{"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/mutual-information-alleviates-hallucinations","title":"Mutual Information Alleviates Hallucinations in Abstractive Summarization","arxiv_id":"2210.13210","date":"2022-10-24","proceeding":null,"authors":["Liam van der Poel","Ryan Cotterell","Clara Meister"],"abstract":"Despite significant progress in the quality of language generated from abstractive summarization models, these models still exhibit the tendency to hallucinate, i.e., output content not supported by the source document. A number of works have tried to fix--or at least uncover the source of--the problem with limited success. In this paper, we identify a simple criterion under which models are significantly more likely to assign more probability to hallucinated content during generation: high model uncertainty. This finding offers a potential explanation for hallucinations: models default to favoring text with high marginal probability, i.e., high-frequency occurrences in the training set, when uncertain about a continuation. It also motivates possible routes for real-time intervention during decoding to prevent such hallucinations. We propose a decoding strategy that switches to optimizing for pointwise mutual information of the source and target token--rather than purely the probability of the target token--when the model exhibits uncertainty. Experiments on the XSum dataset show that our method decreases the probability of hallucinated tokens while maintaining the Rouge and BertS scores of top-performing decoding strategies.","url_abs":"https://arxiv.org/abs/2210.13210v2","url_pdf":"https://arxiv.org/pdf/2210.13210v2.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":"mutual-information-alleviates-hallucinations","repo_url":"https://github.com/vanderpoelliam/cpmi","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"mutual-information-alleviates-hallucinations","repo_url":"https://github.com/ljyflores/loss-library","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"mutual-information-alleviates-hallucinations","repo_url":"https://github.com/iinemo/lm-polygraph","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"abstractive-text-summarization","task_name":"Abstractive Text Summarization"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/2210.13210","atlas_url":"https://app.syntology.ai/?focus=2210.13210","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2210.13210"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-25T09:33:49+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/iinemo/lm-polygraph","reach":null},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/vanderpoelliam/cpmi","reach":{"status":"ok","spdx":"MIT"}},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/ljyflores/loss-library","reach":null}],"summary":{"ran":9,"ran_draft_wrong":1,"unverified":2},"by_repo_kind":{"official":{"samples":8,"ran":8,"repositories":1},"listed":{"samples":4,"ran":2,"repositories":2}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":1,"samples":[{"code_sha256_prefix":"2b4c65c44bfcdd3b","entry":"MeanConditionalPointwiseMutualInformation","repo":"iinemo/lm-polygraph","repo_kind":"listed","path":"src/lm_polygraph/estimators/conditional_pointwise_mutual_information.py","file_url":"https://github.com/iinemo/lm-polygraph/blob/HEAD/src/lm_polygraph/estimators/conditional_pointwise_mutual_information.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":"well_formed","behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"2b4c65c44bfcdd3b"}},{"code_sha256_prefix":"4e5db0a9eb2ef73f","entry":"decode","repo":"vanderpoelliam/cpmi","repo_kind":"official","path":"src/bart_file_bpe_decoder.py","file_url":"https://github.com/vanderpoelliam/cpmi/blob/HEAD/src/bart_file_bpe_decoder.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"4e5db0a9eb2ef73f"}},{"code_sha256_prefix":"62093950a87046d6","entry":"encode","repo":"vanderpoelliam/cpmi","repo_kind":"official","path":"src/detect_hallucination/process_labels.py","file_url":"https://github.com/vanderpoelliam/cpmi/blob/HEAD/src/detect_hallucination/process_labels.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"62093950a87046d6"}},{"code_sha256_prefix":"a380328ed69263c4","entry":"extract_lines","repo":"vanderpoelliam/cpmi","repo_kind":"official","path":"src/evaluate_decoding.py","file_url":"https://github.com/vanderpoelliam/cpmi/blob/HEAD/src/evaluate_decoding.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"a380328ed69263c4"}},{"code_sha256_prefix":"d104b10430234d01","entry":"polygraph_module_init","repo":"iinemo/lm-polygraph","repo_kind":"listed","path":"src/lm_polygraph/estimators/conditional_pointwise_mutual_information.py","file_url":"https://github.com/iinemo/lm-polygraph/blob/HEAD/src/lm_polygraph/estimators/conditional_pointwise_mutual_information.py","link_basis":"first_harvest_node","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"OUTPUT_MISDECLARED","metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"d104b10430234d01"}},{"code_sha256_prefix":"45525b77cc521f6b","entry":"process_raw","repo":"vanderpoelliam/cpmi","repo_kind":"official","path":"src/detect_hallucination/process_labels.py","file_url":"https://github.com/vanderpoelliam/cpmi/blob/HEAD/src/detect_hallucination/process_labels.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"45525b77cc521f6b"}},{"code_sha256_prefix":"97289c22f4f957e9","entry":"read_labels","repo":"vanderpoelliam/cpmi","repo_kind":"official","path":"src/evaluate_decoding.py","file_url":"https://github.com/vanderpoelliam/cpmi/blob/HEAD/src/evaluate_decoding.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"97289c22f4f957e9"}},{"code_sha256_prefix":"70c754919b846528","entry":"read_lines","repo":"vanderpoelliam/cpmi","repo_kind":"official","path":"src/detect_hallucination/process_labels.py","file_url":"https://github.com/vanderpoelliam/cpmi/blob/HEAD/src/detect_hallucination/process_labels.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"70c754919b846528"}},{"code_sha256_prefix":"5685319232b466bb","entry":"read_lines","repo":"vanderpoelliam/cpmi","repo_kind":"official","path":"src/factCC/generate_json_data.py","file_url":"https://github.com/vanderpoelliam/cpmi/blob/HEAD/src/factCC/generate_json_data.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"5685319232b466bb"}},{"code_sha256_prefix":"8d2d77ecb104da85","entry":"read_logs","repo":"vanderpoelliam/cpmi","repo_kind":"official","path":"src/evaluate_decoding.py","file_url":"https://github.com/vanderpoelliam/cpmi/blob/HEAD/src/evaluate_decoding.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"8d2d77ecb104da85"}},{"code_sha256_prefix":"733788d50f9f0f9e","entry":"Estimator","repo":"iinemo/lm-polygraph","repo_kind":"listed","path":"src/lm_polygraph/estimators/conditional_pointwise_mutual_information.py","file_url":"https://github.com/iinemo/lm-polygraph/blob/HEAD/src/lm_polygraph/estimators/conditional_pointwise_mutual_information.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"733788d50f9f0f9e"}},{"code_sha256_prefix":"e03b631822bcf5a3","entry":"mutual_information_loss","repo":"ljyflores/loss-library","repo_kind":"listed","path":"src/loss_library/utils_mutual_info_loss.py","file_url":"https://github.com/ljyflores/loss-library/blob/HEAD/src/loss_library/utils_mutual_info_loss.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"e03b631822bcf5a3"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}