing the likelihood that we would get clean and motivated looks to the target items from the children. Second, it allowed us to compare the relative activation and anticipation of candidate target meanings in response to the agent and action as the sentence unfolds by controlling the relationship of the distractor items to words in the sentence.

Each visual scene consisted of the target and three types of distractors: an object that was associated with the agent (agent-related), an object that was associated with the action (action-related), and an object that was unassociated with either the agent or action (unrelated). Because the target was by definition associated with both the agent and action, this meant that two objects were associated with the action (target and agent-related pictures) and two objects were associated with the agent (target and action-related pictures). Only the target was appropriate given both the agent and action.

This design was chosen because it also allowed us to study in some detail the potential comprehension strategies that children and adults might employ as they interpreted the sentence in real time. One type of possible strategy would involve a staged elimination of potential sentential targets as the sentence unfolds. Consider the sample sentence “The pirate hides the treasure,” with a visual display consisting of TREASURE (target), a (PIRATE) SHIP (agent-related distractor), BONE (action-related distractor), and CAT (unrelated distractor). After hearing pirate, we might expect looks to the two objects that are associated with the agent (SHIP and TREASURE). When the action (hides) is heard, this makes one member of the initial cohort (SHIP) unlikely, and one might expect subsequent looks only to the target (TREASURE). This approach is analogous to the strategy proposed by the Cohort model of word recognition (Marslen-Wilson, 1987), in which the beginning of a word activates all words that are consistent with that beginning; as the rest of the word is spoken, subsequent cues eliminate inconsistent members of the cohort until only one word—the correct word—remains active.

An alternative strategy is suggested by the TRACE and Merge/Shortlist models of speech perception (McClelland and Elman, 1986, Norris, 1994). In those models, subsequent cues may both eliminate members of the initial cohort and activate new words that are consistent with later cues—even if those new words are inconsistent with the initial sound of the word. Thus, as the beginning of a target word such as beaker is heard, listeners might initially activate not only beaker but also words with similar onsets such as beetle and bee. When the second syllable is heard, this might eliminate beetle and bee. But TRACE also predicts that new words that are consistent with that second syllable, such as speaker, will become active. The strategy makes it possible to recover from initial errors in perception or production and, in fact, seems to provide a better fit for the empirical data (Allopenna et al., 1998). In the case of transitive sentence processing, after hearing the agent and action in “The pirate hides …” example, we might then expect fixations not only to the target object (TREASURE) but also to the item that is locally consistent with the action (BONE). There is increasing evidence that adults’ sentential comprehension may be at least partially influenced by local coherence effects analogous to those predicted by TRACE for word recognition (Kukona et al., 2011, Kuperberg, 2007, Tabor et al., 2004). The extent to which such local strategies are applicable sentence processing is debated, and whether children use different strategies than adults is unknown.

Section snippets

Adult participants

A total of 48 native English-speaking college students (30 women and 18 men) between 18 and 28 years of age (M = 21.4 years) took part in this study in return for course credit. An additional 12 participants took part and received credit but were excluded from analysis: 11 for significant exposure to other languages during childhood and 1 for receiving prior speech therapy. Participants had normal or corrected-to-normal vision and normal hearing, and they reported no history of diagnosis or

Behavioral accuracy

To ensure that adults and children understood the sentences and task, we first examined their accuracy to select the correct target picture. Accuracy was very high on the task in both groups, and very few errors were made in selecting the correct target picture. There were 4 incorrect responses across adults (99.5% correct), and there were 20 incorrect responses across children (97.4% correct). All analyses below were conducted, and all results are reported, with these incorrect responses

Discussion

In this investigation, we measured the relationships among anticipatory fixations during a simple spoken sentence comprehension task, age, and linguistic ability. To our knowledge, this study is the first to show an association between vocabulary knowledge and incremental sentence interpretation in both adults and children. The study also replicates and extends a number of prior findings in the incremental sentence processing literature. As reported in previous studies (Altmann and Kamide, 1999

Acknowledgments

We express our sincere gratitude to the families and children throughout the San Diego region who participated in this research. We also thank Sarah Creel and Heather Pelton for assisting with two prior norming studies leading up to the research and thank Kim Sweeney and Alex Huynh for assisting with data collection. In addition, we thank the editor, an anonymous reviewer, and Gerry Altmann for their helpful comments and suggestions. This work was funded by grants received from the National

References (88)

  • A. Fernald et al.

    Blue car, red car: Developing efficiency in online interpretation of adjective–noun phrases

    Cognitive Psychology

    (2010)
  • S.E. Gathercole et al.

    Evaluation of the role of phonological STM in the development of vocabulary in children: A longitudinal study

    Journal of Memory and Language

    (1989)
  • L.A. Hald et al.

    The interaction of discourse context and world knowledge in online sentence comprehension. Evidence from the N400

    Brain Research

    (2007)
  • M. Hare et al.

    Activating event knowledge

    Cognition

    (2009)
  • Y. Kamide et al.

    The time-course of prediction in incremental sentence processing: Evidence from anticipatory eye movements

    Journal of Memory and Language

    (2003)
  • A. Kukona et al.

    The time course of anticipatory constraint integration

    Cognition

    (2011)
  • G.R. Kuperberg

    Neural mechanisms of language comprehension: Challenges to syntax

    Brain Research

    (2007)
  • C. Lew-Williams et al.

    Real-time processing of gender-marked articles by native and non-native Spanish speakers

    Journal of Memory and Language

    (2010)
  • W.D. Marslen-Wilson

    Functional parallelism in spoken word recognition

    Cognition

    (1987)
  • W.D. Marslen-Wilson et al.

    Processing interactions and lexical access during word recognition in continuous speech

    Cognitive Psychology

    (1978)
  • J.L. McClelland et al.

    The TRACE model of speech perception

    Cognitive Psychology

    (1986)
  • K. Nation et al.

    Investigating individual differences in children’s real-time sentence comprehension using language-mediated eye movements

    Journal of Experimental Child Psychology

    (2003)
  • D. Norris

    Shortlist: A connectionist model of continuous speech recognition

    Cognition

    (1994)
  • A.P. Salverda et al.

    The role of prosodic boundaries in the resolution of lexical embedding in speech comprehension

    Cognition

    (2003)
  • J.C. Sedivy et al.

    Achieving incremental semantic interpretation through contextual representation

    Cognition

    (1999)
  • J. Snedeker et al.

    The developing constraints on parsing decisions: The role of lexical biases and referential scenes in child and adult sentence processing

    Cognitive Psychology

    (2004)
  • W. Tabor et al.

    Effects of merely local syntactic coherence on sentence processing

    Journal of Memory and Language

    (2004)
  • J.C. Trueswell et al.

    The kindergarten-path effect: Studying on-line sentence processing in young children

    Cognition

    (1999)
  • G.T.M. Altmann et al.

    Incrementality and prediction in human sentence processing

    Cognitive Science

    (2009)
  • J.M. Anglin

    Vocabulary development: A morphological analysis

    Monographs of the Society for Research in Child Development

    (1993)
  • E. Bates et al.

    On the inseparability of grammar and the lexicon: Evidence from acquisition, aphasia, and real-time processing

    Language and Cognitive Processes

    (1997)
  • E. Bates et al.

    On the emergence of grammar from the lexicon

  • I.L. Beck et al.

    Conditions of vocabulary acquisition

  • P. Boersma

    Praat, a system for doing phonetics by computer

    Glot International

    (2001)
  • J. Bresnan

    Lexical functional syntax

    (2001)
  • E. Carrow-Woolfolk

    Comprehensive assessment of spoken language

    (1999)
  • C. Caselli et al.

    A comparison of the transition from first words to grammar in English and Italian

    Journal of Child Language

    (1999)
  • C.G. Chambers et al.

    Lexical competition during second-language listening: Sentence context, but not proficiency, constrains interference from the native lexicon

    Journal of Experimental Psychology: Learning, Memory, and Cognition

    (2009)
  • A.E. Cunningham et al.

    Early reading acquisition and its relation to reading experience and ability 10 years later

    Developmental Psychology

    (1997)
  • P.S. Dale et al.

    Lexical and grammatical development: A behavioural genetic perspective

    Journal of Child Language

    (2000)
  • K.A. DeLong et al.

    Probabilistic word pre-activation during language comprehension inferred from electrical brain activity

    Nature Neuroscience

    (2005)
  • D.M. Dunn et al.

    Peabody picture vocabulary test

    (2007)
  • S. Ellis Weismer

    Typical talkers, late talkers, and children with specific language impairment: A language endowment spectrum?

  • K.G. Estes et al.

    Can infants map meaning to newly segmented words?

    Psychological Science

    (2007)
  • Cited by (236)

    View all citing articles on Scopus
    View full text