Sie sind auf Seite 1von 17

The effect of prior experience on grammar learning.

Dionne Angela Donnelly Module Code: PSYC337 Word Count: 4030

Abstract Previous research has shown that knowledge of grammar depends upon implicit, not explicit learning (Ferreira & Bock, 2006). One aspect of implicit learning is structural priming, in which people repeat sentence structures they have previously heard (Ferreira, 2003). Sleep has been found to boost the effects of previous learning (Maquet, 2001). The current study aimed to use the structural priming phenomenon (using double object [DO] and prepositional object [PO] datives) to test if adults knowledge could be changed by manipulating their experience of the language. A second aim was to see if this effect was stronger after a night of sleep. It was hypothesised that: 1) The number of PO and DO datives produced by adults in a sentence completion task would be influenced by the relative number of PO and DO datives recently experienced. 2) The effect would be larger when the biased learning and test phases were separated by a nights sleep. 53 participants from a student population completed a sentence completion task, in a procedure similar to that used by Kaschak, Loney & Borreggine (2006). Those in the DO biased group were exposed to a sentence list containing 75% DO datives, and 25% PO. Those in the PO biased group were exposed to 75% PO datives, and 25% DO. They then had to complete 8 non-biased sentence stems in a pre-sleep phase and 8 similar stems in a post-sleep phase. The findings of the study were that hypothesis 1 was accepted; learning was influenced by the proportion of DO or PO datives experienced. Hypothesis 2 was rejected; sleep did not increase the effect. It was concluded that this study only partially supports the implicit theory of grammatical learning.

Introduction People generally repeat sentences similar to those they have previously heard or have said themselves (Ferreira & Bock, 2006; cf. Ferreira, 2003). This priming phenomenon is not
1

specific to any particular structure; it affects structures as diverse as use of transitives/datives and inclusion of the optional complementiser that (Ferreira, 2003). For the purpose of this study, the focus is on priming of dative sentences, which are sentences used to indicate the indirect object of a verb. The focus shall be on priming of double object (DO) datives as compared to prepositional object (PO) datives. For example, if a participant is shown a DO structured sentence such as Leeanne gave her sister a magazine, they are more likely to use the same structure on a target sentence than a PO structure such as Leeanne gave a magazine to her sister (see Kaschak, Loney & Borreggine, 2006, p. 81, and the Methods section below for further examples).

Chang, Dell, Bock and Griffin (2000) argue that structural priming aids us in creating the syntactic structures which we utter; it is a syntactic process. This means it maps the elements of messages into linguistically coded structures and sequences (Chang, Dell, Bock & Griffin, 2000, p. 218). In support of this premise, Bock (1986) showed that syntactic priming can even occur when the prime and target sentences share no open-class words, and when sentences share a similar syntax but dissimilar semantic message. Conversely, priming does not occur when sentences have dissimilar syntax but similar prosody or surface appearance even if the function words in the prime are the same (Bock & Loebell, 1990). Despite this, it does appear that priming is boosted when the target sentence repeats content words from the prime, especially when this shared word is a main verb (Pickering & Branigan, 1998). It also occurs when the content words of prime and target are related in meaning (Cleland & Pickering, 2003). Chang et al. (2000) created a computer model capable of priming certain sentence structures; it was found that priming was not reliant on lexical connections. Bock, Loebell and Morey (1992) argue that repetition of syntactic structures operates above any

semantic-assignment priming effects. However, in the Pickering and Branigan (1998) study the level of priming was not boosted by such factors when neutral trials were introduced.

Support for the existence of structural priming effects across trials comes from a study by Hartsuiker and Westenberg (2000). They concluded that repeated production of particular constructions (such as datives) increased the likelihood that such constructions would be repeated later on in the experiment. Furthermore, Bock & Kroch (1989, cited in Kaschak, Loney & Borreggine, 2006) found that after producing several examples of one construction (e.g. a DO structure), priming on a competing construction (such as a PO structure) was diminished. Kaschak, Loney and Borreggines (2006) main finding in their study of PO vs. DO structures was that strength of priming was dependent upon the relative number of times that each structure had been produced during the priming phase in the study. In concordance with the above results, Bock & Loebell (1990) found that when passive (as opposed to active) structures were used the most during an experiment, they were produced more fluently in the test phase (despite the fact that in English, active structures are most frequently used). However it was also found that this fluency was only present in the short-term, whereas priming has been found to be a longer-term process (Wheeldon & Smith, 2003, cited in Ferreira & Bock, 2006). In support of the argument that priming is a longer-term process, Boyland & Anderson (1992, cited in Chang et al., 2000) found priming still occurred when targets were presented 20 minutes after prime sentences.

Chang et al.s (2000) computer model created a priming effect for datives and transitives that continued over ten sequences. Savage, Lieven, Theakston, and Tomasello (2006) discovered that 4-year-olds demonstrate priming effects for up to month, which suggests that changes in grammatical learning can last for an extended period of time.
3

It is argued that this process by which we construct sentence structures is shaped by experience within the production system. This means that the system learns over time (Chang, Dell, Bock & Griffin, 2000). Bock and Griffin (2000) postulate that priming could represent a short-term reflection of longer-term adaptations within the learning system. They conducted an experiment in which participants were given priming sentences and then asked to describe a picture. The number of neutral sentences between prime and target varied between 0 and 10. They found that priming worked just as well following 10 neutral sentences as it did 0. However, other research, such as Levelt and Kelter (1982) has found that priming effects can disappear after only one neutral sentence, meaning it can also be a short-term effect. A similar result was seen in a written production study by Branigan, Pickering and Cleland (1999; see also, Wheeldon & Smith, 2003), who found that the priming effect was weakened by one neutral sentence between targets.

So, while it is clear from the evidence above that priming is a form of learning, there is much debate over the type of learning the system employs; in relation to the differential time courses priming can take and in particular whether it is implicit or explicit. Ferreira and Bock (2006, p. 3) define implicit learning as incidental tuning or adjustment of the tendencies of a processing system as a function of experience. It involves learning complex, abstract relations, which creates knowledge that does not get accessed consciously (Seger, 1994)

An example of implicit learning under an experimental condition is the stem completion task. Participants are primed with a word (e.g. Cheese), and later given the first few letters and

asked to complete the stem (e.g. Che). Participants often produce the primed word even when they have no explicit memory of learning the word. The word-stem completion task is similar to the method used in the present study, but instead sentence completion is used. Kidd (in submission) postulates that implicit learning is the mechanism for infants speech perception and therefore may play a role in their early syntax acquisition through recognition of statistical regularities (Bannard, Lieven & Tomasello, 2009). Kidd (in submission), conducted an experiment in which one hundred 4-6 year olds were tested on three measures: implicit learning; explicit learning; and syntactic priming. Increased scores on the statistical learning (implicit task) predicted how long the priming effect would last post-test. Priming was not predicted by scores on the explicit learning task. It was concluded that learning such patterns is therefore implicit and gradual, like syntax acquisition, and such learning is seen in the prolonged use of primed structures following the test phase. Rowland (2007) identified frequency of input as playing an important role in syntactic structure acquisition. Bock, Loebell and Morey (1992) found that participants told to remember the syntax of sentences showed significantly greater priming than those who were told to remember the meaning. Further support for syntactic priming being an implicit process comes from Ferreira, Bock, Wilson, and Cohen (2005, cited in Ferreira & Bock, 2006). They found that patients with anterograde amnesia (which affects explicit learning capabilities) showed the same level of priming as controls, but performed poorly on a test of their explicit memory for syntax. This double dissociation points to priming as being an implicit, not explicit process. Despite this, it is possible that explicit learning accounts for the lexical effects of priming (such as the boost seen when content words or main verbs are the same), whereas implicit learning may account for the structural effects, which are less explicitly pronounced than other features of a given sentence (Ferreira & Bock, 2006).

It is commonly thought that sleep helps to consolidate fragile short-term memory traces into long-term memory (Maquet, 2001). Dumay and Gaskell (2007) argue that lexicalisation (integration of recently learned words into long-term lexical memory) occurs during sleep. They recruited 64 subjects, who were tested (on explicit memory and lexical activity) immediately following exposure to words, then at 8pm and 8am (following sleep). Evidence of lexicalisation was only consistently shown following a nights sleep. It logically follows that if words can be implicitly integrated into memory during sleep, then so too can syntactical structures.

There are several possible roles of sleep in memory consolidation. It could be due to lack of interference during sleep. Or that sleep provides specific neural conditions required for Maquet, 2001) found that learning is only impaired when REM is deprived. Koninck & Prvost (1991, cited in Maquet, 2001) found that sleep is different following training when compared to controls, as after training REM is increased. Maquet, Laureys, Peigneux, Fuchs, Petiau et al. (2000, cited in Maquet, 2001) conducted PET scans during sleep and found that those areas activated during a task are also activated during REM sleep following the task. They also found that following sleep, performance on the task was improved. However, such neural patterns are not exclusive to a sleeping state (Maquet, 2001).

The current research aims to consolidate the three streams of research noted above and uncover the strength and direction of the relationship between implicit learning, priming and sleep. Furthermore, the current study is necessary because there is a dearth of empirical research showing that our actual, everyday knowledge of grammar is affected by implicit statistical learning. Therefore, the method employed in this study to change participants
6

experience of the language should also change their production of the language (due to modifications in the underlying grammatical representation of the brain). Finally, whilst there is much evidence for sleep aiding memory, there is little which looks at memory for statistical regularities.

Thus, there were two aims of the present study. The first was to test if adults knowledge and use of grammar could be changed following manipulation of their experience of the language (as hypothesised by the implicit learning theory of grammar). The second aim was to investigate whether a nights sleep causes the effect of prior experience to be stronger. These aims led to two hypotheses. Hypothesis 1 stated that the number of PO and DO datives produced by adults in a sentence completion task (at test) would be influenced by the relative number of PO and DO datives recently experienced (during the biased learning phase). Hypothesis 2 was that the effect of recent experience would be larger when the biased learning and test phases are separated by a nights sleep.

Method Participants 53 participants were recruited using opportunity sampling. The cohort consisted of third year psychology undergraduates with an approximate age range of 20-25. Design The hypotheses presented above produced two independent variables. The first (IV1), which was between-subjects and consisted of two levels, was biased learning group. The second independent variable (IV2), was within-subjects and also consisted of two levels:
7

participation in the pre-sleep test phase, and the post-sleep test phase. The dependent variable in this study was proportion of datives produced that were DO datives. As the study is a mixed design (as it includes both between-subjects and within-subjects variables) the statistical test used was a 2x2 mixed design ANOVA. Stimuli The biased learning phase consisted of 14 datives and 16 fillers. Fillers were all simple sentences, which were designed to resist completion with a dative. Those in the DO group were exposed to a sentence list containing 75% DO datives, and 25% PO. Those in the PO group were exposed to 75% PO datives, and 25% DO. Pre-sleep and post-sleep phases consisted of 8 test sentences and 8 fillers, which, although semantically different, shared the same structure across phases. Procedure Participants were informed of their right to withdraw and assured that their data would be confidential and anonymous (see Appendix). Subjects were asked to participate in a sentence completion task similar to the procedure carried out by Kaschak, Loney & Borreggine (2006). 29 participants were assigned via opportunity sampling to the DO dative biased learning condition. Similarly, 24 participants were assigned to the PO dative biased learning condition. The first test session incorporated both the biased learning test phase and the presleep test phase. In the biased learning phase, 14 target sentence stems were completed. These were either biased towards DO (e.g. The captain issued the old sailor the) or PO (e.g. A government awarded the grant money to ) dative completion (see Appendix). The datives were separated by fillers. These were used to avoid participants recognising the aim of the study and conforming to the demand characteristics inherent in the study. The test took approximately 20 minutes to complete. In the pre-sleep test phase, 8 sentence stems were
8

completed. These sentences had no bias, and were followed the structure: subject-verb (for example, the librarian told...). The second test session consisted of the post-sleep test phase. This was completed more than 24 hours later, so that the first test session (i.e. the biased learning phase and pre-sleep test phase) was followed by a nights sleep. This phase also consisted of 8 sentence stems, in an identical manner to the pre-sleep test phase.

Coding Responses were coded as one of three structure types: 1) PO dative, which follow the structure: PERSON verb OBJECT to PERSON. 2) DO dative, which follow the structure PERSON verb PERSON OBJECT. 3) Other, which refers to anything that is not classified as a dative. For example, transitives such as the diver sold drugs or idioms such as the novelist threw caution to the wind (see Appendix).

Results In order to test the hypotheses stated above, the number of DO datives was divided by the overall number of datives, in order to show the proportion of DO datives produced (the dependent variable). From this proportion it is possible to also see how many PO datives were produced. Other structures (see the Coding section above) were not included in the analysis as they are not incorporated in the hypotheses. Table 1 shows the mean proportions of DOs produced in each condition. This shows that the DO biased group produced more DO datives (M = 0.74, SD = 0.05) than the PO biased group (M = 0.54, SD = 0.07) in the biased learning/pre-sleep phase. These figures changed considerably in the post-sleep phase. The number of DO datives produced by the DO biased group decreased (M = 0.66, SD = 0.05). In

contrast, the number of DO datives that were produced by the PO biased group increased in the post-sleep phase (M = 0.69, SD = 0.06). Table 1: Mean (SD) proportion of DO datives produced by biased learning groups in the pre- and post-sleep phases. Pre-sleep phase 0.74 (0.05) 0.54 (0.07) Post-sleep phase 0.66 (0.05) 0.69 (0.06)

DO biased group PO biased group

The mean interaction plot in Figure 1 shows that there is a clear interaction between biased learning group and sleep phase. This shows that the mean number of DO datives produced by the DO and PO biased groups in the post-sleep phase converged towards a similar mean (0.66 [0.05] and 0.69 [0.06] respectively). Figure 1: Mean interaction plot showing the interaction between biased learning group and sleep phase.

0.8 Mean proportion of DO datives produced 0.75 0.7 0.65 0.6 0.55 0.5 0.45 0.4 Pre-sleep Sleep Phase Post-sleep DO Biased PO Biased

As IV1 and IV2 both consisted of two levels, a 2x2 mixed design ANOVA was used to analyse the data for any main effects. As the data were not normally distributed, an empirical
10

logic transformation was carried out in order to normalise the data for analysis. There was no significant main effect for group (F(1, 51) = 2.08, P = 0.16) No significant main effect was found for sleep phase (F(1, 51) = 0.74, P = 0.4). A significant interaction between sleep phase and group was found (F(1, 51) = 4.21, P = 0.05). Analysis of partial Eta squared (0.8) shows a moderate to large effect size. As a significant interaction effect was found, two oneway ANOVAs were conducted. It was found that biased learning phase had a significant effect in the pre-sleep phase (F(1, 51) = 5.88, P = 0.02), but not in the post-sleep phase (F(1, 51) = 0.06, P = 0.81). Discussion The above results show that Hypothesis 1 was supported as DO production was influenced by the relative number of DO and PO datives exposed to in the biased learning phase. The figures in the pre-sleep phase were as predicted, with the DO biased group producing more DO datives than those in the PO biased group. This supports the premise that prior experience changes our grammatical knowledge. However, despite the PO biased group being exposed to 75% PO datives in the biased learning phase, the proportion of datives they produced was still biased towards DO datives (54% in the biased learning phase). Whilst this is only a marginal bias, it may indicate that in everyday contexts we are more prone to using DO datives. On the other hand, Hypothesis 2 was rejected. The effect of recent sleep was only larger in the PO biased group. Additionally, the increase in DO production by the PO biased group in the post-sleep phase was unexpected. The fact that the proportion of DO datives produced in the post-sleep phase became more similar across the two groups was also unexpected as Hypothesis 2 predicted that the difference in DO datives produced should have become greater in the post-sleep phase. This convergence of the proportion of datives may also be conferred as return to baseline levels of DO dative use. However, this is only a

11

speculatory explanation as a control measure was not taken in order to establish participants baseline levels of dative use.

The results of the present study are similar to those in structural priming studies that only test at the biased learning phase, such as Bock (1986) and Hartsuiker and Westenberg (2000), as syntactic priming did occur (as primes and targets were syntactically not semantically similar) in the biased learning phase. These findings also provide support for the research by Kaschak, Loney & Borreggine (2006), and Bock and Kroch (1989, cited in Kaschak, Loney & Borreggine, 2006) as exposure to either type of dative construction led to diminished production of the contrasting construction in the biased learning phase. As the test took approximately 20 minutes to complete and filler sentences were used, the study also provides support for the research by Boyland & Anderson (1992, cited in Chang et al., 2000), and Chang et al.s (2000) computer model of priming.

However the present study only partially supports studies which have shown priming to continue for longer periods of time, such as that by Savage, Lieven, Theakston, and Tomasello (2006) who found priming effects up to a month later in 4 year olds. On the basis of the above results, in which priming effects across the two groups differed greatly after one only night of sleep, it appears unlikely that priming would last so long. The current study appears to reflect the lack of consensus in the literature regarding how long priming lasts, especially with the inclusion of neutral (or filler) sentences (see studies by Levelt and Kelter, 1982 and Bock and Griffin, 2000, for contrasting results). Therefore the results do not sit well with the well-established assertion that structural priming is a form of long-term implicit learning (as proposed by Kidd [in submission] and Bannard, Lieven & Tomasello, 2009).
12

This could be related to the finding by Rowland (2007) that frequency of input is important. It is possible that this study did not include enough datives to create a long-lasting effect, which is why structural priming was only significant at the biased learning phase and not following sleep. In the short-term this study does support the principle of grammatical learning being implicit as participants showed a structural priming effect whilst they were only told to complete sentences and not explicitly asked to focus on sentence structure. This therefore substantiates research by Ferreira, Bock, Wilson, and Cohen (2005, cited in Ferreira & Bock, 2006) into implicit grammatical learning in patients with anterograde amnesia, who performed poorly on explicit memory for syntax but showed normal levels of priming.

Finally, in relation to relevant literature, this study provides a challenge to the majority of research which states that sleep helps to consolidate learning, in particular Maquets (2001) study. This is the case because sleep was not found to increase the effect of recent experience (i.e. implicit learning of structural primes and therefore a change in grammatical knowledge). A problem of this study, relating to the sleep phase, is that it was impossible standardise the amount of sleep that participants got, or to be sure that they got a certain/minimum amount of sleep. Thus, the seemingly anomalous result of sleep not boosting learning may be due to participants not having enough REM sleep throughout the night, which is when areas activated during the task session are reactivated and appear to be critical in aiding learning (Maquet, 2001)

Possible improvements to this study (and potential future research) could be to include more dative structures in the biased learning phase to see if the length of time priming effects last for is affected by the frequency of input, as suggested by Rowland (2007). Such a study could
13

last over several priming sessions, and have test phases at different points in the future, such as a day, a week, or even a month after the priming sessions were completed. Another possibility for future study (which would also improve upon the methodology of the current study) would be study the brain activity (perhaps with PET or fMRI scanning) which occurs during the priming task and whilst the participants sleep after the task. If this was done over consecutive priming/test sessions then the scans could be compared for structural changes to the brain following grammatical learning, and would also improve upon one of the methodological flaw noted above, by ensuring that participants did get a particular amount of REM sleep.

Ultimately, this study found that the number of datives produced by adults in a sentence completion task was significantly influenced by the proportion of DO and PO datives they had recently experienced. If was also found that a nights sleep did not make this effect larger. These findings both support and contradict previous research into the areas of structural priming/implicit learning and sleeps effects on learning (respectively). This is because structural priming was found to only change participants implicit grammatical knowledge immediately following the biased learning phase and not after a night of sleep. Consequently, this study only partially supports the implicit theory of grammatical learning.

References Bannard, C., Lieven, E., & Tomasello, M. (2009). Modelling Children's Early Grammatical Knowledge. Proceedings of the National Academy of Sciences, 106 (41), 17284-17289.

14

Bock, K. (1986). Syntactic persistence in language production. Cognitive Psychology, 18, 355387. Bock, K. & Griffin, Z. M. (2000). The persistence of structural priming: Transient activation or implicit learning? Journal of Experimental Psychology, 129 (2), 177-192. Bock, K. & Loebell, H. (1990). Framing sentences. Cognition, 35, 139. Bock, K., Loebell, H. & Morey, R. (1992). From conceptual roles to structural relations: Bridging the syntactic cleft. Psychological Review, 99, 150171. Branigan, H. P., Pickering, M. J., & Cleland, A. A. (1999). Structural priming in written production: Evidence for rapid decay. Psychonomic Bulletin & Review, 6, 635 640. Chang, F., Dell, G. S., Bock, K. & Griffin, Z. M. (2000). Structural priming as implicit learning: A comparison of models of sentence production. Journal of Psycholinguistic Research, 29 (2), 217-229. Cleland, A. A., & Pickering, M. J. (2003). The use of lexical and syntactic information in language production: Evidence from the priming of noun-phrase structure. Journal of Memory and Language, 49, 214 230. Dumay, N. & Gaskell, M. G. (2007). Sleep-associated changes in the mental representation of spoken words. Psychological Science, 18, 35-39. Ferreira, V. S. (2003). The persistence of optional complementizer production: Why saying that is not saying that at all. Journal of Memory and Language, 48(2), 379398. Ferreira, V. S. & Bock, K. (2006). The functions of structural priming. Language and Cognitive Processes, 21, 10111029.

15

Hartsuiker, R. J. & Westenberg, C. (2000). Word order priming in written and spoken sentence production. Cognition, 75, 2739. Kaschak, M. P., Loney, R. A. & Borreggine, K. L. (2006). Recent experience affects the strength of structural priming. Cognition, 99, B73B82. Kidd (in submission). Implicit statistical learning predicts natural language learning. Developmental Psychology. Levelt, W. J. M. & Kelter, S. (1982). Surface form and memory in question answering. Cognitive Psychology, 14, 78106. Maquet, P. (2001). The role of sleep in learning and memory. Science, 294, 1048-1052. Pickering, M. J., & Branigan, H. P. (1998). The representation of verbs: Evidence from syntactic priming in language production. Journal of Memory and Language, 39, 633 651. Rowland, C. F. (2007). Explaining errors in childrens questions. Cognition, 104, 106 - 134. Savage, C., Lieven, E. V. M., Theakston, A. L. & Tomasello, M. (2006). Structural priming as implicit learning in language acquisition: The persistence of lexical and structural priming in 4-year-olds. Language Learning and Development, 2, 27 49. Seger, C. A. (1994). Implicit learning. Psychological Bulletin, 115, 163196. Wheeldon L. R. & Smith, M. C. (2003). Phrase structure priming: A short-lived effect. Language and Cognitive Processes, 18(4), 431442. Appendices Appendices are located on the accompanying disc provided.

16

Das könnte Ihnen auch gefallen