Semi-supervised Formality Style Transfer using Language Model Discriminator and Mutual Information Maximization

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 2340–2354 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 2340 Semi-supervised Formality Style Transfer using Language Model Discriminator and Mutual Information Maximization Kunal Chawla Georgia Institute of Technology Atlanta, GA kunalchawla@gatech.edu Diyi Yang Georgia Institute of Technology Atlanta, GA dyang888@gatech.edu Abstract Formality style transfer is the task of con- verting informal sentences to grammatically- correct formal sentences, which can be used to improve performance of many downstream NLP tasks. In this work, we propose a semi- supervised formality style transfer model that utilizes a language model-based discriminator to maximize the likelihood of the output sen- tence being formal, which allows us to use maximization of token-level conditional prob- abilities for training. We further propose to maximize mutual information between source and target styles as our training objective in- stead of maximizing the regular likelihood that often leads to repetitive and trivial gen- erated responses. Experiments showed that our model outperformed previous state-of-the- art baselines significantly in terms of both au- tomated metrics and human judgement. We further generalized our model to unsupervised text style transfer task, and achieved signifi- cant improvements on two benchmark senti- ment style transfer datasets. 1 Introduction Text style transfer is the task of changing the style of a sentence while preserving the content. It has many useful applications, such as changing emo- tion of a sentence, removing biases in natural lan- guage, and increasing politeness in text (Sennrich et al., 2016; Pryzant et al.; Rabinovich et al., 2017; Yang et al., 2019; Chen et al., 2018). There is a wide availability of “informal” data from online sources, yet current Natural Language Processing (NLP) tasks and models could not lever- age or achieve good performance for such data due to informal expressions, and grammatical, spelling and semantic errors. Hence, formality style transfer, a specific style transfer task that aims to preserve the content of an informal sentence while mak- ing it semantically and grammatically correct, has Informal I flippin’ LOVE that movie, sweeeet! Formal I truly enjoy that movie. Informal we was hanging out a little. Formal We were spending a small amount of time together. Table 1: Examples of (formal, informal) sentence pairs. recently received a growing amount of attention. Some examples are given in Table 1. The most widely-used models for formality style transfer are based on a variational auto-encoder ar- chitecture, trained on parallel text data of (informal, formal) style sentence pairs with same content (Jing et al., 2019). However, there is still a lot of incon- sistencies between human-generated sentences and outputs of current models, largely due to the lim- ited availability of parallel data. In contrast, large amount of data consisting of sentences with just ei- ther informal or formal labels is relatively easier to collect. To tackle the training data bottleneck, we propose a semi-supervised approach for formality style transfer, using both human-annotated parallel data and large amount of unlabeled data. Following the success of Generative Adversar- ial …

Content truncated to 500 words