Analogical Reasoning on Chinese Morphological and Semantic Relations

By Shen Li, Zhe Zhao, Renfen Hu, Wensi Li, Tao Liu, Xiaoyong Du. Annual Meeting of the Association for Computational Linguistics, 2018

450 citations recorded by the source corpus.

Analogical reasoning is effective in capturing linguistic regularities. This paper proposes an analogical reasoning task on Chinese. After delving into Chinese lexical knowledge, we sketch 68 implicit morphological relations and 28 explicit semantic relations. A big and balanced dataset CA8 is then built for this task, including 17813 questions. Furthermore, we systematically explore the influences of vector representations, context features, and corpora on analogical reasoning. With the experiments, CA8 is proved to be a reliable benchmark for evaluating Chinese word embeddings.

Subfields: Artificial Intelligence

View Paper: https://www.semanticscholar.org/paper/1f8c70ce22fc5b34ee725d79d4a061b3062f6fc5

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Xipeng Qiu
2. Kyle Richardson
3. Anton van den Hengel
4. Xuanjing Huang
5. Chunhua Shen
6. 刘晗
7. Hang Yan
8. Hai Hu

Citations in the corpus

• FLAT: Chinese NER Using Flat-Lattice Transformer (2020, methodology): “(2) Another line is to convert lattice into graph and use a graph neural network (GNN) to encode it, such as Lexicon-based Graph Network (LGN) (Gui et al., 2019b) and Collaborative Graph Network (CGN) (Sui et al., 2019).”
• On the General Value of Evidence, and Bilingual Scene-Text Visual Question Answering (2020, methodology): “Technically, word embedding of question is first extracted by GloVe [26] for English questions and Word2Vec [19] for Chinese questions.”
• BiPaR: A Bilingual Parallel Dataset for Multilingual and Cross-lingual Reading Comprehension on Novels (2019, methodology): “The 300-dimensional Glove word embeddings trained from 840B Web crawl data (Pennington et al., 2014) were used as our pre-trained English word embeddings while the 300-dimensional SGNS word embeddings trained from mixed-sources data (Li et al., 2018) as our pre-trained Chinese word embeddings.”
• Zero-shot Text Classification via Reinforced Self-training (2020, methodology): “For baselines, we adopt 300-dim GloVe vectors (Pennington et al., 2014) for English words and 300-dim word vectors from (Li et al., 2018) for Chinese words.”
• Induction Networks for Few-Shot Text Classification (2019, methodology): “, 2014) for ARSC dataset and 300-dimension Chinese word embeddings trained by Li et al. (2018) for ODIC.”
• OCNLI: Original Chinese Natural Language Inference (2020, methodology): “In each case, we use Chinese character embeddings from Li et al. (2018) in place of the original GloVe embeddings.”
• Reconstructing Capsule Networks for Zero-shot Intent Classification (2019, methodology): “For SMP-2018, we use 300-dim Chinese word embeddings pre-trained by Li et al. (2018).”
• Censorship of Online Encyclopedias: Implications for NLP Models (2021, methodology): “[21] train 300-dimensional word embeddings on both Baidu Baike and Chinese language Wikipedia using the same algorithm,Word2Vec [24].”
• Mining Dual Emotion for Fake News Detection (2019, methodology): “In practice, as for word embeddings, we use GloVe [31] for English and Chinese Word Vectors for Chinese [25].”
• A Chapter-Wise Understanding System for Text-To-Speech in Chinese Novels (2021, methodology): “The inputs of PerNER are character embeddings [17] of a sentence, and the output is a sequence of BMEIO 1 tags.”