Learning new vocabulary during childhood: Effects of semantic training on lexical consolidation and integration

Abstract

Research suggests that word learning is an extended process, with offline consolidation crucial for the strengthening of new lexical representations and their integration with existing lexical knowledge (as measured by engagement in lexical competition). This supports a dual memory systems account, in which new information is initially sparsely encoded separately from existing knowledge and integrated with long-term memory over time. However, previous studies of this type exploited unnatural learning contexts, involving fictitious words in the absence of word meaning. In this study, 5- to 9-year-old children learned real science words (e.g., hippocampus) with or without semantic information. Children in both groups were slower to detect pauses in familiar competitor words (e.g., hippopotamus) relative to control words 24 h after training but not immediately, confirming that offline consolidation is required before new words are integrated with the lexicon and engage in lexical competition. Children recalled more new words 24 h after training than immediately (with similar improvements shown for the recall and recognition of new word meanings); however, children who were exposed to the meanings during training showed further improvements in recall after 1 week and outperformed children who were not exposed to meanings. These findings support the dual memory systems account of vocabulary acquisition and suggest that the association of a new phonological form with semantic information is critical for the development of stable lexical representations.

Introduction

Numerous studies have demonstrated the efficiency with which infants and young children form mappings between words and their referents (Bloom and Markson, 1998, Carey and Bartlett, 1978, Spiegel and Halberda, 2011). Consequently, vocabulary acquisition has been conceptualized as a “relatively simple affair” (Plunkett & Wood, 2006, p. 165). However, an alternative view is that vocabulary learning during childhood is partial and incremental (e.g., Dockrell et al., 2007, Nagy and Scott, 2000). Recent developmental studies suggest that a prolonged period of time is needed for a novel nonword (e.g., biscal) to become integrated with the existing lexicon (Brown et al., 2012, Henderson et al., 2013), with sleep playing an important role in the “offline consolidation” (stabilization and integration) of new phonological forms (Henderson, Weighall, Brown, & Gaskell, 2012). These results have been explained within the complementary learning systems (CLS) framework, which proposes that new information is initially stored separately from existing knowledge and integrated over time (Davis and Gaskell, 2009, McClelland et al., 1995, Norman and O’Reilly, 2003, O’Reilly and Norman, 2002). However, previous research has tended to study fictitious nonword learning in the absence of word meaning and, crucially, we do not know whether providing meaning during training changes this prolonged time course of lexical integration. Because the purpose of language is to extract and convey meaning, arguably a word cannot be deemed to be fully acquired until it has been integrated with semantic knowledge; hence, it is imperative that we understand how children use meaning in the process of word learning. The current study presents a more naturalistic test of the CLS framework and examines how real (rather than fictitious) words and their meanings are acquired and integrated with the lexicon. Specifically, we address whether providing information about word meaning during training influences the time course of spoken word learning, focusing on the extent to which a new word has been integrated with existing word knowledge and the ease with which it can be retrieved over time.

One way of establishing whether a new word has been acquired is to examine when it begins to exhibit hallmarks of lexical processing typically observed for existing words (Leach & Samuel, 2007). One such hallmark of an established lexical entry acknowledged by numerous models of spoken word recognition is its ability to compete with similar sounding entries for identification (McClelland & Elman, 1986). Hence, a strong test of whether a new speech sequence has been integrated with the lexicon is whether it engages in lexical competition with existing representations during speech perception (e.g., Gaskell & Dumay, 2003).

Gaskell and colleagues have examined how lexical activity changes when adults (Dumay and Gaskell, 2007, Gaskell and Dumay, 2003) and school-aged children (Brown et al., 2012, Henderson et al., 2012, Henderson et al., 2013) learn fictitious novel nonwords (e.g., biscal). Participants made speeded decisions about the presence of a silent pause inserted in existing words with the same onsets (e.g., bisc_uit). Pause detection latency provides an online index of lexical activity at the point in the word where the pause is encountered, with slower pause detection latency indicating more lexical activity (Mattys & Clark, 2002). Gaskell and colleagues showed that pause detection latencies in existing words become slower if participants have recently learned an onset competitor (e.g., biscal). Studies with adults (Dumay & Gaskell, 2007) and 7- to 12-year-old children (Henderson et al., 2012) have shown that this “lexical competition” effect emerges 12 h after exposure to novel nonword competitors, but only if that 12-h period includes sleep. In both studies sleep also benefited explicit recall of the novel nonwords and the children retained this knowledge 1 week later, providing evidence for long-lasting phonological word learning following a short training phase. This finding is consistent with previous reports of delayed improvements in the recognition of newly trained words (without additional exposure) in 3- to 7-year-old children (Dockrell et al., 2007, Storkel, 2001). Together, these findings have led to a significant shift in our understanding of vocabulary acquisition, suggesting that offline consolidation facilitates the integration of novel fictitious words into the developing lexicon and plays a role in the stabilization of new phonological representations.

CLS theory (McClelland et al., 1995, Norman and O’Reilly, 2003, O’Reilly and Norman, 2002) provides a good explanation for these findings. This framework can account for vocabulary acquisition in terms of the operation of general memory systems, whereby new information is initially stored independently from existing knowledge and then integrated over time (Davis & Gaskell, 2009). Under this view, novel word representations are initially sparsely coded in the hippocampus, with offline interactions resulting in strengthening of distributed representations in long-term neocortical memory (Davis et al., 2009, French, 1999, Wilson and McNaughton, 1994). Tamminen, Payne, Stickgold, Wamsley, and Gaskell (2010) reported that sleep spindle activity (11- to 15-Hz oscillations lasting up to 3 s) is positively associated with overnight increases in lexical competition, suggesting that sleep plays an active role in the process of lexical integration, possibly by enabling the transfer from hippocampal to neocortical representations.

The main aim of this study was to address whether previous data on consolidation effects in vocabulary acquisition accurately reflect naturalistic word learning when semantic information is available. We examined whether semantic knowledge shapes the time course with which new lexical information is strengthened such that it can be explicitly retrieved and integrated with existing lexical information (as indexed by the emergence of lexical competition). Previous studies have largely used purely phonological training regimes without providing participants with information pertaining to word meaning (Brown et al., 2012, Dumay and Gaskell, 2007, Gaskell and Dumay, 2003, Henderson et al., 2012, Henderson et al., 2013). It could be argued that the typical finding of a delay in lexical integration of the novel words is an artifact of the impoverished context in which the words have been learned.

Consistent with this position, Leach and Samuel (2007) found that novel nonwords engage with the lexicon immediately after learning (as measured by the retuning of phoneme boundaries) if semantic information has been provided. Participants were trained on novel nonwords containing either an “s” sound or a “sh” sound and then presented with novel nonwords containing an ambiguous phoneme halfway between “s” and “sh”. The novel nonwords engaged with the phoneme category level by adjusting phoneme boundaries. This effect occurred immediately after training that attached meaning to the novel nonwords but not when the training was purely phonological. However, it is not clear whether immediate lexical integration effects would be observed after semantic training when integration is measured by lexical competition. According to Davis and Gaskell (2009), the fast learning hippocampal system has a direct link to lexical phonology, and this might explain why novel words are able to retune phoneme boundaries soon after training even if they have not yet been fully integrated to the extent that they engage in lexical competition.

This explanation is consistent with the findings of Dumay, Gaskell, and Feng (2004), who trained adults on novel nonword competitors (e.g., cathedruke) in isolation in a phoneme monitoring task or in sentence context during a semantic verification task. In the latter condition, the novel nonwords were associated with the name of a conceptual category (e.g., vegetable) and two sentences conveying categorical information (“A cathedruke is a variety of vegetable”) and semantic context (“The cook served the boiled cathedruke with a steak and baked potatoes”). For both training conditions, lexical competition effects (e.g., delayed recognition of “cathedral”) emerged only after a delay of at least 24 h and, importantly, were still observed 1 week after training. Hence, in contrast to Leach and Samuel (2007), there was no suggestion of immediate lexical integration of novel words when a semantic learning environment is used.

The above studies all relate to adult vocabulary learning, and it is unclear how semantic context influences the integration and consolidation of novel spoken words earlier in development. McKague, Pratt, and Johnston (2001) taught 6- and 7-year-olds novel phonological forms alone or their phonological forms plus semantic information (as part of an illustrated story) before their orthographic forms were introduced. The semantic manipulation had no effect on reading times for the novel nonwords over a 2-day period (see also Nation, Angell, & Castles, 2007). In a free recall task, however, semantically trained items were recalled more reliably than phonologically trained items. This suggests that for children, providing semantics during training may enhance the formation of a new phonological representation. This fits with the view that orthographic learning involves the integration of phonological, orthographic, and semantic representations, supported by findings that new written words presented with semantic information are identified more accurately than when words are presented in isolation (e.g., Ouellette & Fraser, 2009; see also McKay et al., 2008, Rueckl and Olds, 1993, Taylor et al., 2011, Wang et al., 2013, Whittlesea and Cantwell, 1987). Importantly, however, studies are yet to examine the influence of semantic training on spoken word learning and lexical integration in children; hence, the current study addressed this important issue.

Evidence is clear in pointing to the importance of offline consolidation for the stabilization of new phonological information, as reflected by continuing improvements in recall and recognition of new phonological forms over the course of a week (e.g., Henderson et al., 2012). However, it is less clear whether consolidation plays a similar role in the stabilization of explicit memory for new orthographic and semantic information. Tamminen and Gaskell (2013) taught adults new meaningful nonwords (using written rather than spoken presentation) and found that explicit recall of the nonword meanings declined rather than improved over the course of a week. Despite this, however, familiar words (that were not presented during training) were primed by the newly related nonwords in both unmasked and masked prime conditions, and this effect was strongest 1 week after training. These data suggest that the nonwords had been integrated with semantic memory (as evidenced by the priming effects) but that the episodic representations of the new meanings that were formed during training weakened over time (as indicated by the decline in explicit recall). Therefore, this suggests a dissociation between the strengthening of explicit memory for new phonological forms over time coupled with the weakening of explicit memory for new semantic features. However, it is unclear whether children will show this same pattern (indeed, a recent article suggests that children show stronger sleep-associated consolidation of explicit aspects of task performance; Wilhelm et al., 2013) and whether explicit memory for new orthographic information strengthens or weakens over time.

In this study, we examined the crucial issue of whether semantic training shapes the time course with which newly learned vocabulary is integrated with existing lexical information. An important contribution of this study is the examination of whether the findings from previous studies with children generalize to real (rather than fictitious) word learning; hence, in this study, children learned 14 unfamiliar science words. We acknowledge that the use of fictitious nonwords (e.g., biscal) is advantageous in many respects; they enable tight control over linguistic variables (e.g., phonotactic probability, frequency, length), ensure that stimuli are truly novel to participants, and are typically designed to be phonotactically indistinguishable from real words. However, it is questionable whether participants treat these nonwords as relevant only in the context of the experiment (Potts, St. John, & Kirson, 1989), particularly when they are not given a meaning (Brown et al., 2012, Dumay and Gaskell, 2007, Gaskell and Dumay, 2003, Henderson et al., 2012, Henderson et al., 2013, Tamminen et al., 2010). Therefore, the current study provides an important opportunity to assess key hypotheses of the CLS account of vocabulary acquisition using real words that are likely to be learned in the classroom.

Much of the established research on vocabulary acquisition in children focuses on concrete early-acquired words, and studies have rarely examined the factors that support subject-related vocabulary acquisition in school-aged children (but see Dockrell et al., 2007, for an exception). Vocabulary acquisition is crucial for academic development, and the importance is particularly marked in subjects such as science that make use of increasingly abstract and sophisticated terms. Thus, evidence from science word learning is important for establishing evidence-based strategies for teaching and learning.

During training, children were exposed to the spoken forms of the words accompanied either by semantic information (the “semantic group”) or orthographic information (the “form-only group”). The form-only training was first and foremost designed to be nonsemantic in nature to allow us to examine the presence or absence of semantic information on the time course of learning a new spoken word form. Our aim was to ensure that in both training conditions children were required to associate the phonological form of each word with another form of linguistic information (i.e., either semantic or orthographic). Children were tested immediately after training (0 h) and after 24 h and 1 week in order to examine the influence of training condition on aspects of word learning before and after periods of offline consolidation.

The current study also sought to rule out the effects of repeated testing on later consolidation effects. Previous studies have used repeated tests to measure changes in explicit memory and lexical competition for novel words (Brown et al., 2012, Dumay and Gaskell, 2007, Gaskell and Dumay, 2003, Henderson et al., 2012, Henderson et al., 2013). Importantly, these studies have shown that improvements in recall and changes in lexical competition are not dependent on reexposure to the novel stimuli or similar sounding existing words in the tests. For instance, Henderson and colleagues (2012) reported improvements in recall and recognition of novel nonwords and the emergence of lexical competition at a 12-h test, but only for children who had been trained during the evening; children who had been trained during the morning and retested after 12 h without sleep did not show significant improvements. This suggests that it was sleep, rather than time or repeated exposure, that was associated with the changes in explicit memory and lexical competition. However, in Tamminen and Gaskell (2013), there was evidence that repeated retests protected explicit recall of nonword meanings from forgetting. Namely, participants who were retested on meaning recall during the 8-day period recalled significantly more novel word meanings during the final session on Day 8 than those who were not tested prior to Day 8. This suggests that repeated administration of the recall task maintained explicit memory of the meanings (see also Bouwmeester and Verkoeijen, 2011, Roediger and Karpicke, 2006). Furthermore, as mentioned, Henderson and colleagues’ (2012) study did not examine word learning in the context of an associated meaning. To assess this issue in the case of learning words with meanings, in the current study, only half of the semantic training group received the immediate (0-h) test, whereas all participants were tested at 24 h. The participants who did not receive the immediate test were confined to the critical semantic training group.

In summary, the aims of this study were threefold. First, the influence of semantic training on the time course of explicit recall for newly learned words was explored. It was expected that children would show improvements in their ability to recall newly learned words (e.g., hippocampus) 24 h after training relative to an immediate (0-h) test and would retain this knowledge 1 week later (similar to previous findings with children; Brown et al., 2012, Henderson et al., 2012, Henderson et al., 2013). Crucially, if the presence of semantic information during training leads to more stable phonological representations (McKague et al., 2001), then improvements in cued recall may be larger (at 24 h or 1 week) after semantic training compared with form-only training.

Second, the time course of lexical integration (the ability of a newly learned word to compete for recognition with its existing base word) was examined. Slower pause detection latencies to similar sounding existing words (e.g., hippopotamus) relative to a control condition were expected 24 h after training on novel competitors (e.g., hippocampus) but not immediately (Henderson et al., 2012, Henderson et al., 2013). However, if the presence of semantic information during training leads to faster or more efficient lexical integration (Leach & Samuel, 2007), then lexical competition effects may emerge sooner (at 0 h) or be stronger (at 24 h or 1 week) after semantic training compared with form-only training.

Third, the consolidation of new semantic and orthographic information over time was explored. If, semantic and orthographic information benefits from a period of offline consolidation to the same extent as newly learned phonological information, (e.g., Brown et al., 2012, Dumay and Gaskell, 2007, Henderson et al., 2012, Henderson et al., 2013), then we may predict that the recognition and recall of new semantic knowledge (measured by picture identification and definition tasks, respectively) and orthographic knowledge (measured by a forced-choice recognition task) will improve at the delayed tests relative to the immediate test. Conversely, however, it is possible that explicit recall of newly learned semantic information would be subject to forgetting over time (cf. Tamminen & Gaskell, 2013).

In addition to these main aims, it was also predicted that performance at the 24-h test would be equivalent regardless of whether participants had received an immediate test and that participants first tested after 24 h would benefit from the consolidation opportunity (showing better recall and stronger lexical integration effects) compared with participants first tested immediately after training.