Expanding, Retrieving and Infilling: Diversifying Cross-Domain Question Generation with Flexible Templates
42492Authors
nan
Abstract
AbstractSequence-to-sequence based models have recently shown promising results in generating high-quality questions. However, these models are also known to have main drawbacks such as lack of diversity and bad sentence structures. In this paper, we focus on question generation over SQL database and propose a novel framework by expanding, retrieving, and infilling that first incorporates flexible templates with a neural-based model to generate diverse expressions of questions with sentence structure guidance. Furthermore, a new activation/deactivation mechanism is proposed for template-based sequence-to-sequence generation, which learns to discriminate template patterns and content patterns, thus further improves generation quality. We conduct experiments on two large-scale cross-domain datasets. The experiments show that the superiority of our question generation method in producing more diverse questions while maintaining high quality and consistency under both automatic evaluation and human evaluation.
Keywords
Full Content
Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics, pages 358–368 April 19 - 23, 2021. ©2021 Association for Computational Linguistics 358 Implicitly Abusive Comparisons – A New Dataset and Linguistic Analysis Michael Wiegand Digital Age Research Center (D!ARC) Alpen-Adria-Universit¨at Klagenfurt AT-9020 Klagenfurt, Austria michael.wiegand@aau.at Maja Geulig Institute of Computational Linguistics Heidelberg University D-69120 Heidelberg, Germany geulig@cl.uni-heidelberg.de Josef Ruppenhofer Leibniz Institute for German Language D-68161 Mannheim, Germany ruppenhofer@ids-mannheim.de Abstract We examine the task of detecting implicitly abusive comparisons (e.g. Your hair looks like you have been electrocuted). Implicitly abu- sive comparisons are abusive comparisons in which abusive words (e.g. dumbass or scum) are absent. We detail the process of creating a novel dataset for this task via crowdsourcing that includes several measures to obtain a suffi- ciently representative and unbiased set of com- parisons. We also present classification exper- iments that include a range of linguistic fea- tures that help us better understand the mecha- nisms underlying abusive comparisons. 1 Introduction Abusive or offensive language is commonly defined as hurtful, derogatory or obscene utterances made by one person to another person.1 Examples are (1)-(3). (1) stop editing this, you dumbass. (2) Just want to slap the stupid out of these bimbos!!! (3) Go lick a pig you arab muslim piece of scum. In the literature, closely related terms include hate speech (Waseem and Hovy, 2016) or cyber bullying (Zhong et al., 2016). While there may be nuanced differences in meaning, they are all compatible with the general definition above. The definition we follow in this work also re- stricts abusive language to those utterances that are made to deliberately insult the target. A second re- quirement of an utterance to be considered abusive is that the target itself has to perceive the utterance as abusive. Due to the rise of user-generated web content, the amount of abusive language is steadily growing. NLP methods are required to focus human review efforts towards the most relevant microposts. 1http://thelawdictionary.org/ Though there has been much work on abusive language detection in general, there is has been comparatively little work focusing on implicit forms of abusive language (4)-(5) (Waseem et al., 2017). (4) I haven’t had an intelligent conversation with a woman in my whole life. (5) Why aren’t there any Mexicans on Star Trek? Because they don’t work in the future either. By implicit we understand abusive language that is not conveyed by (unambiguously) abusive words (e.g. dumbass, bimbo, scum). Detailed analysis on the output of existing classifiers has also revealed that currently only explicit abuse can be reliably detected (Wiegand et al., 2019). Given that implicit abuse is a challenging prob- lem, we believe that the only reasonable approach to solve this problem is to address specific sub- types individually rather than consider all types of implicit abuse at once. In this paper, we examine implicitly abusive comparisons. A comparison is the act of evaluating two or more things by deter- mining the relevant characteristics of each …
Content truncated to 500 words
Labeled Text
AbstractSequence-to-sequence based models have recently shown promising results in generating high-quality questions. However, these models are also known to have main drawbacks such as lack of diversity and bad sentence structures. In this paper, we focus on question generation over SQL database and propose a novel framework by expanding, retrieving, and infilling that first incorporates flexible templates with a neural-based model to generate diverse expressions of questions with sentence structure guidance. Furthermore, a new activation/deactivation mechanism is proposed for template-based sequence-to-sequence generation, which learns to discriminate template patterns and content patterns, thus further improves generation quality. We conduct experiments on two large-scale cross-domain datasets. The experiments show that the superiority of our question generation method in producing more diverse questions while maintaining high quality and consistency under both automatic evaluation and human evaluation.
Publication Details
- Paper ID:
- 42492
- Year:
- 2021
- Publication:
- TIPSTER
- Publication Type:
- conf
- DOI:
- nan