alt Inc. is to present its latest research results on Fine-Grained Named Entity Recognition (FG-NER) at the 27th International Conference on Computational Linguistics (COLING 2018), held from August 20th through 26th in Santa Fe, New Mexico, USA.
This research proposes a neural network-based method for Japanese FG-NER, which is an extension of the work presented in ACL 2017. Experimental results show that the proposed method significantly improves Japanese FG-NER performance (F-score) from 66.75% to 75.18%.
Furthermore, with the title “An Empirical Study on Fine-Grained Named Entity Recognition”, the paper presents the empirical comparison between various FG-NER algorithms for both Japanese and English. It also investigates the characteristics of each algorithm as well as the relationship between the amount of the training data with the performance of the algorithms.
This research not only contributes the method and knowledge on FG-NER to the Natural Language Processing research community but also has a significant importance in alt Inc.’s research and development of its natural language Dialogue Engine and Personal Artificial Intelligence (P.A.I.), a new paradigm of AI which tries to create a clone of the user. This achievement also encourages alt Inc. to strengthen the development of its FG-NER system as well as the research and development of Natural Language Processing and Dialogue Engine technologies.
About COLING （International Conference on Computational Linguistics）
COLING is the one of the oldest and most prestigious Computational Linguistics conference in the World, along with ACL and EMNLP. It is organized by the ICCL (International Committee on Computational Linguistics). Number of papers submitted to COLING2018 is 1017, out of which 37% have been accepted.