Quantum-Inspired Representation for Long-Tail Senses of Word Sense Disambiguation

Authors: Junwei Zhang, Ruifang He, Fengyu Guo

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We theoretically prove the correctness of the method, and verify its effectiveness under the standard WSD evaluation framework and obtain state-of-the-art performance. Furthermore, we also test on the constructed LTS and the latest cross-lingual datasets, and achieve promising results.
Researcher Affiliation Academia 1 Tianjin Key Laboratory of Cognitive Computing and Application, College of Intelligence and Computing, Tianjin University, Tianjin, China. 2 College of Computer and Information Engineering, Tianjin Normal University, Tianjin, China.
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No Other hyperparameters not listed will be given in the published code.
Open Datasets Yes To evaluate the effectiveness of QR-WSD, we carried out experiments under two evaluation settings, namely the standardized evaluation setting and the enhanced evaluation setting. The standardized setting includes only Sem Cor2 in the training set; the enhanced setting includes Sem Cor and WNGT3 in the training set. (2http://lcl.uniroma1.it/wsdeval/training-data, 3https://wordnetcode.princeton.edu/glosstag.shtml)
Dataset Splits Yes Sem Eval07 (SE7; Pradhan et al. (2007)), following convention (Kumar et al. 2019; Blevins and Zettlemoyer 2020), is regarded as the development set.
Hardware Specification Yes The computing platform of the program is Ubuntu 18.04, which is equipped with two Tesla P40 GPUs.
Software Dependencies Yes The program is developed based on the Pytorch 1.8 framework and written in Python 3.6. Moreover, Word Net 3.0 is provided by NLTK 3.5, and bert-base-uncased and bert-large-uncased are provided by Transformers 4.5.
Experiment Setup Yes The learning rate, epoch and batch size of the model are {1e-5, 5e-6}, 20 and 4 respectively. Other hyperparameters not listed will be given in the published code.