{"id":"https://openalex.org/W6963164566","doi":"https://doi.org/10.17615/ewb9-sz07","title":"Inducing and Interpreting Compositionality in Neural NLP Models","display_name":"Inducing and Interpreting Compositionality in Neural NLP Models","publication_year":2024,"publication_date":"2024-08-17","ids":{"openalex":"https://openalex.org/W6963164566","doi":"https://doi.org/10.17615/ewb9-sz07"},"language":"en","primary_location":{"id":"doi:10.17615/ewb9-sz07","is_oa":true,"landing_page_url":"https://doi.org/10.17615/ewb9-sz07","pdf_url":null,"source":{"id":"https://openalex.org/S7407051488","display_name":"UNC Libraries","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"thesis"},"type":"dissertation","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.17615/ewb9-sz07","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Jiang, Yichen","orcid":null},"institutions":[{"id":"https://openalex.org/I114027177","display_name":"University of North Carolina at Chapel Hill","ror":"https://ror.org/0130frc33","country_code":"US","type":"education","lineage":["https://openalex.org/I114027177"]}],"countries":["US"],"is_corresponding":true,"raw_author_name":"Jiang, Yichen","raw_affiliation_strings":["University of North Carolina at Chapel Hill"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of North Carolina at Chapel Hill","institution_ids":["https://openalex.org/I114027177"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I114027177"],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"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":true,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.6567000150680542,"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":0.6567000150680542,"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/T11273","display_name":"Advanced Graph Neural Networks","score":0.1014999970793724,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.04769999906420708,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/principle-of-compositionality","display_name":"Principle of compositionality","score":0.8737999796867371},{"id":"https://openalex.org/keywords/artificial-neural-network","display_name":"Artificial neural network","score":0.6215000152587891},{"id":"https://openalex.org/keywords/modular-design","display_name":"Modular design","score":0.609499990940094},{"id":"https://openalex.org/keywords/adversarial-system","display_name":"Adversarial system","score":0.5652999877929688},{"id":"https://openalex.org/keywords/deep-neural-networks","display_name":"Deep neural networks","score":0.45750001072883606},{"id":"https://openalex.org/keywords/scale","display_name":"Scale (ratio)","score":0.37940001487731934},{"id":"https://openalex.org/keywords/modular-neural-network","display_name":"Modular neural network","score":0.3425999879837036}],"concepts":[{"id":"https://openalex.org/C121375916","wikidata":"https://www.wikidata.org/wiki/Q936559","display_name":"Principle of compositionality","level":2,"score":0.8737999796867371},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7889000177383423},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.7462999820709229},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.6215000152587891},{"id":"https://openalex.org/C101468663","wikidata":"https://www.wikidata.org/wiki/Q1620158","display_name":"Modular design","level":2,"score":0.609499990940094},{"id":"https://openalex.org/C37736160","wikidata":"https://www.wikidata.org/wiki/Q1801315","display_name":"Adversarial system","level":2,"score":0.5652999877929688},{"id":"https://openalex.org/C2984842247","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep neural networks","level":3,"score":0.45750001072883606},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.43790000677108765},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.3865000009536743},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.37940001487731934},{"id":"https://openalex.org/C2781121602","wikidata":"https://www.wikidata.org/wiki/Q3504403","display_name":"Modular neural network","level":4,"score":0.3425999879837036},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.31709998846054077},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.3109999895095825},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.28439998626708984},{"id":"https://openalex.org/C189950617","wikidata":"https://www.wikidata.org/wiki/Q937228","display_name":"Property (philosophy)","level":2,"score":0.2782999873161316},{"id":"https://openalex.org/C44291984","wikidata":"https://www.wikidata.org/wiki/Q1074173","display_name":"Question answering","level":2,"score":0.2750000059604645},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.27219998836517334}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.17615/ewb9-sz07","is_oa":true,"landing_page_url":"https://doi.org/10.17615/ewb9-sz07","pdf_url":null,"source":{"id":"https://openalex.org/S7407051488","display_name":"UNC Libraries","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"thesis"}],"best_oa_location":{"id":"doi:10.17615/ewb9-sz07","is_oa":true,"landing_page_url":"https://doi.org/10.17615/ewb9-sz07","pdf_url":null,"source":{"id":"https://openalex.org/S7407051488","display_name":"UNC Libraries","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"thesis"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Human":[0],"intelligence":[1],"demonstrates":[2],"compositionality,":[3,36],"the":[4,28,98,123,142,145,163,171,199,226,232,306,322,363,368,372,386,417,436,453],"algebraic":[5],"capacity":[6],"that":[7,103,127,256],"enables":[8],"us":[9],"to":[10,33,41,43,92,117,121,133,175,197,209,248,253,265,280,291,320,341,395,452],"understand":[11],"and":[12,37,53,88,100,125,139,152,158,204,221,251,277,282,298,314,325,337,384,411,423,426,448],"produce":[13],"a":[14,58,106,115,154,195,316,377,390],"potentially":[15],"infinite":[16],"number":[17],"of":[18,21,49,136,295,438,459],"novel":[19],"combinations":[20],"known":[22],"components.":[23],"Neural":[24],"NLP":[25,64,84,275,407],"models,":[26],"on":[27,300,334,381],"other":[29],"hand,":[30],"are":[31,131,147,241,246,278],"shown":[32],"lack":[34],"systematic":[35],"thus":[38],"often":[39],"fail":[40],"generalize":[42,290],"expressions":[44],"with":[45,393],"an":[46,263,443],"unseen":[47,74],"composition":[48],"seen":[50,54],"linguistic":[51,296],"structures":[52,297],"constituents.":[55],"This":[56,77,213],"poses":[57],"huge":[59],"risk":[60],"for":[61,66,81,404,430,456],"using":[62],"neural":[63,83,185,190,214,238,406,432],"models":[65,130,191,236],"real-world":[67],"tasks":[68,255,276],"as":[69,244,262],"theirbehavior":[70],"is":[71,222],"unpredictable":[72],"given":[73,442],"textual":[75],"inputs.":[76],"thesis":[78,401],"presents":[79,402],"methods":[80,403],"analyzing":[82,405],"models\u2019":[85,408],"compositional/uncompositional":[86,409],"behavior":[87,410],"improving":[89,412],"their":[90,413],"ability":[91,272],"compostionally":[93],"generalize.":[94],"First,":[95],"I":[96,112,149,180,193,304,327,356,374,434],"reveal":[97,126,435],"biases":[99],"reasoning":[101,137,168,220,259,271],"shortcuts":[102,124,138],"exist":[104],"in":[105,273,310,440],"popular":[107],"multi-hop":[108],"question-answering":[109],"(QA)":[110],"dataset.":[111],"then":[113,328],"propose":[114,181,315,376,449],"method":[116,319],"create":[118],"adversarial":[119,227],"documents":[120],"eliminate":[122],"state-of-the-art":[128],"QA":[129],"susceptible":[132],"such":[134],"kind":[135],"cannot":[140,289],"find":[141],"answer":[143],"when":[144],"adversaries":[146],"present.":[148],"also":[150,375],"collect":[151],"release":[153],"many-hop":[155],"fact":[156],"extraction":[157],"claim":[159],"verification":[160],"dataset":[161],"where":[162],"claims":[164],"embody":[165],"more":[166,223],"diverse":[167],"graphs":[169],"than":[170],"existing":[172],"datasets.":[173,303],"Second,":[174],"tackle":[176],"compositional":[177,184,219,235,428,460],"question":[178,200],"answering,":[179],"two":[182],"explicitly":[183,218,234,425],"frameworks":[186,429],"by":[187,366,415],"combining":[188],"multiple":[189,202,206,301],"together.":[192],"train":[194],"controller":[196],"decompose":[198],"into":[201,362,371],"sub-questions":[203],"assemble":[205],"LSTM":[207],"modules":[208],"address":[210],"each":[211],"sub-questions.":[212],"modular":[215,239,266],"network":[216],"conducts":[217],"robust":[224],"against":[225],"evaluation":[228],"we":[229],"constructed.":[230],"Despite":[231],"success,":[233],"like":[237],"networks":[240],"inherently":[242],"limited":[243],"they":[245,288],"difficult":[247],"scale":[249,281],"up":[250],"transfer":[252],"new":[254],"require":[257],"different":[258],"skills.":[260],"Transformers,":[261],"alternative":[264],"networks,":[267],"demonstrate":[268],"strong":[269],"in-distribution":[270],"many":[274],"easier":[279],"transfer.":[283],"However,":[284],"without":[285],"extensive":[286],"pretraining,":[287],"out-of-distribution":[292],"(OOD)":[293],"compositions":[294,313],"constituents":[299],"diagnosis":[302],"analyze":[305],"Transformer\u2019s":[307],"failure":[308],"mode":[309],"processing":[311],"OOD":[312],"data":[317,323],"augmentation":[318],"increase":[321],"complexity":[324],"diversity.":[326],"introduce":[329],"Mutual":[330],"Exclusivity":[331],"Training":[332],"based":[333],"unlikelihood":[335],"loss":[336],"auxiliary":[338],"regularization":[339,421],"objectives":[340],"induce":[342],"stronger":[343],"implicit":[344,351],"compositionality":[345,352,414,439],"from":[346],"Transformers.":[347],"To":[348],"further":[349],"encourage":[350],"under":[353],"low-complexity":[354],"data,":[355,419],"study":[357],"injecting":[358],"structural":[359],"inductive":[360],"bias":[361],"Transformer":[364,454],"architecture":[365,455],"incorporating":[367],"tensor-product":[369],"representation":[370],"attention.":[373],"structure-oriented":[378],"vector":[379],"quantization":[380],"lexical":[382],"embeddings":[383],"regularize":[385],"attention":[387],"outputs":[388],"towards":[389],"soft":[391],"invariance":[392],"respect":[394],"syntactic":[396],"structure.":[397],"In":[398],"summary,":[399],"this":[400],"augmenting":[416],"training":[418,431,446],"proposing":[420],"objectives,":[422],"developing":[424],"implicitly":[427],"networks.":[433],"emergence":[437],"Transformers":[441],"increasingly":[444],"complex":[445],"set,":[447],"promising":[450],"improvements":[451],"data-efficient":[457],"learning":[458],"natural":[461],"language":[462],"processing.":[463]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2025-10-10T00:00:00"}