Article Structure
Abstract
Probabilistic models of sentence comprehension are increasingly relevant to questions concerning human language processing.
Introduction
Probabilistic grammars have been found to be useful for investigating the architecture of the human sentence processing mechanism (J urafsky, 1996; Crocker and Brants, 2000; Hale, 2003; Boston et al., 2008; Levy, 2008; Demberg and Keller, 2009).
Cognitive Experiments 2.1 Discourse and Ambiguity Resolution
There is a fairly large literature on garden path experiments involving context (Crain and Steedman, 1985; Mitchell et al., 1992, ibid).
Model
The model comprises three parts: a parser, a coreference resolution system, and a pragmatics subsystem.
Evaluation 4.1 Method
When modeling the garden path experiment we presented in Section 2.1, we compute Surprisal values on the word ‘by’ , which is the earliest point at which there is evidence for a relative clause interpretation.
Conclusions
The main result of this paper is that it is possible to produce a Surprisal-based sentence processing model which can simulate the influence of discourse on syntax in both garden path and unambiguous sentences.
Topics
coreference
- This paper introduces a novel sentence processing model that consists of a parser augmented with a probabilistic logic-based model of coreference resolution, which allows us to simulate how context interacts with syntax in a reading task.Page 1, “Abstract”
- This is the first model we know of which introduces a broad-coverage sentence processing model which takes the effect of coreference and discourse into account.Page 1, “Introduction”
- There are three main parts of the model: a syntactic processor, a coreference resolution system, and a simple pragmatics processor which computes certain limited forms of discourse coherence.Page 1, “Introduction”
- The coreference resolution system is implementedPage 1, “Introduction”
- The model comprises three parts: a parser, a coreference resolution system, and a pragmatics subsystem.Page 3, “Model”
- However, as the coreference processor takes trees as input, we must therefore unpack parses before resolving referential ambiguity.Page 4, “Model”
- the agent), get the -LGS label; (iv) non-recursive NPs are renamed NPbase (the coreference system treats each NPbase as a markable).Page 4, “Model”
- The primary function of the discourse processing module is to perform coreference resolution for each mention in an incrementally processed text.Page 4, “Model”
- Because each mention in a coreference chains is transitive, we cannot use a simple classifier, as they cannot enforce global transitivity constraints.Page 4, “Model”
- Coref (x, y) c is coreferent with y.Page 4, “Model”
- Two of these predicates, Coref and First, are the output of the MLN — they provide a labelling of coreference mentions into entity classes.Page 4, “Model”
See all papers in Proc. ACL 2010 that mention coreference.
See all papers in Proc. ACL that mention coreference.
Back to top.
coreference resolution
- This paper introduces a novel sentence processing model that consists of a parser augmented with a probabilistic logic-based model of coreference resolution , which allows us to simulate how context interacts with syntax in a reading task.Page 1, “Abstract”
- There are three main parts of the model: a syntactic processor, a coreference resolution system, and a simple pragmatics processor which computes certain limited forms of discourse coherence.Page 1, “Introduction”
- The coreference resolution system is implementedPage 1, “Introduction”
- The model comprises three parts: a parser, a coreference resolution system, and a pragmatics subsystem.Page 3, “Model”
- The primary function of the discourse processing module is to perform coreference resolution for each mention in an incrementally processed text.Page 4, “Model”
- Note that, unlike Huang eta1., we assume an ordering on c and y if Coref(x, y) is true: 3/ must occur earlier in the document than c. The remaining predicates in Table 1 are a subset of features used by other coreference resolution systems (cf.Page 4, “Model”
- As our main focus is not to produce a state-of-the-art coreference system, we do not include predicates which are irrevelant for our simulations even if they have been shown to be effective for coreference resolution .Page 4, “Model”
- To test that the coreference resolution system was producing meaningful results, we evaluated our system on the test section of the ACE-2 dataset.Page 5, “Model”
- The effect of context in the experiments described in Section 2 cannot be fully explained using a coreference resolution system alone.Page 5, “Model”
See all papers in Proc. ACL 2010 that mention coreference resolution.
See all papers in Proc. ACL that mention coreference resolution.
Back to top.
probabilistic modeling
- Probabilistic models of sentence comprehension are increasingly relevant to questions concerning human language processing.Page 1, “Abstract”
- For example, probabilistic models shed light on so-called locality effects: contrast the non-probabilistic hypothesis that dependants which are far away from their head always cause processing difficulty for readers due to the cost of storing the intervening material in memory (Gibson, 1998), compared to the probabilistic prediction that there are cases when faraway dependants facilitate processing, because readers have more time to predict the head (Levy, 2008).Page 1, “Introduction”
- So far, probabilistic models of sentence processing have been largely limited to syntactic factors.Page 1, “Introduction”
- In the literature on probabilistic modeling , though, the bulk of this work is focused on lexical semantics (e.g.Page 1, “Introduction”
- Although our main results above underscore the usefulness of probabilistic modeling , this observation emphasizes the importance of finding a tenable link between probabilities and behaviours.Page 9, “Conclusions”
See all papers in Proc. ACL 2010 that mention probabilistic modeling.
See all papers in Proc. ACL that mention probabilistic modeling.
Back to top.
entity type
- c has entity type 6 (person, organization, etc.)Page 4, “Model”
- Entity Type (x, e)Page 4, “Model”
- The predicates we use involve matching strings (checking if two mentions share a head word or if they are exactly the same string), matching argreement features (if the gender, number or person of pairs of NPs are the same; especially important for pronouns), the distance between mentions, and if mentions have the same entity type (i.e.Page 4, “Model”
See all papers in Proc. ACL 2010 that mention entity type.
See all papers in Proc. ACL that mention entity type.
Back to top.
