Probabilistic Labeling for Efficient Referential Grounding based on Collaborative Discourse
Liu, Changsong and She, Lanbo and Fang, Rui and Chai, Joyce Y.

Article Structure

Abstract

When humans and artificial agents (e.g.

Introduction

In situated human-robot dialogue, humans and robots have mismatched capabilities of perceiving the shared environment.

Related Work

Previous works on situated referential grounding have mainly focused on computational models that connect linguistic referring expressions to the perceived environment (Gomiak and Roy, 2004; Gor-niak and Roy, 2007; Siebert and Schlangen, 2008; Matuszek et al., 2012; J ayant and Thomas, 2013).

Data

Previously, we have collected a set of human-human dialogues on an object-naming task (Liu et al., 2012).

Probabilistic Labeling for Reference Grounding

4.1 System Overview

Evaluation and Discussion

Our dataset has 62 dialogues, each of which contains an average of 25 valid utterances from the director.

Conclusion

In this paper, we have presented a probabilistic labeling based approach for referential grounding in situated dialogue.

Topics

coreference

Appears in 15 sentences as: coref (2) coreference (18)
In Probabilistic Labeling for Efficient Referential Grounding based on Collaborative Discourse
  1. Our system first processes the data using automatic semantic parsing and coreference resolution.
    Page 2, “Probabilistic Labeling for Reference Grounding”
  2. We then perform pairwise coreference resolution on the discourse entities to find out the discourse relations between entities from different utterances.
    Page 2, “Probabilistic Labeling for Reference Grounding”
  3. Based on the semantic parsing and pairwise coreference resolution results, our system further builds a graph representation to capture the collaborative discourse and formulate referential grounding as a probabilistic labeling problem, as described next.
    Page 3, “Probabilistic Labeling for Reference Grounding”
  4. It is constructed based on the semantic parsing and coreference resolution results.
    Page 3, “Probabilistic Labeling for Reference Grounding”
  5. coreference ) between the two nodes.
    Page 3, “Probabilistic Labeling for Reference Grounding”
  6. The edge attributes can be either from parsing results, such as a spatial relation between two entities (e.g., RightO f (a1, a2)); Or from pairwise coreference resolution results, i.e., two entities are coreferen-tial (coref = +) or not ( coref = —).
    Page 3, “Probabilistic Labeling for Reference Grounding”
  7. using annotated 66.4% 74.8% 81.9% coreference
    Page 4, “Probabilistic Labeling for Reference Grounding”
  8. using manually annotated coreference .
    Page 4, “Probabilistic Labeling for Reference Grounding”
  9. If an edge is used to encode the discourse relation between two entities (i.e., the pairwise coreference results), the compatibility coefficient can be defined as (suppose edge CLin encodes a positive coreference relation between entities CL, and aj):
    Page 4, “Probabilistic Labeling for Reference Grounding”
  10. which can be calculated based on the results from the coreference classifier (Section 4.1).
    Page 4, “Probabilistic Labeling for Reference Grounding”
  11. We first applied the semantic parser and coreference classifier as described in Section 4.1 to process each dialogue, and then built a graph representation based on the automatic processing results at the end of the dialogue.
    Page 4, “Evaluation and Discussion”

See all papers in Proc. ACL 2014 that mention coreference.

See all papers in Proc. ACL that mention coreference.

Back to top.

coreference resolution

Appears in 9 sentences as: coreference resolution (9)
In Probabilistic Labeling for Efficient Referential Grounding based on Collaborative Discourse
  1. Our system first processes the data using automatic semantic parsing and coreference resolution .
    Page 2, “Probabilistic Labeling for Reference Grounding”
  2. We then perform pairwise coreference resolution on the discourse entities to find out the discourse relations between entities from different utterances.
    Page 2, “Probabilistic Labeling for Reference Grounding”
  3. Based on the semantic parsing and pairwise coreference resolution results, our system further builds a graph representation to capture the collaborative discourse and formulate referential grounding as a probabilistic labeling problem, as described next.
    Page 3, “Probabilistic Labeling for Reference Grounding”
  4. It is constructed based on the semantic parsing and coreference resolution results.
    Page 3, “Probabilistic Labeling for Reference Grounding”
  5. The edge attributes can be either from parsing results, such as a spatial relation between two entities (e.g., RightO f (a1, a2)); Or from pairwise coreference resolution results, i.e., two entities are coreferen-tial (coref = +) or not (coref = —).
    Page 3, “Probabilistic Labeling for Reference Grounding”
  6. With no surprise, the coreference resolution performance plays an important role in the final grounding performance (see the grounding performance of using manually annotated coreference in the bottom part of Table 1).
    Page 5, “Evaluation and Discussion”
  7. Due to the simplicity of our current coreference classifier and the flexibility of the human-human dialogue in the data, the pairwise coreference resolution only achieves 0.74 in precision and 0.43 in recall.
    Page 5, “Evaluation and Discussion”
  8. The low recall of coreference resolution makes it difficult to link interrelated referring expressions and resolve them jointly.
    Page 5, “Evaluation and Discussion”
  9. So it is important to develop more sophisticated coreference resolution and dialogue management components to reliably track the discourse relations and other dynamics in the dialogue to facilitate referential grounding.
    Page 5, “Evaluation and Discussion”

See all papers in Proc. ACL 2014 that mention coreference resolution.

See all papers in Proc. ACL that mention coreference resolution.

Back to top.

semantic parsing

Appears in 7 sentences as: semantic parser (1) semantic parsing (6)
In Probabilistic Labeling for Efficient Referential Grounding based on Collaborative Discourse
  1. These works have provided valuable insights on how to manually and/or automatically build key components (e.g., semantic parsing , grounding functions between visual features and words, mapping procedures) for a situated referential grounding system.
    Page 2, “Related Work”
  2. Our system first processes the data using automatic semantic parsing and coreference resolution.
    Page 2, “Probabilistic Labeling for Reference Grounding”
  3. For semantic parsing , we use a rule-based CCG parser (Bozsahin et al., 2005) to parse each utterance into a formal semantic representation.
    Page 2, “Probabilistic Labeling for Reference Grounding”
  4. Based on the semantic parsing and pairwise coreference resolution results, our system further builds a graph representation to capture the collaborative discourse and formulate referential grounding as a probabilistic labeling problem, as described next.
    Page 3, “Probabilistic Labeling for Reference Grounding”
  5. It is constructed based on the semantic parsing and coreference resolution results.
    Page 3, “Probabilistic Labeling for Reference Grounding”
  6. The node attributes are from the semantic parsing results, i.e., the unary properties associated to a discourse entity.
    Page 3, “Probabilistic Labeling for Reference Grounding”
  7. We first applied the semantic parser and coreference classifier as described in Section 4.1 to process each dialogue, and then built a graph representation based on the automatic processing results at the end of the dialogue.
    Page 4, “Evaluation and Discussion”

See all papers in Proc. ACL 2014 that mention semantic parsing.

See all papers in Proc. ACL that mention semantic parsing.

Back to top.

significantly outperforms

Appears in 4 sentences as: significantly outperforms (4)
In Probabilistic Labeling for Efficient Referential Grounding based on Collaborative Discourse
  1. Our empirical results have shown the probabilistic labeling approach significantly outperforms a previous graph-matching approach for referential grounding.
    Page 1, “Abstract”
  2. Our empirical results have shown that the probabilistic labeling approach significantly outperforms the state-space search approach in both grounding accuracy and efficiency.
    Page 1, “Introduction”
  3. significantly outperforms state-space search (S.S.S.
    Page 5, “Evaluation and Discussion”
  4. Although probabilistic labeling significantly outperforms the state-space search, the grounding performance is still rather poor (less than 50%)
    Page 5, “Evaluation and Discussion”

See all papers in Proc. ACL 2014 that mention significantly outperforms.

See all papers in Proc. ACL that mention significantly outperforms.

Back to top.

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