Neural coreference resolution for Slovene language

Abstract

Coreference resolution systems aim to recognize and cluster together mentions of the same underlying entity. While there exist large amounts of research on broadly spoken languages such as English and Chinese, research on coreference in other languages is comparably scarce. In this work we first present SentiCoref 1.0 - a coreference resolution dataset for Slovene language that is comparable to English-based corpora. Further, we conduct a series of analyses using various complex models that range from simple linear models to current state-of-the-art deep neural coreference approaches leveraging pre-trained contextual embeddings. Apart from SentiCoref, we evaluate models also on a smaller coref149 Slovene dataset to justify the creation of a new corpus. We investigate robustness of the models using cross-domain data and data augmentations. Models using contextual embeddings achieve the best results - up to 0.92 average F1 score for the SentiCoref dataset. Cross-domain experiments indicate that SentiCoref allows the models to learn more general patterns, which enables them to outperform models, learned on coref149 only.

Incorporating Dropped Pronouns into Coreference Resolution: The case for Turkish

“…The neural Turkish CR performances are reported using a neural coreference resolution architecture (Klemen and Žitnik, 2021) which was introduced for Slovene, another PD-MRL. The model uses a mention-ranking approach and resolves coreferential relations on gold-standard mentions.…”

“…As a result, the produced mention embedding comprises information about the head token in the mention and its right and left contexts. This paper replicates the neural CR model with the default hyper-parameters from Klemen and Žitnik (2021). Documents are split into train/validation/test parts by considering their genres.…”

“…Guarasci et al (2021) used ELECTRA (Clark et al, 2020) on the neural structure for Italian. Klemen and Žitnik (2021) proposed a neural CR model focusing on only mention clustering stage for Slovene.…”

“…The pro-drop nature of such languages reveals the need for mention annotation on other tokens (i.e., artificially inserted (Pradhan et al, 2012a;Nedoluzhko et al, 2022) or existing tokens (Rodrıguez et al, 2010;Klemen and Žitnik, 2021) other than the dropped pronouns, such as verbs carrying personal suffixes). The morphological richness in these languages may reveal the appearance of multiple coreference relations on a single token which is illustrated below.…”

“…As a final step, the paper reports the first neural Turkish coreference resolution results in the literature providing a strong baseline for future studies in this field. 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.…”