With Measured Words: Simple Sentence Selection for Black-Box Optimization of Sentence Compression Algorithms

Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics, pages 1625–1634 April 19 - 23, 2021. ©2021 Association for Computational Linguistics 1625 With Measured Words: Simple Sentence Selection for Black-Box Optimization of Sentence Compression Algorithms Yotam Shichel Department of Software and Information Systems Engineering Ben Gurion University yotamsh@post.bgu.ac.il Meir Kalech Department of Software and Information Systems Engineering Ben Gurion University kalech@bgu.ac.il Oren Tsur Department of Software and Information Systems Engineering Ben Gurion University orentsur@bgu.ac.il Abstract Sentence Compression is the task of gen- erating a shorter, yet grammatical version of a given sentence, preserving the essence of the original sentence. This paper pro- poses a Black-Box Optimizer for Compression (B-BOC): given a black-box compression al- gorithm and assuming not all sentences need be compressed – find the best candidates for compression in order to maximize both com- pression rate and quality. Given a required compression ratio, we consider two scenar- ios: (i) single-sentence compression, and (ii) sentences-sequence compression. In the first scenario, our optimizer is trained to predict how well each sentence could be compressed while meeting the specified ratio requirement. In the latter, the desired compression ratio is applied to a sequence of sentences (e.g., a paragraph) as a whole, rather than on each in- dividual sentence. To achieve that, we use B-BOC to assign an optimal compression ra- tio to each sentence, then cast it as a Knapsack problem, which we solve using bounded dy- namic programming. We evaluate B-BOC on both scenarios on three datasets, demonstrat- ing that our optimizer improves both accuracy and Rouge-F1-score compared to direct appli- cation of other compression algorithms. 1 Introduction Sentence Compression is the task of generating a short, accurate, and fluent sentence that preserves the essence of a given original sentence by re- moving nonessential words and/or rephrasing it in a compact form. Compression can take many forms, ranging from Extractive and Abstractive Summarization (Jing, 2000; Madnani et al., 2007; Cohn and Lapata, 2008, 2009; Galanis and An- droutsopoulos, 2010; Rush et al., 2015; Chopra et al., 2016) to Text Simplification and Paraphras- ing (Bannard and Callison-Burch, 2005; Xu et al., 2012; Klerke et al., 2016; Narayan et al., 2017; Aharoni and Goldberg, 2018; Botha et al., 2018), among others. On the sentential level, compression is often viewed as a word deletion task (Knight and Marcu, 2000, 2002; Filippova and Strube, 2008; Filippova et al., 2015; Wang et al., 2016, 2017; Zhou and Rush, 2019). However, not all sentences could, or should be compressed as part of compressing a longer text they reside in. Consider the familiar scenario in which a full paragraph needs to be com- pressed in order to have an EACL paper meet the page restriction specified in the submission guide- lines. A common approach by LATEX users is to first identify paragraphs ending with a short line, (e.g., this very paragraph), then choose one or more sentences that could be compressed with a mini- mal loss of information – shaving the …

Content truncated to 500 words