{"id":"https://openalex.org/W7214978022","doi":"https://doi.org/10.48550/arxiv.2609.37391","title":"Rethinking Soft Tokens for Parallel Decoding in Diffusion Language Models","display_name":"Rethinking Soft Tokens for Parallel Decoding in Diffusion Language Models","publication_year":2026,"publication_date":"2026-09-29","ids":{"openalex":"https://openalex.org/W7214978022","doi":"https://doi.org/10.48550/arxiv.2609.37391"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.37391","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.37391","pdf_url":"https://arxiv.org/pdf/2609.37391","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2609.37391","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5153642345","display_name":"Kodai Kawamura","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I165932596","display_name":"National University of Singapore","ror":"https://ror.org/01tgyzw49","country_code":"SG","type":"education","lineage":["https://openalex.org/I165932596"]}],"countries":["SG"],"is_corresponding":false,"raw_author_name":"Kawamura, Kodai","raw_affiliation_strings":["National University of Singapore"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"National University of Singapore","institution_ids":["https://openalex.org/I165932596"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5003184366","display_name":"Kenji Kawaguchi","orcid":"https://orcid.org/0000-0003-1839-7504","observed_orcids":["https://orcid.org/0000-0003-1839-7504","https://orcid.org/0000-0002-0791-0870","https://orcid.org/0000-0002-5361-9793"]},"institutions":[{"id":"https://openalex.org/I165932596","display_name":"National University of Singapore","ror":"https://ror.org/01tgyzw49","country_code":"SG","type":"education","lineage":["https://openalex.org/I165932596"]}],"countries":["SG"],"is_corresponding":false,"raw_author_name":"Kawaguchi, Kenji","raw_affiliation_strings":["National University of Singapore"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"National University of Singapore","institution_ids":["https://openalex.org/I165932596"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5153617806","display_name":"Anji Liu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I165932596","display_name":"National University of Singapore","ror":"https://ror.org/01tgyzw49","country_code":"SG","type":"education","lineage":["https://openalex.org/I165932596"]}],"countries":["SG"],"is_corresponding":false,"raw_author_name":"Liu, Anji","raw_affiliation_strings":["National University of Singapore"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"National University of Singapore","institution_ids":["https://openalex.org/I165932596"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I165932596"],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.5687999725341797,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.5687999725341797,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.33329999446868896,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.03440000116825104,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/parallel-decoding","display_name":"parallel decoding","score":0.8610000014305115},{"id":"https://openalex.org/keywords/diffusion-language-models","display_name":"diffusion language models","score":0.9980000257492065},{"id":"https://openalex.org/keywords/embedding-space","display_name":"embedding space","score":0.7039999961853027},{"id":"https://openalex.org/keywords/uncertainty-preservation","display_name":"uncertainty preservation","score":0.7229999899864197}],"concepts":[{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.8165000081062317},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7354000210762024},{"id":"https://openalex.org/C41608201","wikidata":"https://www.wikidata.org/wiki/Q980509","display_name":"Embedding","level":2,"score":0.6883000135421753},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.6646000146865845},{"id":"https://openalex.org/C2780801425","wikidata":"https://www.wikidata.org/wiki/Q5164392","display_name":"Construct (python library)","level":2,"score":0.5931000113487244},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.5569000244140625},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.5188999772071838},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.414900004863739},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.4115999937057495},{"id":"https://openalex.org/C69357855","wikidata":"https://www.wikidata.org/wiki/Q163214","display_name":"Diffusion","level":2,"score":0.38989999890327454},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.38359999656677246},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.3765000104904175},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.37540000677108765},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.35920000076293945},{"id":"https://openalex.org/C28719098","wikidata":"https://www.wikidata.org/wiki/Q44946","display_name":"Point (geometry)","level":2,"score":0.31049999594688416},{"id":"https://openalex.org/C129782007","wikidata":"https://www.wikidata.org/wiki/Q162886","display_name":"Euclidean geometry","level":2,"score":0.3068000078201294},{"id":"https://openalex.org/C43126263","wikidata":"https://www.wikidata.org/wiki/Q128751","display_name":"Source code","level":2,"score":0.2912999987602234},{"id":"https://openalex.org/C149441793","wikidata":"https://www.wikidata.org/wiki/Q200726","display_name":"Probability distribution","level":2,"score":0.29109999537467957},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.2825999855995178}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.37391","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.37391","pdf_url":"https://arxiv.org/pdf/2609.37391","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2609.37391","doi":"https://doi.org/10.48550/arxiv.2609.37391","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.37391","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2609.37391","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.37391","pdf_url":"https://arxiv.org/pdf/2609.37391","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7214978022.pdf","grobid_xml":"https://content.openalex.org/works/W7214978022.grobid-xml"},"referenced_works_count":10,"referenced_works":["https://openalex.org/W3177813494","https://openalex.org/W3195140297","https://openalex.org/W3210904070","https://openalex.org/W4317539509","https://openalex.org/W4416051103","https://openalex.org/W4416143562","https://openalex.org/W4417529490","https://openalex.org/W7133299406","https://openalex.org/W7152996795","https://openalex.org/W7202111018"],"related_works":["https://openalex.org/W7164886286","https://openalex.org/W7131895179","https://openalex.org/W7117881514","https://openalex.org/W7202347597","https://openalex.org/W7136370632","https://openalex.org/W7163534824","https://openalex.org/W7223027578","https://openalex.org/W7117967698","https://openalex.org/W7214976080","https://openalex.org/W7223389275"],"abstract_inverted_index":{"Diffusion":[0],"language":[1],"models":[2],"(DLMs)":[3],"enable":[4],"parallel":[5,70,159,185],"generation":[6],"by":[7,37],"predicting":[8],"and":[9,115,175,178,187],"committing":[10],"multiple":[11],"tokens":[12,58],"at":[13,50],"each":[14],"denoising":[15],"step,":[16],"yet":[17],"they":[18],"can":[19,33],"generate":[20],"individually":[21],"plausible":[22],"but":[23],"mutually":[24],"inconsistent":[25],"tokens.":[26,132],"Recent":[27],"work":[28],"shows":[29],"that":[30,139,165],"\\emph{soft":[31],"tokens}":[32],"mitigate":[34],"this":[35,78,82,122],"issue":[36],"representing":[38],"uncertain":[39],"positions":[40],"with":[41],"continuous":[42],"embeddings":[43],"built":[44],"from":[45],"the":[46,51,93,116],"model's":[47],"predictive":[48,64],"distribution":[49],"previous":[52],"decoding":[53,71,186],"step.":[54],"However,":[55],"although":[56],"soft":[57,131],"are":[59],"commonly":[60],"understood":[61],"as":[62],"preserving":[63],"uncertainty,":[65],"how":[66,147,155],"soft-token":[67,90,100,113,136,156,191],"feedback":[68,91,137,157],"improves":[69,158],"has":[72],"not":[73,144],"been":[74],"systematically":[75],"examined.":[76],"In":[77],"paper,":[79],"we":[80,106,124,161],"investigate":[81],"question":[83],"in":[84,102],"frozen":[85],"pretrained":[86,117,173],"DLMs":[87,174],"to":[88],"examine":[89],"without":[92],"effects":[94],"of":[95,130,135],"additional":[96],"training.":[97],"To":[98,152],"construct":[99],"inputs":[101],"a":[103,108,126,188],"training-free":[104,189],"setting,":[105],"identify":[107],"geometric":[109],"mismatch":[110],"between":[111],"conventional":[112],"construction":[114,129],"embedding":[118],"space.":[119],"Based":[120],"on":[121],"observation,":[123],"propose":[125],"training-free,":[127],"geometry-aware":[128],"Our":[133],"analysis":[134],"suggests":[138],"uncertainty":[140],"preservation":[141],"alone":[142],"does":[143],"fully":[145],"explain":[146,154],"it":[148,166],"reshapes":[149],"subsequent":[150],"predictions.":[151],"better":[153],"decoding,":[160],"provide":[162],"empirical":[163],"evidence":[164],"favors":[167],"coherent":[168],"token":[169],"sequences.":[170],"Across":[171],"four":[172,176],"math":[177],"code":[179],"benchmarks,":[180],"our":[181],"method":[182],"outperforms":[183],"standard":[184],"Euclidean":[190],"baseline.":[192],"Code:":[193],"https://github.com/kodaikawamura/rethinking-soft-tokens":[194]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-10-01T00:00:00"}