Neural Character-level Dependency Parsing for Chinese
Authors: Haonan Li, Zhisong Zhang, Yuqi Ju, Hai Zhao
AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments Data We use Chinese Penn Treebank 5.1 (CTB5) for evaluation. Dataset splitting follows (Zhang and Clark 2008). ... Evaluation Although at present, there is no general standards to evaluate full character-level dependency parsing, we can still follow the practice of word-level parsing and take UAS/LAS (unlabeled/labeled attachment scores) on character-level tokens as the metrics. ... Table 1: Results (on Dev set) of different tagging strategies ... Table 2: Character-level evaluation. ... Table 3: Pipelined analysis. ... Table 4: Word-level evaluation (with restoration), Dep. indicates UAS for dependency parsing. |
| Researcher Affiliation | Academia | Haonan Li,1,2 Zhisong Zhang,1,2 Yuqi Ju,1,2 Hai Zhao1,2, 1Department of Computer Science and Engineering, Shanghai Jiao Tong University 2Key Laboratory of Shanghai Education Commission for Intelligent Interaction and Cognitive Engineering Shanghai Jiao Tong University, Shanghai, 200240, China nathan l@163.com, {zzs2011, tongkong}@sjtu.edu.cn, zhaohai@cs.sjtu.edu.cn |
| Pseudocode | No | The paper describes the models and methods in text and with diagrams (Figure 2, Figure 3, Figure 4, Figure 6), but does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper mentions 'The treebank is available at http://bcmi.sjtu.edu.cn/ zebraform/scdt.html' which refers to the dataset (SCDT), not the open-source code for the neural parser or the methodology described in the paper. |
| Open Datasets | Yes | To train the character-level parsing model, we bundle two treebanks to construct the required full character-level dependency treebank. ... We use Chinese Penn Treebank 5.1 (CTB5) for evaluation. ... SCDT (Shanghai Jiao Tong University) Chinese Character Dependency Treebank ... The treebank is available at http://bcmi.sjtu.edu.cn/ zebraform/scdt.html |
| Dataset Splits | Yes | Dataset splitting follows (Zhang and Clark 2008). ... Table 1: Results (on Dev set) of different tagging strategies (# stands for the number of POS tags or dependency labels). |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory amounts, or detailed computer specifications) used for running its experiments. |
| Software Dependencies | No | Malt Parser (Nivre, Hall, and Nilsson 2006) with default settings for the traditional model is exploited, while for the neural one, the LSTM parser in (Dyer et al. 2015) are utilized. ... character POS tags are learned and annotated with a Conditional Random Fields (CRFs) tagger using the same features and settings of (Chen, Zhang, and Sun 2008)... |
| Experiment Setup | Yes | The hyper-parameters are set as follows: 50 for character embedding8, 10 for POS embedding, 20 for action embedding and 3 for LSTM hidden layers. |