Conditional Random Fields for Responsive Surface Realisation using Global Features
Dethlefs, Nina and Hastie, Helen and Cuayáhuitl, Heriberto and Lemon, Oliver

Article Structure

Abstract

Surface realisers in spoken dialogue systems need to be more responsive than conventional surface realisers.

Introduction

Surface realisation typically aims to produce output that is grammatically well-formed, natural and cohesive.

Related Work

Our approach is most closely related to Lu et al.

Cohesion across Utterances

3.1 Tree-based Semantic Representations

Evaluation

To evaluate our approach, we focus on a subjective human rating study which aims to determine whether CRF-based surface realisation that takes the full generation context into account, called CRF (global), is perceived better by human judges than one that uses a CRF but just takes local context into account, called CRF (local).

Incremental Surface Realisation

Recent years have seen increased interest in incremental dialogue processing (Skantze and Schlangen, 2009; Schlangen and Skantze, 2009).

Conclusion and Future Directions

We have presented a novel technique for surface realisation that treats generation as a sequence labelling task by combining a CRF with tree-based semantic representations.

Topics

CRF

Appears in 27 sentences as: CRF (32)
In Conditional Random Fields for Responsive Surface Realisation using Global Features
  1. Our main hypothesis is that the use of global context in a CRF with semantic trees can lead to surface realisations that are better phrased, more natural and less repetitive than taking only local features into account.
    Page 1, “Introduction”
  2. This paper focuses on surface realisation from these trees using a CRF as shown in the surface realisation module.
    Page 3, “Cohesion across Utterances”
  3. As shown in the architecture diagram in Figure 1, a CRF surface realiser takes a semantic tree as input.
    Page 3, “Cohesion across Utterances”
  4. Figure 2: (a) Graphical representation of a linear-chain Conditional Random Field (CRF), where empty nodes correspond to the labelled sequence, shaded nodes to linguistic observations, and dark squares to feature functions between states and observations; (b) Example semantic trees that are updated at each time step in order to provide linguistic features to the CRF (only one possible surface realisation is shown and parse categories are omitted for brevity); (c) Finite state machine of phrases (labels) for this example.
    Page 3, “Cohesion across Utterances”
  5. thl} is a set of production rules of the form n —> 04, where n E N, 04 E T U N. The production rules represent alternatives at each branching node where the CRF is consulted for the best available expansion from the subset of possible ones.
    Page 3, “Cohesion across Utterances”
  6. We use the linear-chain Conditional Random Field ( CRF ) model for statistical phrase-based surface realisation, see Figure 2 (a).
    Page 4, “Cohesion across Utterances”
  7. The following features define the generation context used during training of the CRF .
    Page 4, “Cohesion across Utterances”
  8. In this way, a sequence of surface form constituents is generated corresponding to latent states in the CRF .
    Page 4, “Cohesion across Utterances”
  9. If the generation history already contains a semantic attribute, e. g. the restaurant name, the CRF may afterwards choose a pronoun, e.g.
    Page 4, “Cohesion across Utterances”
  10. Similarly, the CRF may decide to realise a new attribute as constituents of different order, such as a sentence or PP, depending on the length, number and parse categories of previously generated output.
    Page 4, “Cohesion across Utterances”
  11. To train the CRF , we used a data set of 552 restaurant recommendations from the website The
    Page 4, “Cohesion across Utterances”

See all papers in Proc. ACL 2013 that mention CRF.

See all papers in Proc. ACL that mention CRF.

Back to top.

CRFs

Appears in 10 sentences as: CRFs (10)
In Conditional Random Fields for Responsive Surface Realisation using Global Features
  1. We formulate surface realisation as a sequence labelling task and combine the use of conditional random fields ( CRFs ) with semantic trees.
    Page 1, “Abstract”
  2. Due to their extended notion of context, CRFs are able to take the global utterance context into account and are less constrained by local features than other realisers.
    Page 1, “Abstract”
  3. to surface realisation within incremental systems, because CRFs are able to model context across full as well as partial generator inputs which may undergo modifications during generation.
    Page 2, “Introduction”
  4. (2009) who also use CRFs to find the best surface realisation from a semantic tree.
    Page 2, “Related Work”
  5. This grammar defines the surface realisation space for the CRFs .
    Page 5, “Cohesion across Utterances”
  6. CRFs and other state-of-the-art methods, we also compare our system to two other baselines:
    Page 5, “Evaluation”
  7. Since CRFs are not restricted by the Markov condition, they are less constrained by local context than other models and can take nonlocal dependencies into account.
    Page 8, “Incremental Surface Realisation”
  8. While their extended context awareness can often make CRFs slow to train, they are fast at execution and therefore very applicable to the incremental scenario.
    Page 8, “Incremental Surface Realisation”
  9. We have argued that CRFs are well suited for this task because they are not restricted by independence assumptions.
    Page 9, “Conclusion and Future Directions”
  10. In addition, we may compare different sequence labelling algorithms for surface realisation (Nguyen and Guo, 2007) or segmented CRFs (Sarawagi and Cohen, 2005) and apply our method to more complex surface realisation domains such as text generation or summarisation.
    Page 9, “Conclusion and Future Directions”

See all papers in Proc. ACL 2013 that mention CRFs.

See all papers in Proc. ACL that mention CRFs.

Back to top.

sequence labelling

Appears in 5 sentences as: sequence labelling (5)
In Conditional Random Fields for Responsive Surface Realisation using Global Features
  1. We formulate surface realisation as a sequence labelling task and combine the use of conditional random fields (CRFs) with semantic trees.
    Page 1, “Abstract”
  2. In this paper, we propose to formulate surface realisation as a sequence labelling task.
    Page 1, “Introduction”
  3. The main idea of our approach is to treat surface realisation as a sequence labelling task in which a sequence of semantic inputs needs to be labelled with appropriate surface realisations.
    Page 3, “Cohesion across Utterances”
  4. We have presented a novel technique for surface realisation that treats generation as a sequence labelling task by combining a CRF with tree-based semantic representations.
    Page 9, “Conclusion and Future Directions”
  5. In addition, we may compare different sequence labelling algorithms for surface realisation (Nguyen and Guo, 2007) or segmented CRFs (Sarawagi and Cohen, 2005) and apply our method to more complex surface realisation domains such as text generation or summarisation.
    Page 9, “Conclusion and Future Directions”

See all papers in Proc. ACL 2013 that mention sequence labelling.

See all papers in Proc. ACL that mention sequence labelling.

Back to top.

Grammar Induction

Appears in 4 sentences as: Grammar Induction (2) grammar induction (2)
In Conditional Random Fields for Responsive Surface Realisation using Global Features
  1. 3.4 Grammar Induction
    Page 5, “Cohesion across Utterances”
  2. The grammar g of surface realisation candidates is obtained through an automatic grammar induction algorithm which can be run on unlabelled data and requires only minimal human intervention.
    Page 5, “Cohesion across Utterances”
  3. We provide the human corpus of restaurant recommendations from Section 3.3 as input to grammar induction .
    Page 5, “Cohesion across Utterances”
  4. Algorithm 1 Grammar Induction .
    Page 5, “Evaluation”

See all papers in Proc. ACL 2013 that mention Grammar Induction.

See all papers in Proc. ACL that mention Grammar Induction.

Back to top.

n-grams

Appears in 4 sentences as: n-grams (4)
In Conditional Random Fields for Responsive Surface Realisation using Global Features
  1. In addition, we compare our system with alternative surface realisation methods from the literature, namely, a rank and boost approach and n-grams .
    Page 1, “Introduction”
  2. 0 n-grams represents a simple 5- gram baseline that is similar to Oh and Rudnicky (2000)’s system.
    Page 5, “Evaluation”
  3. CRF global 3.65 3.64 3.65 CRF local 310* 319* 313* CLASSiC 353* 3.59 348* n-grams 301* 309* 332*
    Page 6, “Evaluation”
  4. This difference is significant for all categories compared with CRF (local) and n-grams (using a 1-sided Mann Whitney U-test, p < 0.001).
    Page 6, “Evaluation”

See all papers in Proc. ACL 2013 that mention n-grams.

See all papers in Proc. ACL that mention n-grams.

Back to top.

Semantic Representations

Appears in 3 sentences as: semantic representation (1) Semantic Representations (1) semantic representations (1)
In Conditional Random Fields for Responsive Surface Realisation using Global Features
  1. 3.1 Tree-based Semantic Representations
    Page 2, “Cohesion across Utterances”
  2. In this way, each nonterminal symbol has a semantic representation and an associated parse category.
    Page 5, “Cohesion across Utterances”
  3. We have presented a novel technique for surface realisation that treats generation as a sequence labelling task by combining a CRF with tree-based semantic representations .
    Page 9, “Conclusion and Future Directions”

See all papers in Proc. ACL 2013 that mention Semantic Representations.

See all papers in Proc. ACL that mention Semantic Representations.

Back to top.

PDF available from www.aclweb.org.
Send us an email