This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.
A central problem in the field of language acquisition is how children learn to use an adult like production of verbs without creating overgeneralization errors such as This'll hot it up (from Bowerman, 1982) which uses the intransitive verb hot in a transitive construction. These kinds of overgeneralization errors have been observed in children from around the age of 3 up to the age of around 7 years in their speech (Bowerman, 1982, Pinker, 1989) and are often judged as grammatical by the children themselves (Hochberg, 1986). These are not the only overgeneralizations children make however, for example, in locative constructions it is deemed acceptable to say Adam poured water into the cup or I filled the cup with water but not I poured the cup with water or Adam filled water into the cup. In order for a verb to be used in a locative construction it must be formed as affecting the direct object by causing it to change its state for verbs such as fill, or a change of location for verbs as pour (Gropen et al 1991) yet children readily violate these rules.
A number of theories have been put forward in an attempt to explain how children avoid making overgeneralization errors and eventually gain an adult like grammar. Negative evidence is one such possibility. It suggests that children use verbs in all syntactic constructions and then are corrected by parents in order to learn what constructions are considered to be grammatical and which are not. This kind of feedback may alert children to some errors (Saxton, 2000) however it does not seem plausible to explain a total retreat from overgeneralizations as feedback appears to be too weak (Marcus, 1993), and in many cases this feedback isn't given by parents when overgeneralizations occur.
Another such explanation is pre-emption; this suggests that children will create errors in their spontaneous speech if they have not yet acquired an alternative construction to express the intended meaning of their utterance. So children will be likely to make errors such as This'll hot it up until they acquire the phrase This will heat it up to express the intended meaning (Brain & Brooks, 1995). Experimental evidence for pre-emption supports its claim, Brooks and Tomasello (1999) taught children aged 2, 4 and 6-7 years two novel verbs that were either modelled in either transitive or intransitive constructions, with half of the children were presented with pre-empting constructions. The results show no significant pre-emption effects in the 2 and 4 year old groups with only the oldest group having any effects from being presented with the pre-empting constructions. The 2 year old group actually showed a significant trend in the opposite direction. These results show that pre-emption does have an effect on retreat from overgeneralization errors yet is only available in later development when much of the retreat has already occurred.
Many of the overgeneralizations are easily overcome, for example past-tense overregularizations such as go-ed. When children learn the correct past tense, in this case went, children cease making the error via processes like pre-emption. Some more problematic overgeneralizations, such as argument-structure overgeneralization errors (e.g. *The clown giggled the boy) are more difficult to explain as children cannot simply exchange the sentence with competing input.
A recent study by Ambridge, Pine, Rowland, and Young (2008) has given evidence for two alternative explanations of how children retreat from overgeneralization which have extensive support in the literature: the entrenchment hypothesis (Brain and Brooks, 1995) and the semantic class hypothesis (pinker, 1989).
According to the entrenchment hypothesis (Braine & Brooks, 1995) children retreat from overgeneralization errors and gain an adult like grammar based on the distributional probability of the input data. The likelihood that children will produce an overgeneralization with a verb is linked to the frequency that a particular verb is heard in a range of constructions in the input. In other words, the more often a child hears a verb in a construction in its input the less likely the child will produce an overgeneralization error when creating novel constructions using that verb. As children develop the amount of input they receive is expected to increase, incorporating the majority of verbs, which increases the familiarity of individual verbs, reducing the number of errors produced in spontaneous speech to the levels of adult grammar.
Brooks, Tomasello, Dodson and Lewis (1999) provided support for this hypothesis as children aged between 3; 4 and 9 were less likely to produce errors violating the transitivity status of higher frequency verbs (e.g. come and take)compared to lower frequency verbs of similar semantic meaning (e.g. arrive and remove).
Theakston (2004) conducted a similar frequency test using a grammaticality-judgement paradigm. Adults and children (aged 5:1 - 9:2) were asked to rate sentences containing argument-structure overgeneralization errors with semantically matched high and how frequency verbs (e.g. *Shall i shout/whisper you something, *How do you unsquash/unsqueeze it?). Using a binary scale for children and a seven point scale for adults the results showed that both groups rated overgeneralization errors more acceptable for lower frequency verbs than their higher frequency counterpart.
Ambridge et al (2008) conducted a grammaticality-judgement paradigm using a novel five point 'smiley face' scale on adults and children (aged 5;3-6;4, 9;3-10;3). Using high, low and novel frequency verbs for the semantic classes of ''directed motion'', ''going out of existence'' and ''semi voluntary expression of emotion' (from Pinker, 1989a: 303) in intransitive and transitive causatives. The results showed the pattern expected by the entrenchment hypothesis. However the results lacked a significant interaction between verb frequency and age, suggesting that verb frequency had a similar effect on all age groups judgments throughout development. This is not what the hypothesis expected as the adult group should have had larger frequency effects than the younger groups due to being exposed to more input data. Thus, this would reinforce the probability of sentences being acceptable compared to the younger groups much weaker probabilistic inference abilities. The results also showed a preference for the intransitive over transitive causative uses of novel verbs which cannot be explained by entrenchment, further more why this effect should increase with age.
One aim of the presented study...
Pinker (1989) put forward the verb-semantics hypothesis, suggesting that verbs appear in specific syntactic constructions which define its semantic class. It is these semantic classes that gradually reduce children's overgeneralization errors as they are formed through development. Once a child has correctly established the meaning of the verbs and which narrow semantic class it belongs to overgeneralizations should cease to be created with that specific verb.
Evidence from Brooks and Tomasello (1999) is consistent with this theory. By teaching two novel verbs; one with a semantic class that can be used in both transitive and intransitive constructions
DRAFTand one which can only be used in a transitive construction (to two, four and six to seven year olds), it would show the effect of semantic verbs classes. The results showed that children did observe the transitive construction for the transitively fixed verb and would produce both transitive and intransitive utterances for the verb that allowed it. However this effect was not present in children younger than 4.5 years of age, suggesting that the narrow semantic rules doe help children but are only present later in development so children must use an alternative mechanism until then.
Similar results were found by Ambridge et al (2008) when teaching three novel verbs (Tamm, Fudd, Meek, each representing a verb with semantics representing laughing, falling and disappearing respectively) with semantics only consistent with intransitive constructions. Adults and older children confirmed the semantic verb prediction for all three verbs. Younger children however only followed the prediction for the novel verb representing laughing and not the other two novel verbs. It was argued that this was due to the novel laughing verb being more inherently an internal action compared to falling or disappearing.
One aim of the presented study...
The entrenchment hypothesis and the semantic verb class hypothesis are both well supported in the literature yet weaknesses are inherent in each. Neither account can explain the findings that are clearly visible in the other. Both accounts clearly are important for reduction of overgeneralizations and both frequency and verb semantics must be considered when looking at both children's and adults judgements (Theakston, 2004).
The findings in Ambridge et al (2008) that laughing verbs were rated less acceptable than disappearing or falling verbs regardless of both being ungrammatical suggest that the constructions (transitive or intransitive) verbs are in will affect how grammatical they are perceived to be. This idea suggests that verbs appear to lie on a scale with respect to the extent to which they are semantically consistent with the meaning of the construction they are in. Ambridge, Pine, Rowland, Jones and Clark (2009) suggests that for transitive causative constructions the scale will depend on how the verbs appears to describe an action that is caused by an external agent, where as for intransitive prohibited constructions the verb must denote an action that has an internal cause.
The ILVACS account explains this occurrence being due to an incremental acquisition of verb semantics and constructions. In layman terms, children would learn the actions that each verb
DRAFTdescribes and the relevant sentence constructions that the verb is allowed to be used in. This would decrease overgeneralization errors to the same levels of adult errors as the children understand which verb semantics are associated with which constructions and any mismatching errors would be noticed and would cease to be produced.
The semantics of the verbs and constructions are in this account learned incrementally by the use of an entrenchment mechanism. This mechanism is not a statistical tally as that by Braine and Brooks (1995) but way of increasing understanding of a verb by increasing the semantic representation of the action the verb describes. This would also show effects that could be explained in the form of semantic verb classes as verbs with similar semantic meaning will inevitably be compatible within a similar set of constructions.
The presented study was designed to test out the validity of the ILVACS account...
Participants consisted of 90 monolingual speakers of British English; 30 ages 5-6 (Mean age = NN), 30 aged 9-10 (M= NN) and 30 adults (all undergraduate students, mostly aged 18-19).
The experiment used a correlation design with a separate regression analysis conducted for each age group. There were 20 verbs of each type (alternating, contents-only and container-only). Each verb was rated in the contents and the container locative construction. Adults completed these rating for all 60 verbs. Each child rated 1/3 of the total number of verbs as it was unlikely children could have completed 120 trials.
The experiment used two predictor variables
The extent to which each verb was rated as describing "something that happens to A more then B" as rated by 10 independent adults. A and B were defined as follows
All the verbs in this checklist describe scenarios where liquid, mass, substance, small item, or a grouped of small items (e.g., water, paint, seeds, juice, nails, paper cloths, bandages, stickers, bullets) moves to/into/onto/around a container, surface or location (e.g., a cup, a bin, a suitcase, a shelf, a box, a sink, a filed, a table, the floor, a sponge). In what follows we use A to denote the mass/substance/small items/group of small items and B to denote the container/surface/location.
Thus higher numbers mean "doing something to the contents", lower numbers mean "doing something to the container". This semantic rating was obtained for each of 60 verbs - 1/3 of which are contents-only, 1/3 container only and 1/3 alternating.
The frequency of each verb in the British National Corpus (a corpus of adult-adult speech). We assume that verbs occur (almost) exclusively in the constructions in which they are permitted (i.e., there are no overgeneralization errors). For example contents-locative-only verbs will not appear in container locative sentences; they will appear in contents locatives (and also simple in/transitives).
DV1 Mean rating for the contents-locative
DV2 Mean rating for the container-locative
For some analysis a difference score is used instead of these two. Difference score is mean preference for the contents-over the container locative.
Animations were created and presented using Microsoft powerpoint 2007, running on a laptop PC. Stereo audio recordings were created and implemented using one speaker (hidden inside a "talking dog" toy) to play the recorded sentences, except for adults which received the sentences directly. This was to allow the experimenter to engage the younger children in the task in a format similar to a game in which the children attempted to help the "talking dog" learn to speak English.
The children were first introduced to the 'talking dog' toy and his attempt to learn to speak English and they would play a game. The experimenter explained that the dog was going to say what was visible on the adjacent laptops screen and that the child's task would be to 'help' the 'talking dog' by using a five point 'face scale'. By using the five point face scale* the children would rate how grammatical the sentence produced by the talking dog was.
Five trial sentences were used to introduce the children to the game of rating the dog's sentences. If a child rated a test sentence incorrectly the experimenter would prompt* the child to change their answer and explained why. Subjects who reliably answered the test sentences as expected were then shown the test sentences.
The test phase was split into two sessions of twenty three test sentences. Both sessions had the trial sentences used before the test phases to ensure the child fully understood the task. Each rating was then noted down for each test sentence.
<ADD MORE DETAIL>
A preliminary analysis was conducted to determine if the participants were able to correctly rate the verb forms as expected. All participant ratings should show a significant preference for contents over container locatives for verbs that are contents-only, container over contents locatives for verbs that are container-only and neither for verbs that alternate.
The data was entered into a 3x2x3 mixed analysis by items with within-item factors of age (5-6, 9-10, adult) and sentence type (contents vs. container) and between-item factor of verb-type (contents-only, container-only and alternating). DV = mean rating (collapsing across all participants at each age group).
DRAFTAll main effects and interaction are significant at p<0.01 or better, however many of these interactions are uninterpritable so only the interaction of age by verb-type by sentence type (F (4, 114) = 27.65; PE= 0.49, p < 0.001) was considered.
ADD TABLE OF MEANS!!!!
Post-hoc tests revealed that at age group 5-6; contents-locatives (M=4.06 (0.09)) were significantly preferred over container-locatives (M= 3.42(0.11)) for contents-only verbs (p<0.001) and container-locatives (M= 3.89 (0.11)) were significantly preferred over contents locatives (M=3.563(0.9)) for container-only verbs (p< 0.02). Age group 9-10 showed contents-locatives (M=4.56 (0.13)) were significantly preferred over container-locatives (M= 2.81(0.16)) for contents-only verbs (p<0.001) and container-locatives (M= 4.13 (0.16)) were significantly preferred over contents locatives (M=3.26(0.13)) for container-only verbs (p< 0.001). Adults showed contents-locatives (M=6.24 (0.15)) were significantly preferred over container-locatives (M= 3.37(0.25)) for contents-only verbs (p<0.001) and container-locatives (M= 5.77 (0.25)) were significantly preferred over contents locatives (M=4.247(0.15)) for container-only verbs (p< 0.001). As predicted there is no difference for alternating verbs, although adults show a trend towards preferring the contents- over container-locative. Thus it appears that both younger and adult participants were able to correctly rate verb forms using the provided scale.
Testing the ILVACS account
The ILVACS account predicts a correlation between (1) extent to which event is something that "happens to the contents as opposed to the container" and (2) preference for contents over container locative (difference score)
POSSIBLY ADD CORRELATION TABLE
Conducting a Pearson correlation showed that these predictions are correct. The test was significant for the youngest child group [r=0.357, n=60, p= 0.005], oldest child group [r=0.543, n=60, p<0.001] and adults [r=0.633, n=60, p, 0.001].
Whilst the correlations are significant it is difficult to explain a large portion of the variance (year 1 R^2 = 0.13, year 5 R^2 = 0.30, adult R^2= 0.40). An apparent reason for this appears to be that the semantic ratings don't capture the semantic differences between contents-only, container-only and alternating verbs. Looking at the ratings for the contents-only verbs, the ratings for "happening to the contents more than the container" are considerably lower than expected (rated 3.18 as opposed to the expected 9).
According to this account another reason for the lack of verb frequency is that verb frequency is important. This is due to both entrenchment effects and because high frequency verbs have better learned semantics associated with them.
The ILVACS account predicts a significant positive correlation between the frequency of each contents-locative only verb in the British national corpus in contents-locative sentences (and in simple in/transitive sentences) and the preference for contents over container locatives.
The account also predicts a significant negative correlation between the frequency of each container-locative only verb in the BNC and the preference for contents over container locatives. This is because of the way the difference scores are calculated.
Testing the first prediction a Pearson's correlation was run. Looking at the difference scores the predictions were tested.
For year 1 children the correlation was not significant between contents-locative only verbs and preference for contents over container locative [R=0.200, R^2=0.040, P=0.397]. Year 5 children the correlation was significant between contents-locative only verbs and preference for contents over container locative [R=0.448, R^2=0.201, P=0.048]. For adults the correlation was significant between contents-locative only verbs and preference for contents over container locative [R=0.621, R^2=0.386, P=0.004]
Testing the second prediction a Pearson's correlation was run using the difference scores.
For year 1 children the correlation was not significant between container-locative only verbs and the preference for contents over container locatives [R=-0.228, R^2=0.052 ,P=0.333].For year 5 children the correlation was significant between container-locative only verbs and the preference for contents over container locatives [R=-0.463, R^2=0.214 ,P=0.40].For adults the correlation was significant between container-locative only verbs and the preference for contents over container locatives [R=-0.446, R^2=0.199 ,P=0.49].
The ILVACS hypothesis also predicts a negative correlation between: frequency of contents-only verbs and the raw acceptability of container locatives, and the frequency of container-only verbs and the raw acceptability of contents locatives.
Testing the prediction of a negative correlation between frequency of contents-only verbs and the raw acceptability of container locatives the 3 age groups were considered individually. For the year 1 group this correlation was not significant [R=-0.278, R^2=0.077, P=0.235]. For year 5 this correlation was not significant [R=-0.289, R^2=0.084, P=0.216]. For the adult group this correlation was significant [R=-0.542, R^2=0.294, P=0.014].
Testing the prediction of a negative correlation between the frequency of container-only verbs and the raw acceptability of contents locatives a Pearson's correlation was run. For the year 1 groups this correlation was not significant [R=-0.065, R^2=0.004, P=0.786]. For the year 5 groups this correlation was significant [R=-0.444, R^2=0.197, P=0.50]. For the adult group this correlation was not significant [R=-0.122, R^2=0.015, P=0.609].
As predicted by the ILVACS account both verb semantics and frequency are significant indicators of "preference for A over B". In order to find out which indicator is a better predictor a regression analysis was run. This was a forced-entry regression with both predictors entered simultaneously.
For the 5-6 year old group the overall model isn't significant for either container-only [R= 0.25, R^2=0.63, p=0.576] or contents-only [R=0.2, R^2 = 0.4, P= 0.706] so no further analysis was conducted.
For 9-10 year olds the overall model is only significant for the contents-only verbs [R=0.562, R^2 = 0.316, p =0.40] and not significant for container-only verbs [R=0.488, R^2=0.238 P=0.99].
As contents-only verbs are significant further analysis is required.
From a <INSERT TEST NAME> test it is clear that frequency is a better predictor than semantics. Frequency [B=0.664, SE=0.269, Beta=0.501, t=2.469, P=0.24] Semantics [B=0.203, SE=0.12, Beta=0.344, t=1.693, P=0.109]
For adults the overall model is significant for both container-only [R=0.666, R^2=0.443, P=0.007] and content-only verbs [R=0.697, R^2=0.486, P=0.003] individually.
From a <INSERT NAME> test it is clear that for the container-only verbs semantics is a better predictor for adults [B=0.547, SE=0.2, Beta=0.515, t=2.735, P=0.014]than frequency [B=-0.467, SE=0.289, Beta=-0.304, t=-1.615, P=0.125]
For content-only verbs frequency is the better predictor for adults [B=1.482, SE=0.389, Beta=0.670, t=3.810, P=0.001] then semantics [B=0.317, SE=0.173, Beta=3.22, t=1.831, P=0.085].
Overall it is clear from the results that both verb frequency and semantics seem to be important in the retreat from overgeneralization errors, as predicted by the ILVACS hypothesis.
The entrenchment hypothesis is supported to an extent by the results but it can't account for the semantics effects observed. Similarly an account based solely on semantics can't explain all the results observed so an account like the ILVACS hypothesis should be considered