Controlling the Focus of Pretrained Language Generation Models

Findings of the Association for Computational Linguistics: ACL 2022, pages 3291 - 3306 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Controlling the Focus of Pretrained Language Generation Models Jiabao Ji Shanghai Jiao Tong University jiyi0115@gmail.com Yoon Kim Massachusetts Institute of Technology yoonkim@mit.edu James Glass Massachusetts Institute of Technology glass@mit.edu Tianxing He Massachusetts Institute of Technology tianxing@mit.edu Abstract The finetuning of pretrained transformer-based language generation models are typically con- ducted in an end-to-end manner, where the model learns to attend to relevant parts of the input by itself. However, there does not exist a mechanism to directly control the model’s fo- cus. This work aims to develop a control mech- anism by which a user can select spans of con- text as “highlights” for the model to focus on, and generate relevant output. To achieve this goal, we augment a pretrained model with train- able “focus vectors” that are directly applied to the model’s embeddings, while the model itself is kept fixed. These vectors, trained on automatic annotations derived from attribution methods, act as indicators for context impor- tance. We test our approach on two core genera- tion tasks: dialogue response generation and ab- stractive summarization. We also collect evalu- ation data where the highlight-generation pairs are annotated by humans. Our experiments show that the trained focus vectors are effective in steering the model to generate outputs that are relevant to user-selected highlights. 1 Introduction Transformer-based models pretrained on large- scale text data have become the dominant paradigm for natural language generation (NLG) tasks (Roller et al., 2020; Lewis et al., 2019; Raffel et al., 2020). The attention module (Bahdanau et al., 2016; Vaswani et al., 2017), which aggregates infor- mation via a weighted average over word-level em- beddings, plays a vital role in these models. The at- tention mechanism serves two major purposes: (1) It captures linguistic phenomena in the input (Clark et al., 2019; Kovaleva et al., 2019; Kobayashi et al., 2020); (2) It helps the model focus on relevant portions of the input (e.g., alignment in machine translation (Bahdanau et al., 2016) and abstractive summarization (Rush et al., 2015)). The attention module is particularly useful as it does not require any explicit supervision: the Figure 1: Illustration of our motivation: different high- lights in the input (including persona) lead to different generations. This example is from our collected dia- logue data for evaluation (Section 3). model learns to attend to relevant parts of the input by itself through end-to-end training. However, this property makes it difficult to explicitly con- trol the model’s focus. If the model happens to put focus on some span of context that the user thinks is not so important, we currently do not have a mechanism to correct it. This is especially sub-optimal in some NLG applications involving a relatively long input such as dialogue or summa- rization: focusing on different spans of the input could result in completely different generations (il- lustrated in Figure 1). It would be attractive to give the …

Content truncated to 500 words