{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/evaluating-compositionality-in-sentence","title":"Evaluating Compositionality in Sentence Embeddings","arxiv_id":"1802.04302","date":"2018-02-12","proceeding":null,"authors":["Ishita Dasgupta","Demi Guo","Andreas Stuhlmüller","Samuel J. Gershman","Noah D. Goodman"],"abstract":"An important challenge for human-like AI is compositional semantics. Recent\nresearch has attempted to address this by using deep neural networks to learn\nvector space embeddings of sentences, which then serve as input to other tasks.\nWe present a new dataset for one such task, `natural language inference' (NLI),\nthat cannot be solved using only word-level knowledge and requires some\ncompositionality. We find that the performance of state of the art sentence\nembeddings (InferSent; Conneau et al., 2017) on our new dataset is poor. We\nanalyze the decision rules learned by InferSent and find that they are\nconsistent with simple heuristics that are ecologically valid in its training\ndataset. Further, we find that augmenting training with our dataset improves\ntest performance on our dataset without loss of performance on the original\ntraining dataset. This highlights the importance of structured datasets in\nbetter understanding and improving AI systems.","url_abs":"http://arxiv.org/abs/1802.04302v2","url_pdf":"http://arxiv.org/pdf/1802.04302v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"evaluating-compositionality-in-sentence","repo_url":"https://github.com/ishita-dg/ScrambleTests","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"natural-language-inference","task_name":"Natural Language Inference"},{"task_slug":"sentence","task_name":"Sentence"},{"task_slug":"sentence-embeddings","task_name":"Sentence Embeddings"},{"task_slug":null,"task_name":"valid"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1802.04302","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}