Japanese Parsing Using Smaller BERT
نویسندگان
چکیده
BERT は fine-tuning することで様々な NLP タスクに対して高性能な結果を出した事前学習済みモデルであるが,多くのパラメータを調整する必要があるため学習や推論に時間がかかるという問題がある.本論文では日本語構文解析に対して,BERT の一部の層を削除した簡易小型化 の利用を提案する.実験では,京都大学ウェブ文書リードコーパスと京都大学テキストコーパスを混合したデータを用いて,京大版の とそこから構築した簡易小型化 の正解率と処理時間を比較した.提案する簡易小型化 では,京大版の からの正解率の劣化をウェブコーパスで 0.87 ポイント,テキストコーパスで 0.91 ポイントに押さえながら,学習時間は 83%,推論時間はウェブコーパスで 65%,テキストコーパスで 85%まで削減することができた.
منابع مشابه
Using Smaller Constituents Rather Than Sentences in Active Learning for Japanese Dependency Parsing
We investigate active learning methods for Japanese dependency parsing. We propose active learning methods of using partial dependency relations in a given sentence for parsing and evaluate their effectiveness empirically. Furthermore, we utilize syntactic constraints of Japanese to obtain more labeled examples from precious labeled ones that annotators give. Experimental results show that our ...
متن کاملJapanese Dependency Parsing Using a Tournament Model
In Japanese dependency parsing, Kudo’s relative preference-based method (Kudo and Matsumoto, 2005) outperforms both deterministic and probabilistic CKY-based parsing methods. In Kudo’s method, for each dependent word (or chunk) a loglinear model estimates relative preference of all other candidate words (or chunks) for being as its head. This cannot be considered in the deterministic parsing me...
متن کاملJapanese Dependency Parsing using Relative Preference of Dependency
This paper presents a new statistical Japanese dependency parser which models a relative preference of dependency. Most conventional Japanese dependency parsers are based on binary classifications where all possible pairs of segments are classified into positive (dependent) or negative (non-dependent) examples. However, such methods are not suitable for dependency parsing, since the goal of thi...
متن کاملParsing Myanmar (Burmese) by Using Japanese as a Pivot
As Myanmar (Burmese) and Japanese share similar syntactic structures, we explore an approach for parsing Myanmar by using Japanese as a pivot. Specifically, we first apply a statistical machines translation (SMT) system to translate Myanmar sentences into Japanese, and then we parse the Japanese translations. Finally, the Japanese syntactic structures are mapped to the original Myanmar sentence...
متن کاملJapanese Dependency Parsing Using Sequential Labeling for Semi-spoken Language
The amount of documents directly published by end users is increasing along with the growth of Web 2.0. Such documents often contain spoken-style expressions, which are difficult to analyze using conventional parsers. This paper presents dependency parsing whose goal is to analyze Japanese semi-spoken expressions. One characteristic of our method is that it can parse selfdependent (independent)...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: Shizen gengo shori
سال: 2022
ISSN: ['1340-7619', '2185-8314']
DOI: https://doi.org/10.5715/jnlp.29.854