{"id":883221,"date":"2022-10-05T13:05:01","date_gmt":"2022-10-05T20:05:01","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/"},"modified":"2023-06-19T10:12:17","modified_gmt":"2023-06-19T17:12:17","slug":"the-whole-truth-and-nothing-but-the-truth-faithful-and-controllable-dialogue-response-generation-with-dataflow-transduction-and-constrained-decoding","status":"publish","type":"msr-research-item","link":"https:\/\/www.microsoft.com\/en-us\/research\/publication\/the-whole-truth-and-nothing-but-the-truth-faithful-and-controllable-dialogue-response-generation-with-dataflow-transduction-and-constrained-decoding\/","title":{"rendered":"The Whole Truth and Nothing But the Truth: Faithful and Controllable Dialogue Response Generation with Dataflow Transduction and Constrained Decoding"},"content":{"rendered":"<p><span dir=\"ltr\" role=\"presentation\">In a real-world dialogue system, generated text <\/span><span dir=\"ltr\" role=\"presentation\">must be truthful and informative while remain<\/span><span dir=\"ltr\" role=\"presentation\">ing fluent and adhering to a prescribed style. <\/span><span dir=\"ltr\" role=\"presentation\">Satisfying these constraints simultaneously is <\/span><span dir=\"ltr\" role=\"presentation\">difficult for the two predominant paradigms in <\/span><span dir=\"ltr\" role=\"presentation\">language generation: neural language model<\/span><span dir=\"ltr\" role=\"presentation\">ing and rule-based generation. We describe a <\/span><span dir=\"ltr\" role=\"presentation\">hybrid architecture for dialogue response gen<\/span><span dir=\"ltr\" role=\"presentation\">eration that combines the strengths of both <\/span><span dir=\"ltr\" role=\"presentation\">paradigms. The first component of this archi<\/span><span dir=\"ltr\" role=\"presentation\">tecture is a rule-based content selection model <\/span><span dir=\"ltr\" role=\"presentation\">defined using a new formal framework called <\/span><span dir=\"ltr\" role=\"presentation\">dataflow transduction<\/span><span dir=\"ltr\" role=\"presentation\">,<\/span> <span dir=\"ltr\" role=\"presentation\">which uses declara<\/span><span dir=\"ltr\" role=\"presentation\">tive rules to transduce a dialogue agent\u2019s ac<\/span><span dir=\"ltr\" role=\"presentation\">tions and their results (represented as dataflow <\/span><span dir=\"ltr\" role=\"presentation\">graphs) into context-free grammars represent<\/span><span dir=\"ltr\" role=\"presentation\">ing the space of contextually acceptable re<\/span><span dir=\"ltr\" role=\"presentation\">sponses.<\/span> <span dir=\"ltr\" role=\"presentation\">The second component is a con<\/span><span dir=\"ltr\" role=\"presentation\">strained decoding procedure that uses these <\/span><span dir=\"ltr\" role=\"presentation\">grammars to constrain the output of a neu<\/span><span dir=\"ltr\" role=\"presentation\">ral language model, which selects fluent utter<\/span><span dir=\"ltr\" role=\"presentation\">ances. Our experiments show that this system <\/span><span dir=\"ltr\" role=\"presentation\">outperforms both rule-based and learned ap<\/span><span dir=\"ltr\" role=\"presentation\">proaches in human evaluations of fluency, rel<\/span><span dir=\"ltr\" role=\"presentation\">evance, and truthfulness.<\/span><\/p>\n","protected":false},"excerpt":{"rendered":"<p>In a real-world dialogue system, generated text must be truthful and informative while remaining fluent and adhering to a prescribed style. Satisfying these constraints simultaneously is difficult for the two predominant paradigms in language generation: neural language modeling and rule-based generation. We describe a hybrid architecture for dialogue response generation that combines the strengths of [&hellip;]<\/p>\n","protected":false},"featured_media":0,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"footnotes":""},"msr-content-type":[3],"msr-research-highlight":[],"research-area":[13556,13545],"msr-publication-type":[193716],"msr-product-type":[],"msr-focus-area":[],"msr-platform":[],"msr-download-source":[],"msr-locale":[268875],"msr-field-of-study":[263203],"msr-conference":[],"msr-journal":[],"msr-impact-theme":[],"msr-pillar":[],"class_list":["post-883221","msr-research-item","type-msr-research-item","status-publish","hentry","msr-research-area-artificial-intelligence","msr-research-area-human-language-technologies","msr-locale-en_us","msr-field-of-study-computation-and-language"],"msr_publishername":"","msr_edition":"","msr_affiliation":"","msr_published_date":"2023-7-1","msr_host":"","msr_duration":"","msr_version":"","msr_speaker":"","msr_other_contributors":"","msr_booktitle":"","msr_pages_string":"","msr_chapter":"","msr_isbn":"","msr_journal":"","msr_volume":"","msr_number":"","msr_editors":"","msr_series":"","msr_issue":"","msr_organization":"","msr_how_published":"","msr_notes":"","msr_highlight_text":"","msr_release_tracker_id":"","msr_original_fields_of_study":"","msr_download_urls":"","msr_external_url":"","msr_secondary_video_url":"","msr_longbiography":"","msr_microsoftintellectualproperty":1,"msr_main_download":"","msr_publicationurl":"","msr_doi":"","msr_publication_uploader":[{"type":"url","viewUrl":"false","id":"false","title":"https:\/\/arxiv.org\/abs\/2209.07800","label_id":"243109","label":0}],"msr_related_uploader":"","msr_attachments":[],"msr-author-ordering":[{"type":"user_nicename","value":"Hao Fang","user_id":39465,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Hao Fang"},{"type":"user_nicename","value":"Anusha Balakrishnan","user_id":39549,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Anusha Balakrishnan"},{"type":"user_nicename","value":"Harsh Jhamtani","user_id":42132,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Harsh Jhamtani"},{"type":"guest","value":"john-bufe","user_id":765268,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=john-bufe"},{"type":"user_nicename","value":"Jean Crawford","user_id":39486,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Jean Crawford"},{"type":"user_nicename","value":"Jayant Krishnamurthy","user_id":39519,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Jayant Krishnamurthy"},{"type":"user_nicename","value":"Adam Pauls","user_id":39483,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Adam Pauls"},{"type":"user_nicename","value":"Jason Eisner","user_id":39970,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Jason Eisner"},{"type":"user_nicename","value":"Jacob Andreas","user_id":39138,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Jacob Andreas"},{"type":"user_nicename","value":"Dan Klein","user_id":40909,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Dan Klein"}],"msr_impact_theme":[],"msr_research_lab":[],"msr_event":[945648],"msr_group":[907656],"msr_project":[],"publication":[],"video":[],"download":[],"msr_publication_type":"inproceedings","_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/883221"}],"collection":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item"}],"about":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-research-item"}],"version-history":[{"count":2,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/883221\/revisions"}],"predecessor-version":[{"id":944565,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/883221\/revisions\/944565"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=883221"}],"wp:term":[{"taxonomy":"msr-content-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-content-type?post=883221"},{"taxonomy":"msr-research-highlight","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-highlight?post=883221"},{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=883221"},{"taxonomy":"msr-publication-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-type?post=883221"},{"taxonomy":"msr-product-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-product-type?post=883221"},{"taxonomy":"msr-focus-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-focus-area?post=883221"},{"taxonomy":"msr-platform","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-platform?post=883221"},{"taxonomy":"msr-download-source","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-download-source?post=883221"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=883221"},{"taxonomy":"msr-field-of-study","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-field-of-study?post=883221"},{"taxonomy":"msr-conference","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-conference?post=883221"},{"taxonomy":"msr-journal","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-journal?post=883221"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=883221"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=883221"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}