Turkish Coreference Resolution

Abstract

Representation of coreferential relations is a challenging and actively studied topic for prodrop and morphologically rich languages (PD-MRLs) due to dropped pronouns (e.g., null subjects and omitted possessive pronouns). These phenomena require a representation scheme at the morphology level and enhanced evaluation methods. In this paper, we propose a representation & evaluation scheme to incorporate dropped pronouns into coreference resolution and validate it on the Turkish language. Using the scheme, we extend the annotations on the only existing Turkish coreference dataset, which originally did not contain annotations for dropped pronouns. We provide publicly available pre and post processors to enhance the prominent CoNLL coreference scorer also to cover coreferential relations arising from dropped pronouns. As a final step, the paper reports the first neural Turkish coreference resolution results in the literature. Although validated on Turkish, the proposed scheme is languageindependent and may be used for other PD-MRLs.

Incorporating Dropped Pronouns into Coreference Resolution: The case for Turkish

“…The model may utilize either word embeddings (word2vec 11 and fastText 12 ) or contextual neural language models (ELMo 11 and BERT 13 ). In addition to dense representations, we also extended the replicated model by including hand-crafted features used in previous Turkish studies (Schüller et al, 2017;Pamay and Eryigit, 2018) to analyze their representation power for morphological richness. A mention is considered as a sequence of tokens so that its embedding is created from its words' embeddings.…”

“…Previous works on Turkish CR are based on traditional machine learning algorithms Kılıçaslan et al, 2009;Küçük and Yöndem, 2015;Schüller et al, 2017;Pamay and Eryigit, 2018). The most recent Turkish coreference dataset (MTCC -Marmara Turkish Coreference Corpus) is from Schüller et al (2017), and consists of a document subset extracted from METU Turkish Corpus (MTC) (Say et al, 2002).…”

“…The most recent Turkish coreference dataset (MTCC -Marmara Turkish Coreference Corpus) is from Schüller et al (2017), and consists of a document subset extracted from METU Turkish Corpus (MTC) (Say et al, 2002). The dataset had been later extended by morpho-syntactic features by Pamay and Eryigit (2018) using an automated Turkish NLP pipeline (Eryigit, 2014). This dataset does not contain annotations for dropped pronouns.…”

“…In this paper, we propose a representation & evaluation scheme using existing tokens to incorporate dropped pronouns into coreference resolution and validate it on Turkish. Using this scheme, we extend the annotations on the only existing Turkish coreference dataset (Schüller et al, 2017;Pamay and Eryigit, 2018), which originally did not contain annotations for dropped pronouns. We provide publicly available pre and post processors 4 to enhance the prominent CoNLL coreference scorer 5 (Pradhan et al, 2014) to also cover multiple coreferential relations arising from dropped pronouns.…”

“…The preliminary results are reported on a neural coreference resolution model with a mention-ranking approach (Klemen and Žitnik, 2021), which was introduced for Slovene, another PD-MRL. Since the coreference information is coded at the morphology level, we investigate the impact of different word embeddings (Mikolov et al, 2013a,b;Grave et al, 2018), neural language models (Peters et al, 2018;Devlin et al, 2018;Schweter, 2020), and the inclusion of hand-crafted features used in previous studies (Schüller et al, 2017;Pamay and Eryigit, 2018) to analyze their representation power for morphological richness. Although validated on Turkish, the proposed scheme is language-independent and may be used for other PD-MRLs.…”