Dialogue focus tracking for zero pronoun resolution

23591

Authors

nan

Abstract

We take a novel approach to zero pronoun res- olution in Chinese: our model explicitly tracks the flow of focus in a discourse. Our approach, which generalizes to deictic references, is not reliant on the presence of overt noun phrase antecedents to resolve to, and allows us to ad- dress the large percentage of “non-anaphoric” pronouns filtered out in other approaches. We furthermore train our model using read- ily available parallel Chinese/English corpora, allowing for training without hand-annotated data. Our results demonstrate improvements on two test sets, as well as the usefulness of linguistically motivated features. 1

Keywords

nan

Full Content

Human Language Technologies: The 2015 Annual Conference of the North American Chapter of the ACL, pages 494–503, Denver, Colorado, May 31 – June 5, 2015. c⃝2015 Association for Computational Linguistics Dialogue focus tracking for zero pronoun resolution Sudha Rao1,3, Allyson Ettinger2, Hal Daum´e III1,3, Philip Resnik2,3 1Computer Science, 2Linguistics, 3UMIACS University of Maryland raosudha@cs.umd.edu, aetting@umd.edu, hal@cs.umd.edu, resnik@umd.edu Abstract We take a novel approach to zero pronoun res- olution in Chinese: our model explicitly tracks the flow of focus in a discourse. Our approach, which generalizes to deictic references, is not reliant on the presence of overt noun phrase antecedents to resolve to, and allows us to ad- dress the large percentage of “non-anaphoric” pronouns filtered out in other approaches. We furthermore train our model using read- ily available parallel Chinese/English corpora, allowing for training without hand-annotated data. Our results demonstrate improvements on two test sets, as well as the usefulness of linguistically motivated features. 1 Introduction “Pro-drop” languages like Chinese, Japanese and Turkish allow for dropping of pronouns when the referents of those pronouns can be inferred. English is typically not pro-drop, but is unusual in that re- gard: two thirds of languages documented in WALS (Haspelmath et al., 2005) can be categorized as pro- drop. In such languages, sentences are frequently characterized by “zero pronouns”: gaps in the sen- tence which in English would hold an overt pro- noun. In some languages, verbal morphology or cl- itics elsewhere in the sentence are sufficient to re- solve the ambiguity of dropped pronouns; in other languages, there is no overt marking at all in the sen- tence and the referent of the dropped pronoun must be resolved using pragmatic information. Our work departs from mainstream work on zero pronoun resolution in that we focus primarily on the resolution of deictic zero pronouns. Unlike an Figure 1: A conversation between a student and a teacher. The text has been translated from Mandarin, but zero pronouns are retained and indexed with their referent: (T)eacher or (S)tudent. anaphoric zero pronoun (Section 2), whose refer- ence must be specified by a noun phrase occurring previously in the text, a non-anaphoric zero pro- noun refers to an entity that is salient from larger units of discourse (such as full sentences or pas- sages) or from the extralinguistic environment (out- side of the text altogether). Although anaphoric zero pronoun resolution has been the focus of most past work (Yeh and Chen, 2007; Chen and Ng, 2014), 50% or fewer of zero pronouns in natural Chinese text are anaphoric (Zhao and Ng, 2007; Kong and Zhou, 2010). Our approach allows for generaliza- tion to non-anaphoric pronouns, focusing in partic- ular on deictic non-anaphoric zero pronouns, which 494 refer to salient entities in the environment (such as the speaker, hearer or pragmatically accessible ref- erent) without requiring any introduction in the pre- ceding text. Figure 1 shows an example conversa- tion in which zero pronouns are frequently used to refer to speaker or listener, and would be translated to …

Content truncated to 500 words

Labeled Text

We take a novel approach to zero pronoun resolution in Chinese: our model explicitly tracks the flow of focus in a discourse. Our approach, which generalizes to deictic references, is not reliant on the presence of overt noun phrase antecedents to resolve to, and allows us to address the large percentage of “non-anaphoric” pronouns filtered out in other approaches. We furthermore train our model using readily available parallel Chinese/English corpora, allowing for training without hand-annotated data. Our results demonstrate improvements on two test sets, as well as the usefulness of linguistically motivated features. 1

Publication Details

Paper ID:
23591
Year:
2015
Publication:
TIPSTER
Publication Type:
conf
DOI:
nan