2021/01/25 by Yotam Shichel, Shichel, Yotam, Meir Kalech +3
Computer Science · #Computation and Language (cs.CL) #FOS: Computer and information sciences #Multimodal Machine Learning Applications #Natural Language Processing Techniques #Topic Modeling
paper · pdf · doi:10.48550/arxiv.2101.10096
openalex publication_date 2021/01/25 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Sentence Compression is the task of generating a shorter, yet grammatical\nversion of a given sentence, preserving the essence of the original sentence.\nThis paper proposes a Black-Box Optimizer for Compression (B-BOC): given a\nblack-box compression algorithm and assuming not all sentences need be\ncompressed -- find the best candidates for compression in order to maximize\nboth compression rate and quality. Given a required compression ratio, we\nconsider two scenarios: (i) single-sentence compression, and (ii)\nsentences-sequence compression. In the first scenario, our optimizer is trained\nto predict how well each sentence could be compressed while meeting the\nspecified ratio requirement. In the latter, the desired compression ratio is\napplied to a sequence of sentences (e.g., a paragraph) as a whole, rather than\non each individual sentence. To achieve that, we use B-BOC to assign an optimal\ncompression ratio to each sentence, then cast it as a Knapsack problem, which\nwe solve using bounded dynamic programming. We evaluate B-BOC on both scenarios\non three datasets, demonstrating that our optimizer improves both accuracy and\nRouge-F1-score compared to direct application of other compression algorithms.\n