Definition Modelling for Appropriate Specificity

Abstract

Definition generation techniques aim to generate a definition of a target word or phrase given a context. In previous studies, researchers have faced various issues such as the out-of-vocabulary problem and over/underspecificity problems. Over-specific definitions present narrow word meanings, whereas under-specific definitions present general and context-insensitive meanings. Herein, we propose a method for definition generation with appropriate specificity. The proposed method addresses the aforementioned problems by leveraging a pre-trained encoderdecoder model, namely Text-to-Text Transfer Transformer, and introducing a re-ranking mechanism to model specificity in definitions. 1 Experimental results on standard evaluation datasets indicate that our method significantly outperforms the previous state-of-theart method. Moreover, manual evaluation confirms that our method effectively addresses the over/under-specificity problems.

LitMind Dictionary: An Open-Source Online Dictionary

“…NIST focuses on content words by giving more weightage to them. This makes NIST more informative than solely assigning an equal weight to each n-grams as BLEU (Huang et al, 2021).…”

“…Bevilacqua et al (2020) employed a novel span-based encoding scheme to fine-tune a pre-trained English Encoder-Decoder system to generate definitions. Huang et al (2021) leveraged the T5 model (Raffel et al, 2019) for this task and introduced a re-ranking mechanism to model specificity in definitions.…”

COMPILING: A Benchmark Dataset for Chinese Complexity Controllable Definition Generation

“…Due to the complex training objectives and large hyperparameter groups, large-scale pre-training models can effectively extract features from a large amount of supervised and unsupervised data. By storing the learned knowledge in parameters and fine-tuning the model for specific tasks, the same model can be applied to a series of downstream natural language processing tasks (Han et al, 2021a).…”

“…Prompt learning is a method of fully learning knowledge by adding additional text to the model's input. Prompt can be divided into artificial and automatic construction according to the text attached to the input (Han et al, 2021a). Among them, automatically constructed prompts are divided into discrete and continuous ones.…”

BLCU-ICALL at SemEval-2022 Task 1: Cross-Attention Multitasking Framework for Definition Modeling

“…Many previous work used additional data to improve the performance of generation, such as example sentences (Gadetsky et al, 2018;Chang et al, 2018;Ishiwatari et al, 2019;Kong et al, 2020) and semantic features (Yang et al, 2020). Some studies also investigated how to employ PLMs for this task (Reid et al, 2020;Bevilacqua et al, 2020;Huang et al, 2021;Kong et al, 2022).…”