The Best of Both Worlds: Combining Human and Machine Translations for Multilingual Semantic Parsing with Active Learning

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 912–928 July 9-14, 2023 ©2023 Association for Computational Linguistics From Key Points to Key Point Hierarchy: Structured and Expressive Opinion Summarization Arie Cattan1∗ Lilach Eden2∗ Yoav Kantor2 Roy Bar-Haim2 1Computer Science Department, Bar Ilan University 2IBM Research arie.cattan@gmail.com {lilache, yoavka, roybar}@il.ibm.com Abstract Key Point Analysis (KPA) has been recently pro- posed for deriving fine-grained insights from collections of textual comments. KPA extracts the main points in the data as a list of concise sentences or phrases, termed key points, and quantifies their prevalence. While key points are more expressive than word clouds and key phrases, making sense of a long, flat list of key points, which often express related ideas in varying levels of granularity, may still be chal- lenging. To address this limitation of KPA, we introduce the task of organizing a given set of key points into a hierarchy, according to their specificity. Such hierarchies may be viewed as a novel type of Textual Entailment Graph. We develop THINKP, a high quality benchmark dataset of key point hierarchies for business and product reviews, obtained by consolidat- ing multiple annotations. We compare different methods for predicting pairwise relations be- tween key points, and for inferring a hierarchy from these pairwise predictions. In particular, for the task of computing pairwise key point relations, we achieve significant gains over ex- isting strong baselines by applying directional distributional similarity methods to a novel dis- tributional representation of key points, and further boost performance via weak supervi- sion. https://github.com/IBM/kpa-hierarchy 1 Introduction Many organizations face the challenge of extract- ing insights from large collections of textual com- ments, such as user reviews, survey responses, and feedback from customers or employees. Current text analytics tools summarize such datasets via word clouds (Heimerl et al., 2014) or key phrases (Hasan and Ng, 2014; Merrouni et al., 2019), which are often too crude to capture fine-grained insights. ∗ Equal contribution. Work done while the first author was an intern at IBM Research. Multi-document summarization methods, on the other hand (Chu and Liu, 2019; Bražinskas et al., 2020a,b; Angelidis et al., 2021; Louis and Maynez, 2022), do not quantify the prevalence of each point in the summary, and are not well-suited for repre- senting conflicting views (Bar-Haim et al., 2021). Key Point Analysis (KPA) is a recent opinion summarization framework that aims to address the above limitations (Bar-Haim et al., 2020b). KPA extracts concise sentences and phrases termed Key Points (KPs), which represent the most salient points in the data, and quantifies the prevalence of each KP as the number of its matching input sen- tences. One remaining shortcoming of KPA, how- ever, is that it generates a flat list, which does not capture the relations between the key points. For example, consider the sample set of key points in Figure 1 (left), which was automatically extracted from reviews of one of the hotels in the Yelp Open Dataset1. The results …

Content truncated to 500 words