ParaDetox: Detoxification with Parallel Data

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 610 - 627 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Controllable Dictionary Example Generation: Generating Example Sentences for Specific Targeted Audiences Xingwei He, Siu Ming Yiu Department of Computer Science, The University of Hong Kong, Hong Kong, China hexingwei15@gmail.com, smyiu@cs.hku.hk Abstract Example sentences for targeted words in a dic- tionary play an important role to help read- ers understand the usage of words. Tradition- ally, example sentences in a dictionary are usu- ally created by linguistics experts, which are labor-intensive and knowledge-intensive. In this paper, we introduce the problem of dic- tionary example sentence generation, aiming to automatically generate dictionary example sentences for targeted words according to the corresponding definitions. This task is chal- lenging especially for polysemous words, be- cause the generated sentences need to reflect different usages and meanings of these tar- geted words. Targeted readers may also have different backgrounds and educational levels. It is essential to generate example sentences that can be understandable for different back- grounds and levels of audiences. To solve these problems, we propose a controllable target-word-aware model for this task. Our proposed model can generate reasonable exam- ples for targeted words, even for polysemous words. In addition, our model allows users to provide explicit control over attributes re- lated to readability, such as length and lexical complexity, thus generating suitable examples for targeted audiences. Automatic and human evaluations on the Oxford dictionary dataset show that our model can generate suitable ex- amples for targeted words with specific defini- tions while meeting the desired readability. 1 Introduction A dictionary usually consists of targeted words, part-of-speech (POS) tags, definitions and corre- sponding example sentences. Definitions and their corresponding examples enable audiences to better master new words, understand unfamiliar texts and the usage of the words in typical sentences, where a definition is a simple description for the meaning of the targeted word, and an example shows audi- ences how to use the word under this definition. Both definitions and examples are critical, playing an important role in language acquisition and nat- ural language understanding. However, it is often the case that audiences cannot find satisfactory ex- ample sentences for rarely used or newly coined words. On the other hand, it is time-consuming for experts to create dictionary examples for these words. With the advancement of AI technologies, it is a natural direction to study how to generate dictionary examples automatically, to assist dictio- nary compilation and help humans understand the corresponding targeted words. Dictionary example sentence generation aims to generate example sentences for targeted words to reflect their definitions and usages automati- cally. Recently, definition generation (Noraset et al., 2017; Gadetsky et al., 2018; Ishiwatari et al., 2019) has been extensively studied, yet generating example sentences is not well-studied. To the best of our knowledge, we are the first group to intro- duce this challenging problem. One main challenge for this task is that targeted words …

Content truncated to 500 words