Implicit artificial grammar learning: effects of complexity and usefulness of the structure

Implicit artificial grammar learning: effects of complexity and usefulness

of the structure

Bos, E.J. van den

Citation

Bos, E. J. van den. (2007, June 6). Implicit artificial grammar learning: effects of complexity

Institute for Psychological Research, Faculty of Social Sciences, Leiden University. Retrieved

from https://hdl.handle.net/1887/12037

Version: Corrected Publisher’s Version

License:

Downloaded from:

Note: To cite this publication please use the final published version (if applicable).

Chapter 6

Implicit learning of a useful structure by adults and children

Abstract

The present study investigated whether implicit learning of a useful structure is invariant with age. Adults and 10 to 11-year-old children were presented with non- word sentences, generated by an artificial grammar, and three pictures of ice-creams.

For one half of the participants, the grammatical structure was useful to their task of identifying which ice-cream each sentence referred to; for the other half it was not. A subsequent grammaticality judgment test showed that adults had acquired more knowledge of the grammar than children, particularly about second-order

dependencies. The conditions under which implicit learning occurred, however, where the same for adults and children: they learned when the structure was useful to their current task, but not when it was useless.

Introduction

Implicit learning is often defined as a process that occurs without intention to learn, which results in knowledge that is not completely accessible to consciousness (e.g. Gomez, 1997; Reber, 1989; Seger, 1994). The process is tuned to patterns in the environment (Reber, 1989; Seger, 1994) and has been suggested to be involved in the acquisition of social rules (Reber & Allen, 2000; Seger, 1994) and the grammars of natural languages (Reber & Allen, 2000). An implication of this suggestion is that implicit learning has to be effective at an early age (Reber & Allen, 2000; Saffran, Newport, Aslin, Tunick & Barrueco, 1997).

In line with this prediction, implicit learning by 9 to 11-year-old children was demonstrated in the Artificial Grammar Learning (AGL) paradigm (Reber, 1967). The children memorized a set of letter strings without being informed that the strings had been generated by an artificial grammar. Nevertheless, their performance on a subsequent grammaticality judgment test with new letter strings indicated that they had acquired knowledge of the grammar (Fischer, 1997). (See Figure 1 on page 86 for an example of an artificial grammar that generates strings of non-words). A similar ability was demonstrated in one-year-old infants who had been exposed to

grammatical sequences of non-words presented auditorily for less than two minutes.

They kept their heads turned towards the speaker for a longer time when new grammatical sequences were presented than when ungrammatical sequences were played (Gomez & Gerken, 1999).

These studies provide evidence that implicit learning is, indeed, operational early in life. In addition, Reber (1992) predicted that implicit learning is invariant with age. He proposed that implicit learning is evolutionarily older than explicit forms of learning and would therefore be relatively stable. Implicit learning would be shared with other species, less prone to individual differences between humans, robust with respect to disorders and independent of IQ as well as age. Studies testing the age- invariance hypothesis have compared children and adults on several implicit learning tasks and have obtained various results.

In the AGL-paradigm, no correlation was obtained between age and

performance on a grammaticality judgment test for participants from 9 to 50 years old (Don, Schellenberg, Reber, DiGirolamo &Wang, 2003). Similarly, adults and 6 to 7- year-old children, who were presented with a continuous stream of spoken non-words during a drawing task, were equally good at distinguishing non-words that had been presented in the stream from new non-words composed of the same syllables (Saffran et al., 1997). These findings are in line with the proposal that implicit learning is invariant with age.

In contrast, Maybery, Taylor and O’Brien-Malone (1995) found a difference in implicit covariation learning between 5 to 7 and 10 to 12-year-old children. After being exposed to a contingency in the learning phase, the older children were better than the younger ones at predicting the location of a target picture on a matrix board based on the color of its cover and the side from which it was handed to them. When intellectually gifted children and children with mental retardation were added to the sample, however, performance was shown to improve with mental rather than chronological age (Fletcher, Maybery & Bennet, 2000).

Vinter and Detable (2003) suggested that these effects of (mental) age were due to the conceptual difficulty of the task designed by Maybery et al. (1995). They found no age related differences in implicit learning with a conceptually simple task.

In this task, participants unlearned the natural tendency to draw a figure clockwise when they start at the lower left and counter-clockwise when they start at the upper right. Tracing figures in an obligatory direction from an obligatory starting point,

which combination was predominantly incongruent, led to an equal decrease of this tendency for participants of 4 to 5 years old, 6 to 10 years old and adults (Vinter &

Perruchet, 2000). In addition, the same decrease was found for children with mental retardation and control participants matched for chronological and mental age (Vinter

& Detable, 2003).

However, conceptual difficulty cannot account for discrepant findings within the Serial Reaction Time (SRT) paradigm (Nissen & Bullemer, 1987). In this task, a stimulus appears at one location on a computer screen and participants have to respond as quickly and accurately as possible by pressing the key corresponding to that location. On most trials, the location is determined by a regular sequence. One study showed a similar decrease in reaction times on regular as opposed to random trials for children of 6 to 7 years old, 10 to 11 years old and adults (Meulemans, Van der Linden & Perruchet, 1998). Another study, however, indicated that the difference in reaction times on regular and random trials became larger for adults than for 7 to 11-year-old children (Thomas, Hunt, Vizueta, Sommer, Durston, Yang & Worden, 2004).

In summary, although most studies indicated that children perform as well as adults on implicit learning tasks, some differences have been observed. The question of whether implicit learning is invariant with age is still open. Most studies have focused on differences in the amount of knowledge acquired by adults and children.

However, there may be qualitative differences in implicit learning as well. For example, Thomas et al. (2004) suggested that the process may be slower in children than in adults. If implicit learning is subject to development, such differences are likely to exist.

In a previous study (reported in Chapter 5 of this thesis), we investigated the conditions under which implicit learning occurs. We proposed that implicit structure learning is reliably induced whenever the structure is useful to one’s current task, but not when it is useless. Two AGL-experiments with adults provided evidence in line with this prediction. The results may be different for children, however. One

possibility is that children show implicit learning even when the structure is useless.

Such a finding would suggest that implicit learning occurs automatically, but that the adults engaged in activities (e.g. hypothesis testing) that interfered with the process (Reber, 1989). Another possibility is that children fail to learn the structure even when

it is useful to their task. This would suggest that children’s attention is not reliably drawn to useful structures yet.

In the present study, adults and children performed a similar AGL-task as in our previous study, in which the grammar could be either useful or useless to the task in the induction phase. This allowed us to explore qualitative differences in the conditions under which adults and children show implicit learning as well as to test the generality of our previous finding. If implicit learning is invariant with age, both adults and children are expected to learn the structure implicitly when it is useful to their current task, but not when it is useless. If implicit learning is a developing process, however, both the amount of knowledge acquired by adults and children and the conditions in which learning is observed are likely to differ.

0 0

1

2

3

4 DIR

DIR

DIR BOG

BOG

BOG

TAF

TAF

NUP

KES KES

Figure 1. The artificial grammar used in the present study (based on Howard & Ballas, 1980).

Method Participants

The adult participants were 32 undergraduate students of Leiden University (8 male, 24 female; 18-34 years of age), who received either course credits or € 4 for their participation. The children who participated in the experiment were recruited at

two Dutch elementary schools, in Den Haag and Haarlem. Thirty-six children (16 boys, 20 girls; 10-11 years old) completed the experiment without interruptions. The data from five other children were discarded, because they went to the bathroom (1), returned to the classroom for a birthday celebration (1) or were disturbed by a third person (3) during the experiment.

Materials

The stimuli in the experiment were sentences of non-words (in Dutch) that were generated by a finite-state grammar adapted from Howard and Ballas (1980), see Figure 1. The grammar could generate 59 unique sentences of three to eight non- words. Fifteen sentences were assigned to the induction phase and 40 to the test phase, so that the paths of the grammar were equally represented in both phases of the experiment (see Appendix B). The four remaining sentences were presented as examples in the cover-story and as stimuli on the practice trials. The instructions were illustrated by a picture of an ice-cream van and by four pictures of ice-creams.

In the induction phase, each sentence was accompanied by three pictures of ice-creams. The pictures were different for the two conditions, but the sentences were the same. In the structure-useless condition, each non-word referred to a color (BOG

= yellow, DIR = red, KES = green, NUP = brown, TAF = pink). For the ice-cream the sentence referred to, the color of the balls corresponded with the meaning of the non- words. The two incorrect alternatives were created by substituting the color of one ball by one of the other colors.

In the structure-useful condition, BOG, DIR and NUP referred to the same colors as in the structure-useless condition. KES and TAF, however, had different referents: KES meant ‘with sprinkles’ and referred to the word preceding it, TAF meant ‘extra large’ and referred to the word following it. One incorrect alternative was created by either putting sprinkles on the ball following (instead of preceding) KES or enlarging the ball preceding (instead of following) TAF. The other incorrect alternative was again created by substituting the color of one ball by one of the other colors. For one sentence, that contained neither KES nor TAF, two incorrect

alternatives were created in this way.

In the test phase, one half of the stimuli were unaltered grammatical sentences.

The other half was made ungrammatical by switching two adjacent non-words in the middle of each sentence (excluding the first and last). This resulted in two types of ungrammatical sentences. Six sentences contained 3 violations of first-order

dependencies (illegal sequences of two words) and 14 sentences contained 2 first- order and 1 second-order violation (a sequence of two words that is illegal given the word preceding it). No pictures of ice-creams were presented, but each sentence was followed by a picture of a customer, who either paid or walked away. All stimuli in the experiment were displayed on a computer monitor as black text (Courier New 18, bold) against a white background. Participants were seated in front of the computer screen. They reacted by pressing keys on the keyboard.

Procedure

Adult participants were tested on a desktop computer in a test booth at Leiden University. The children were tested on a laptop in quiet rooms at their schools. All participants were tested individually. The experimenter stayed with them to check that they carried out the instruction of reading aloud, but remained out of the participants’

sight.

The experiment was introduced in a cover-story, informing participants that they would be employed at an ice-cream van on an imaginary planet, where people spoke an unfamiliar language. They were presented with four examples of sentences from this language and pictures of the corresponding ice-creams. Then, they were informed that the boss of the ice-cream van was aware that they did not speak the language yet and that they would be allowed to guess which ice-cream each client wanted. They were instructed to read aloud the sentences of 30 clients and to guess for each whether it referred to ice-cream 1, 2 or 3.

The participants first completed 2 practice trials. Then they were presented with 2 consecutive blocks of 15 experimental trials. The order of the trials was random within each block. Each trial started with the appearance of the trial number.

After 1 second, the number was replaced by a non-word sentence and three pictures of ice-creams. Participants indicated which ice-cream they thought the sentence referred to by pressing 1, 2 or 3 on the keyboard. The location of the correct ice-cream (left, center, or right) was balanced within each of the two blocks. The location where each picture appeared was balanced across blocks. After pressing a button, participants were shown the correct ice-cream for 2 seconds, accompanied by the word ‘good!’ if they had chosen that ice-cream.

After the induction phase, participants were informed that, so far, they had only seen sentences referring to ice-creams that were for sale, but that not all ice- creams were for sale on the planet. It was announced that they would be presented

with sentences from new customers. Some would order ice-creams that were for sale:

their sentences would be highly similar to the sentences presented before (i.e. the grammatical sentences). Others would order ice-creams that were not for sale: their sentences would be less similar to the sentences presented before (i.e. the

ungrammatical sentences). The participants’ task was to decide whether or not the sentence referred to an ice-cream that was for sale, by pressing a green button for yes or a red button for no. They were informed that each time they pressed the green button, they would see the customer pay and each time they pressed the red button, they would see the customer walk away.

Participants completed 40 test trials. Grammatical and ungrammatical sentences were presented in random order. On each trial, a non-word sentence was presented in the middle of the screen until the participant pressed either the green or the red button. After the participant had pressed a button, one of two pictures was shown for one second, depending on the response (but not on its accuracy). After the test phase, participants were thanked for their participation. The experiment took about 25 minutes.

Results

Induction phase

Preliminary analyses showed that 3 children and 1 adult in the structure-useful condition had been unable to carry out the task in the induction phase (they scored at chance in the first block and failed to improve in the second). The data from these participants were not included in the main analyses. To keep the group sizes equal, the data from the 3 children and 1 adult with the lowest scores on the task in the induction phase of the structure-useless condition were excluded as well.

A univariate analysis of variance (ANOVA) on the proportion of ice-creams identified correctly by the remaining 60 participants showed a significant effect of condition (F(1,56) = 47.315, MSE = .026, p < .001). Participants in the structure- useless condition (M = .778, SD = .134) more often chose the correct ice-cream than participants in the structure-useful condition (M = .494, SD = .183). In short,

identifying the correct ice-cream was easier in the structure useless than in the structure useful condition.

Test phase

A univariate ANOVA on the proportion of correct grammaticality judgments showed significant main effects of condition (F(1,56) = 9.608, MSE = .003, p = .003) and of group (F(1,56) = 4.605, MSE = .003, p = .036), but no interaction. The

proportion of correct grammaticality judgments was higher for participants in the structure-useful than in the structure-useless condition and higher for adults than for children (see Figure 2). One sample t-tests showed that, in the structure-useful

condition, the proportion of correct grammaticality judgments was significantly above chance for both adults (t(14) = 4.987, p < .001) and children (t(14) = 2.320, p = .036).

In the structure-useless condition, in contrast, the proportion of correct grammaticality judgments was at chance for both groups.

Figure 2. Mean proportion of correct grammaticality judgments with 95% confidence interval for each group and condition.

To further investigate these effects, we performed a mixed-model ANOVA on the number of correct responses with condition (structure-useful vs. structure-useless) and group (adults vs. children) as between-subject variables and type of test sentence (grammatical vs. first-order violations vs. second-order violation) as within-subjects factor. The analysis showed that the effect of group was modified by an interaction with type of test sentence (multivariate: F(2,55) = 3.966, p = .025). Independent samples t-tests showed that there was no difference in performance between adults and children for grammatical sentences (t(58) < 1) and sentences containing only first- order violations (t(58) <1). On sentences that contained a second-order violation, however, adults did better than children (t(58)= 2.430, p = .018).

Discussion

In a previous study, we found that adult participants implicitly learned the structure of an artificial grammar when it was useful to their task in the induction phase, but not when it was useless (Chapter 5). The present study replicated this finding. Moreover, it showed the same pattern of results for 10 to 11-year-old children. When the structure of non-word sentences was useful in identifying which ice-cream they referred to, children showed evidence of structure learning. When the structure was useless, because the ice-creams could be identified by the meaning of the individual non-words, however, their performance was at chance.

It could be argued that performance by children is a purer measure of implicit learning than performance by undergraduate students, which is more likely to be (positively or negatively) affected by their expectations concerning the experiment.

Therefore, it seemed possible that children would learn the structure of the non-word sentences implicitly from reading them aloud, as an automatic consequence of

processing structured stimuli (Reber, 1989). The results of the present study, however, indicated that structure learning requires selective attention to the structure (c.f.

Whittlesea & Wright, 1997; Wright & Whittlesea, 1998). For children as well as adults, the structure had to be useful to their current task to draw enough attention to be learned. This provides further evidence that usefulness of the structure is an important factor in inducing implicit learning.

Although there was no qualitative difference in the conditions under which implicit learning occurred, adults acquired more knowledge of the grammar than children. This finding could be taken as evidence that implicit learning is affected by

age. An alternative explanation, however, is that the advantage of adults over children was due to explicit learning. A previous experiment using a process dissociation procedure (Jacoby, 1991) showed that knowledge acquired from decoding non-word sentences was not as much under voluntary control as knowledge acquired from memorizing or looking for rules (Chapter 5). Nevertheless, adults may have acquired more explicit knowledge than children. In particular, the adults’ more extensive knowledge of second-order dependencies may have been explicit. Gomez (1997) showed that participants became aware of second-order dependencies, whereas first- order dependencies could be learned implicitly. This suggests that implicit learning may be invariant with age, but restricted to the acquisition of first-order dependencies.

Regardless of explicit contamination, it has also been suggested that a

difference in performance between age groups may reflect an effect of age on another cognitive process, on which implicit learning depends (Fletcher, Maybery & Bennet, 2000; Salthouse, McGuthry & Hambrick, 1999). For example, the diminished ability of elderly participants to learn higher-order dependencies in SRT-experiments has been attributed to a decrease in working memory capacity (Curran, 1997; Howard &

Howard, 1997). The number of elements that can simultaneously be held in working memory sets a limit on the order of the dependencies that can be learned (Curran, 1997). As working memory capacity continues to increase until the age of 15 (Gathercole et al., 2004), the difference in performance between adults and 10 to 11- year-old children in the present study may also reflect a difference in this capacity.

Another process that has been suggested to underlie differences in

performance on implicit learning tasks is attention (Reber & Allen, 2000; Vinter &

Detable, 2003). If implicit learning involves selective attention to useful structures, better performance by adults than by children may be related to developmental changes in selective attention. Indeed, some components of selective attention have been shown to be more effective in young adults than in children (see Plude, Enns &

Brodeur, 1994 for a review). Children are not as good as adults at selecting stimuli that are relevant to their task and they are particularly poor at maintaining attention to relevant information (Lane & Pearson, 1982; Plude et al., 1994). This suggests that children may not have performed as well as adults on the present grammaticality judgment task, because the useful structure captured less of their attention in the induction phase.

In short, the difference in artificial grammar learning between adults and children may be due to developmental changes in working memory and attention rather than to age related changes in implicit learning per se. However, this

interpretation depends on specifying the unique role of implicit learning. Perruchet, Vinter, Pacteau and Gallego (2002) suggested that implicit learning can be fully explained as an interaction of selective attention and memory. In their account of AGL, shifting attention during reading causes a letter string to be segmented into (initially arbitrary) chunks. The representations of those chunks decay unless they are repeated, but while they are active, they guide attention in segmenting new letter strings. Such a view implies that implicit learning improves in the course of development, at least on tasks that are demanding with respect to attention and working memory.

In conclusion, adults acquired more knowledge of the structure than 10 to 11- year-old children in the present study, in particular about second-order dependencies.

This difference may be due to explicit learning by the adults, which would suggest that implicit learning is invariant with age, but restricted to the acquisition of first- order dependencies. Alternatively, the finding may be due to developmental

differences in working memory and attention. In spite of the quantitative difference in performance, the conditions under which implicit learning occurred were the same.

Both adults and children learned the structure of an artificial grammar when it was useful to their current task, but not when it was useless.

Acknowledgements

We thank the staff and pupils of the St. Paschalis school in Den Haag and of a school in Haarlem for their participation in this study. We are also grateful to Saskia de Ridder for extensive pilot work.