Article Structure
Abstract
Analyses of filler-gap dependencies usually involve complex syntactic rules or heuristics; however recent results suggest that filler-gap comprehension begins earlier than seemingly simpler constructions such as ditransitives or passives.
Introduction
The phenomenon of filler-gap, where the argument of a predicate appears outside its canonical position in the phrase structure (e.g.
Background
The developmental timeline during which children acquire the ability to process filler-gap constructions is not well-understood.
Assumptions
The present work restricts itself to acquiring filler-gap comprehension in English.
Model
The model represents the preferred locations of semantic roles relative to the verb as distributions over real numbers.
Evaluation
The model in this work is trained using transcribed child-directed speech (GDS) from the BabySRL portions (Connor et al., 2008) of GHILDES (MacWhinney, 2000).
Comparison to BabySRL
The acquisition of semantic role labelling (SRL) by the BabySRL model (Connor et al., 2008; Connor et al., 2009; Connor et al., 2010) bears many similarities to the current work and is, to our knowledge, the only comparable line of inquiry to the current one.
Discussion
This paper has presented a simple cognitive model of filler-gap acquisition, which is able to capture several findings from developmental psychology.
Topics
semantic roles
- Specifically, this model, trained on part-of-speech tags, represents the preferred locations of semantic roles relative to a verb as Gaussian mixtures over real numbers.Page 1, “Abstract”
- In particular, the model described in this paper takes chunked child-directed speech as input and learns orderings over semantic roles .Page 1, “Introduction”
- This finding suggests both that learners will ignore canonical structure in favor of using all possible arguments and that children have a bias to assign a unique semantic role to each argument.Page 2, “Background”
- BabySRL is a computational model of semantic role acquistion using a similar set of assumptions to the current work.Page 2, “Background”
- to acquire semantic role labelling while still exhibiting 1-1 role bias.Page 2, “Background”
- The model presented here learns a single, non-recursive ordering for the semantic roles in each sentence relative to the verb since several studies have suggested that early child grammars may consist of simple linear grammars that are dictated by semantic roles (Diessel and Tomasello, 2001; J ackendoff and Wittenberg, in press).Page 3, “Assumptions”
- how many semantic roles it confers).Page 3, “Assumptions”
- Since infants infer the number of semantic roles , this work further assumes they already have expectations about where these roles tend to be realized in sentences, if they appear.Page 3, “Assumptions”
- These positions may correspond to different semantic roles for different predicates (e.g.Page 3, “Assumptions”
- Therefore, this work uses syntactic and semantic roles interchangeably (e.g.Page 3, “Assumptions”
- Finally, following the finding by Gertner and Fisher (2012) that children interpret intransitives with conjoined subjects as transitives, this work assumes that semantic roles have a one-to-one correspondence with nouns in a sentence (similarly used as a soft constraint in the semantic role labelling work of Titov and Klementiev, 2012).Page 3, “Assumptions”
See all papers in Proc. ACL 2014 that mention semantic roles.
See all papers in Proc. ACL that mention semantic roles.
Back to top.
role labelling
- to acquire semantic role labelling while still exhibiting 1-1 role bias.Page 2, “Background”
- Finally, following the finding by Gertner and Fisher (2012) that children interpret intransitives with conjoined subjects as transitives, this work assumes that semantic roles have a one-to-one correspondence with nouns in a sentence (similarly used as a soft constraint in the semantic role labelling work of Titov and Klementiev, 2012).Page 3, “Assumptions”
- These annotations were obtained by automatically semantic role labelling portions of CHILDES with the system of Punyakanok et al.Page 5, “Evaluation”
- The acquisition of semantic role labelling (SRL) by the BabySRL model (Connor et al., 2008; Connor et al., 2009; Connor et al., 2010) bears many similarities to the current work and is, to our knowledge, the only comparable line of inquiry to the current one.Page 6, “Comparison to BabySRL”
- The primary function of BabySRL is to model the acquisition of semantic role labelling While making an idiosyncratic error Which infants also make (Gertner and Fisher, 2012), the 1-1 role bias error (John and Mary gorped interpreted as J ohn go'r’ped M a'r’y).Page 6, “Comparison to BabySRL”
- (2008) demonstrate that a supervised perceptron classifier, based on positional features and trained on the silver role label annotations of the BabySRL corpus, manifests 1-1 role bias errors.Page 6, “Comparison to BabySRL”
- Training significantly improves role labelling in the case of object-extractions, which improves the overall accuracy of the model.Page 8, “Discussion”
See all papers in Proc. ACL 2014 that mention role labelling.
See all papers in Proc. ACL that mention role labelling.
Back to top.
lexicalized
- Since the model is not lexicalized , these roles correspond to the semantic roles most commonly associated With subject and object.Page 5, “Evaluation”
- The difference in transitive settings stems from increased lexicalization, as is apparent from their results alone; the model presented here initially performs close to their weakly lexicalized model, though training impedes agent-prediction accuracy due to an increased probability of non-canonical objects.Page 7, “Comparison to BabySRL”
- In sum, the unleXicalized model presented in this paper is able to achieve greater labelling accuracy than the lexicalized BabySRL models in intransitive settings, though this model does perform slightly worse in the less common transitive setting.Page 8, “Comparison to BabySRL”
- This could also be an area where a lexicalized model could do better.Page 8, “Discussion”
- In future, it would be interesting to incorporate lexicalization into the model presented in this paper, as this feature seems likely to bridge the gap between this model and BabySRL in transitive settings.Page 8, “Discussion”
- Lexicalization should also help further distinguish modifiers from arguments and improve the overall accuracy of the model.Page 8, “Discussion”
See all papers in Proc. ACL 2014 that mention lexicalized.
See all papers in Proc. ACL that mention lexicalized.
Back to top.
semantic role labelling
- to acquire semantic role labelling while still exhibiting 1-1 role bias.Page 2, “Background”
- Finally, following the finding by Gertner and Fisher (2012) that children interpret intransitives with conjoined subjects as transitives, this work assumes that semantic roles have a one-to-one correspondence with nouns in a sentence (similarly used as a soft constraint in the semantic role labelling work of Titov and Klementiev, 2012).Page 3, “Assumptions”
- These annotations were obtained by automatically semantic role labelling portions of CHILDES with the system of Punyakanok et al.Page 5, “Evaluation”
- The acquisition of semantic role labelling (SRL) by the BabySRL model (Connor et al., 2008; Connor et al., 2009; Connor et al., 2010) bears many similarities to the current work and is, to our knowledge, the only comparable line of inquiry to the current one.Page 6, “Comparison to BabySRL”
- The primary function of BabySRL is to model the acquisition of semantic role labelling While making an idiosyncratic error Which infants also make (Gertner and Fisher, 2012), the 1-1 role bias error (John and Mary gorped interpreted as J ohn go'r’ped M a'r’y).Page 6, “Comparison to BabySRL”
See all papers in Proc. ACL 2014 that mention semantic role labelling.
See all papers in Proc. ACL that mention semantic role labelling.
Back to top.
error rate
- Error rate Initial .36 Trained .11 Initial (given 2 args) .66 Trained (given 2 args) .13 2008 arg—arg position .65 2008 arg-verb position 0 2009 arg—arg position .82 2009 arg-verb position .63Page 7, “Comparison to BabySRL”
- The model presented in this paper does not share this restriction, so the raw error rate for this model is presented in the first two lines; the error rate once this additional restriction is imposed is given in the second two lines.Page 7, “Comparison to BabySRL”
- The 1-1 role bias error rate (before training) of the model presented in this paper is comparable to that of Connor et al.Page 7, “Comparison to BabySRL”
See all papers in Proc. ACL 2014 that mention error rate.
See all papers in Proc. ACL that mention error rate.
Back to top.
significant improvement
- This slight, though significant in Eve, deficit is counterbalanced by a very substantial and significant improvement in object-extraction labelling accuracy.Page 6, “Evaluation”
- Similarly, training confers a large and significant improvement for role assignment in wh-relative constructions, but it yields less of an improvement for that-relative constructions.Page 6, “Evaluation”
- Training significantly improves role labelling in the case of object-extractions, which improves the overall accuracy of the model.Page 8, “Discussion”
See all papers in Proc. ACL 2014 that mention significant improvement.
See all papers in Proc. ACL that mention significant improvement.
Back to top.
