A Mention-Ranking Model for Abstract Anaphora Resolution

Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 221–232 Copenhagen, Denmark, September 7–11, 2017. c⃝2017 Association for Computational Linguistics A Mention-Ranking Model for Abstract Anaphora Resolution Ana Marasovi´c, Leo Born†, Juri Opitz†, and Anette Frank† Research Training Group AIPHES Department of Computational Linguistics Heidelberg University 69120 Heidelberg, Germany {marasovic,born,opitz,frank}@cl.uni-heidelberg.de Abstract Resolving abstract anaphora is an impor- tant, but difficult task for text understand- ing. Yet, with recent advances in represen- tation learning this task becomes a more tangible aim. A central property of ab- stract anaphora is that it establishes a re- lation between the anaphor embedded in the anaphoric sentence and its (typical- ly non-nominal) antecedent. We propose a mention-ranking model that learns how abstract anaphors relate to their antece- dents with an LSTM-Siamese Net. We overcome the lack of training data by generating artificial anaphoric sentence– antecedent pairs. Our model outperforms state-of-the-art results on shell noun re- solution. We also report first benchmark results on an abstract anaphora subset of the ARRAU corpus. This corpus presents a greater challenge due to a mixture of nominal and pronominal anaphors and a greater range of confounders. We found model variants that outperform the base- lines for nominal anaphors, without train- ing on individual anaphor data, but still lag behind for pronominal anaphors. Our model selects syntactically plausible can- didates and – if disregarding syntax – dis- criminates candidates using deeper fea- tures. 1 Introduction Current research in anaphora (or coreference) res- olution is focused on resolving noun phrases re- ferring to concrete objects or entities in the real †Leo Born, Juri Opitz and Anette Frank contributed equally to this work. world, which is arguably the most frequently oc- curring type. Distinct from these are diverse types of abstract anaphora (AA) (Asher, 1993) where reference is made to propositions, facts, events or properties. An example is given in (1) below.1 While recent approaches address the resolution of selected abstract shell nouns (Kolhatkar and Hirst, 2014), we aim to resolve a wide range of abstract anaphors, such as the NP this trend in (1), as well as pronominal anaphors (this, that, or it). Henceforth, we refer to a sentence that contains an abstract anaphor as the anaphoric sentence (AnaphS), and to a constituent that the anaphor refers to as the antecedent (Antec) (cf. (1)). (1) Ever-more powerful desktop computers, designed with one or more microprocessors as their ”brains”, are ex- pected to increasingly take on functions carried out by more expensive minicomputers and mainframes. ”[Antec The guys that make traditional hardware are really being obsoleted by microprocessor-based ma- chines]”, said Mr. Benton. [AnaphS As a result of this trendAA, longtime powerhouses HP, IBM and Digital Equipment Corp. are scrambling to counterattack with microprocessor-based systems of their own.] A major obstacle for solving this task is the lack of sufficient amounts of annotated training data. We propose a method to generate large amounts of training instances covering a wide range of ab- stract anaphor types. This enables us …

Content truncated to 500 words