Upgrade to PRO for Only $50/Year—Limited-Time Offer! 🔥

時間情報表現抽出とルールベース解析器のこれから / Temporal Expression A...

yag_ays
April 08, 2022

時間情報表現抽出とルールベース解析器のこれから / Temporal Expression Analysis in Japanese and Future of Rule-based Approach

【NLP Hacks vol.3】『実装』に特化した、NLP勉強会コミュニティ開催!
https://connpass.com/event/241079/

yag_ays

April 08, 2022
Tweet

More Decks by yag_ays

Other Decks in Research

Transcript

  1. 3 ࠓ೔͓࿩͢Δ͜ͱ • ࣌ؒ৘ใදݱΛղੳ͢Δja-timexʹ͍ͭͯ • ࣌ؒ৘ใදݱͷநग़/ن֨Խͱ͍͏λεΫʹ͍ͭͯઆ໌͠·͢ • ࢲ͕࡞੒ͨ࣌ؒ͠৘ใදݱΛղੳ͢ΔϥΠϒϥϦΛ঺հ͠·͢ • ͲͷΑ͏ʹ࣮૷͍͔ͯͬͨ͠ͷഎܠ΍ۤ࿑΋؆୯ʹ঺հ͠·͢

    • ϧʔϧϕʔεղੳثͱ࣌ؒ৘ใදݱͷ͜Ε͔Β • ϧʔϧϕʔεͬͯࠓͲ͖Ͳ͏ͳͷʁେن໛ݴޠϞσϧʹஔ͖׵Θ͍ͬͯ͘ͷʁͱ͍͏ٙ໰Λɹ ߟ͑ͯΈ·͢ • طଘݚڀͳͲ΋౿·͑ͯɺ࣌ؒ৘ใදݱͷࠓޙʹ͍ͭͯߟ͑ͯΈ·͢
  2. 7 ࣌ؒ৘ใදݱͰλά෇͚͢ΔTIMEX3ܗࣜ [1] ൴͸2008೥4݄͔Βिʹ3ճͷδϣΪϯάΛ ே8͔࣌Β1࣌ؒߦ͖ͬͯͨ 2008೥4݄ ே8࣌ 1࣌ؒ िʹ3ճ "2008-04-XX"

    "T08-XX-XX" "PT1H" "P1W", "3X" DATE ೔෇දݱ TIME ࣌ؒදݱ DURATION ࣋ଓ࣌ؒදݱ SET ස౓ू߹දݱ ن֨Խ நग़
  3. 8 • ࠷ॳͷ͖͔͚ͬ (2021/07ࠒ) • NAISTߥ຀ݚओ࠵ͷNTCIR16 Real-MedNLPͱ͍͏γΣΞʔυλεΫ͕ެ։͞Εͨ • ຊλεΫͷݻ༗දݱநग़ͷΤϯςΟςΟͷ1ͭʹ࣌ؒ৘ใදݱ͕͋ͬͨ •

    Real-MedNLPͰͷར༻΋ߟ͑ɺ͍ͣΕࣗࣾͷϓϩμΫτ։ൃ౳ͰඞཁʹͳΔͩΖ͏ͳͱࢥͬͯ நग़ΤϯδϯΛ࡞Γ࢝ΊͨΒɺ൚༻తͳ΋ͷʹͨ͘͠ͳͬͨ (खஈͷ໨తԽ) • ҩྍυϝΠϯʹ͓͚Δ࣌ؒ৘ใදݱͷॏཁ͞ • ిࢠΧϧςจॻɺ࿦จͷ঱ྫใࠂͳͲͰ͸ɺ࣌ؒ৘ใදݱͱࣄ৅͕ϖΞͰهࡌ͞ΕΔ • ױऀ͕ૌ͑Δ঱ঢ়ͷൃੜ࣌ظ΍͔ͦ͜Βͷܦա (׮ղ/૿ѱ) ͸ɺ਍࡯ʹ͓͍ͯॏཁ • Temporal Relation Extraction, Temporal reasoningͳͲͱݺ͹ΕΔλεΫ [2] [3] ja-timex࡞੒ͷϞνϕʔγϣϯ - ཧ೦ฤ
  4. 11 • ༰қʹར༻Ͱ͖Δ΋ͷ͕౰࣌ݶΒΕ͍ͯͨ • ϧʔϧϕʔε࣮૷͸੒ᖒΒ[4] ͷ࿦จ࣮૷Ͱ͋ΔnormalizeNumexp ͷΈ • C++࣮૷ͰɺPythonόΠϯσΟϯά͸͋Δ΋ͷͷखݩͰಈ͔͢͜ͱ͕Ͱ͖ͳ͔ͬͨ •

    ͳ͓ɺݱࡏͰ͸ pynormalizenumexp ͱ͍͏PythonͰͷ࠶࣮૷͕ଘࡏ͢Δ • spaCyͷݻ༗දݱநग़Ϟσϧ͸༰қʹར༻Ͱ͖ͨ • DATEͱTIMEʹରԠ͍͕ͯͨ͠ɺ೔෇දݱʹऑ͍ͳͲਫ਼౓ͷ՝୊Λײͨ͡ • ࣌ؒ৘ใදݱΛநग़͢ΔͷΈͰɺ͔ͦ͜ΒPythonͷdatetime౳΁ͷม׵͸Ͱ͖ͳ͍ • σʔληοτ͕ͳ͔ͬͨ • TIMEX3ܗࣜ౳Ͱ࣌ؒ৘ใදݱ͕Ξϊςʔγϣϯ͞Ε͍ͯΔσʔληοτ͕ར༻Ͱ͖ͳ͔ͬͨ • CRF/ܥྻϞσϧ/BERT౳Ͱɺؾܰʹڭࢣ͋ΓֶशͷNERΛߏஙͰ͖ͳ͔ͬͨ ja-timex࡞੒ͷϞνϕʔγϣϯ - ࣮૷ฤ
  5. 14 ͲͷΑ͏ʹϧʔϧΛهड़͍ͯ͠Δ͔ʁ 1. ࣌ؒ৘ใදݱͷ਺஋෦෼ΛάϧʔϓԽ • ਖ਼نදݱͰͷ਺஋දݱͷऔಘͱՄಡੑ ୲อͷͨΊʹɺ໊લ෇͖άϧʔϓΛ ࡞͓ͬͯ͘ 2. ࣌ؒ৘ใදݱͷύλʔϯΛਖ਼نදݱͰߏங

    • จࣈྻ্Ͱදݱ͞Ε͏Δ࣌ؒ৘ใදݱ Λਖ਼نදݱʹམͱ͠ࠐΉ • ରԠ͢Δղੳ༻ͷؔ਺ͱඥ෇͚Δ 3. நग़͞Εͨ࣌ؒ৘ใදݱΛղੳ • நग़ͨ࣌ؒ͠৘ใදݱͷจࣈྻΛ TIMEX3λάͷ࢓༷ʹ߹Θͤͯղੳ͠ɹ ग़ྗ͢Δ
  6. 15 ͲͷΑ͏ʹϧʔϧΛهड़͍ͯ͠Δ͔ʁ 1. ࣌ؒ৘ใදݱͷ਺஋෦෼ΛάϧʔϓԽ • ਖ਼نදݱͰͷ਺஋දݱͷऔಘͱՄಡੑ ୲อͷͨΊʹɺ໊લ෇͖άϧʔϓΛ ࡞͓ͬͯ͘ 2. ࣌ؒ৘ใදݱͷύλʔϯΛਖ਼نදݱͰߏங

    • จࣈྻ্Ͱදݱ͞Ε͏Δ࣌ؒ৘ใදݱ Λਖ਼نදݱʹམͱ͠ࠐΉ • ରԠ͢Δղੳ༻ͷؔ਺ͱඥ෇͚Δ 3. நग़͞Εͨ࣌ؒ৘ใදݱΛղੳ • நग़ͨ࣌ؒ͠৘ใදݱͷจࣈྻΛ TIMEX3λάͷ࢓༷ʹ߹Θͤͯղੳ͠ɹ ग़ྗ͢Δ
  7. 16 ͲͷΑ͏ʹϧʔϧΛهड़͍ͯ͠Δ͔ʁ 1. ࣌ؒ৘ใදݱͷ਺஋෦෼ΛάϧʔϓԽ • ਖ਼نදݱͰͷ਺஋දݱͷऔಘͱՄಡੑ ୲อͷͨΊʹɺ໊લ෇͖άϧʔϓΛ ࡞͓ͬͯ͘ 2. ࣌ؒ৘ใදݱͷύλʔϯΛਖ਼نදݱͰߏங

    • จࣈྻ্Ͱදݱ͞Ε͏Δ࣌ؒ৘ใදݱ Λਖ਼نදݱʹམͱ͠ࠐΉ • ରԠ͢Δղੳ༻ͷؔ਺ͱඥ෇͚Δ 3. நग़͞Εͨ࣌ؒ৘ใදݱΛղੳ • நग़ͨ࣌ؒ͠৘ใදݱͷจࣈྻΛ TIMEX3λάͷ࢓༷ʹ߹Θͤͯղੳ͠ɹ ग़ྗ͢Δ
  8. 17 ࣌ؒ৘ใදݱΛѻ͏্Ͱͷͦͷଞͷػೳ • ΞϥϏΞ਺ࣈ/׽਺ࣈɺ੢ྐྵ/࿨ྐྵͳͲͷଟ࠼ͳϑΥʔϚοτʹରԠ • ׽਺ࣈˠΞϥϏΞ਺ࣈ΁ͷม׵͢Δ͜ͱͰ࣮ݱ • ࿨ྐྵͷදݱ΋ରԠ • Pythonͷ೔෇ܕ/࣌ؒܕ΁ͷม׵

    • datetime΍timedeltaܗࣜʹม׵͠ɺϓϩάϥϜ͔Βར༻Ͱ͖ΔΑ͏ʹ͢Δ • ج४࣌Λઃఆ͠ɺ૬ରతͳ࣌ؒ৘ใදݱͰͷิ׬͕Մೳ • ࠓ೔Λج४೔ͱͨ͠ͱ͖ɺʮ4/8ʯˠ ʮ2022-04-08ʯͷΑ͏ʹ೥͕ิ׬͞ΕΔ
  9. 18 ։ൃͷొΓํ (1) • طଘݚڀΛಡΈࠐΉ • ͱʹ͔͘࢓༷͕ෳࡶͰཧղ͕೉͍͠ • ͜ΕΛҰ͔ΒखಈΞϊςʔγϣϯ͠ΖͱݴΘΕΔͱɺ͔ͳΓ೉͍͠Μ͡Όͳ͍͔ͱࢥ͏ •

    ແݶʹٙ໰͕ग़ͯ͘ΔͷͰɺͳΜͱ͔໌จԽͰ͖Δܗʹམͱ͠ࠐΜͰ͍͘ • e.g. DATEͱTIMEͬͯԿ͕ҧ͏ͷʁ → 1೔ͱ͍͏ɺ஍ٿ্ͷपظతͳ࠷খ୯ҐͰ۠ผ • e.g.ʮே8࣌ʯͷʮேʯ͸ؚΊΔ΂͖ʁ → طଘݚڀͩͱؚΊ͍ͯΔͷͰϤγʂ • ·ͣ͸ߟ͑ΒΕΔදݱΛ۪௚ʹ࣮૷ • ςετۦಈ։ൃͰɺ࣌ؒ৘ใදݱͱ͋Δ΂͖ղੳ݁ՌΛେྔʹྻڍ͢Δ • ͦͷςετʹ௨ΔΑ͏ʹͻͨ͢Βຊମͷ࣮૷ΛਐΊΔ
  10. 19 ։ൃͷొΓํ (2) • Livedoor χϡʔείʔύεશจʹର࣮ͯ͠ߦͯ͠False PositiveΛ௵͍ͯ͘͠ɺΛ܁Γฦ͢ • ͜ΕͰ͸False Negative͸վળ͕೉͍͕͠ɺϧʔϧ

    ϕʔεͷ࣮૷Ͱ͸͋·Γى͜Βͳ͍ͱ൑அ • ੨ۭจݿͰ׽਺ࣈͷදݱΛ୳ͯ͠ɺςετʹ௥Ճ͍ͯ͘͠ • ׽਺ࣈˠΞϥϏΞ਺ࣈม׵ࣗମͷ࣮૷΋ฒߦ • ༧૝͠ͳ͍਺ࣈදݱ͕େྔʹݟ͔ͭΓࠔ࿭ • ׽਺ࣈ͔ΒΞϥϏΞ਺ࣈͷม׵ʹࠔΔྫ • ʮҰɺʓʓ࢛ઍࣣޒʓʯ • ʮࡾೋʓઍʓʓʓʯ ʮޒϱ೥ܭըͱιϰΣτಉໍͷจԽతඈ༂ʯ ٶຊඦ߹ࢠ ΑΓ
  11. 20 ja-timexͷݶքͱ՝୊ • จ຺Λߟྀͨ͠೔෇දݱͷநग़͕Ͱ͖ͳ͍ • False Positive: ʮੴͷ্ʹ΋ࡾ೥ʯʮҙؾࠐΈ͸े෼Ͱ͢ʯʮ12/5 ͸ 2.4Ͱ͢ʯ

    • False Negative: DD/MM/YYYYͳͲͷඇ೔ຊޠݍͷ೔෇දه • TIMEX3࢓্༷ͷݶք • ͋Δಛఆͷ1೔Λද͍ͯ͠Δͷʹ΋ؔΘΒͣɺෳ਺ʹ෼ׂ͞Εͯ͠·͏ • ʮ4݄8೔༵ۚ೔ʯˠ <TIMEX3 “4݄8೔”> <TIMEX3 “༵ۚ೔”> • ᐆດ͕͋͞Δ೔෇දݱΛදݱ͖͠Εͳ͍ • ʮઌ݄ͷ೔༵೔ʯʮ4݄7,8೔ʯ • ਫ਼౓ΛͪΌΜͱܭଌͰ͖͍ͯͳ͍ • ධՁσʔλΛ࡞Δͷ͕໘౗ष͗͢Δͱ͍͏ͷͰɺ͜Ε͸ࢲͷଵຫͰ͢͝ΊΜͳ͍͞
  12. 22 ࣌ؒ৘ใදݱͷநग़͸ࠓޙͲ͏ͳ͍ͬͯ͘ͷ͔ʁ • લड़ͷ௨Γɺจ຺Λߟྀ͠ͳ͍ͱؒҧ͑Δྫ͕͋Δͷ͸͔֬ • ԿΒ͔จ຺Λߟྀͨࣗ͠વݴޠॲཧͷΞϓϩʔν͕ඞཁ • ৭ʑͳߟ͑ํ͕͋Γ͏Δ • ݫີͳऔಘʹ͸ਖ਼نදݱͰͷܾఆతͳॲཧ͕ඞਢͩΑ೿

    • ϧʔϧʹج͍࣮ͮͯ֬ʹಈ͘ͱ͍͏҆৺ײɺϛεΛଈ࠲ʹमਖ਼Ͱ͖ΔରԠྗ • TransformerͳͲͷDNNʹΑΔݻ༗දݱநग़͕༏੎ʹͳΔΑ೿ • จ຺Λߟྀͨ͠ݻ༗දݱநग़ͱͯ͠ɺநग़ثͷ෦෼Λ୅ସ͢Δ • ͦ΋ͦ΋End2EndͳγεςϜͰ͸࣌ؒ৘ใදݱΛऔಘ͢ΔͳΜͯ͜ͱ͸͠ͳ͍Α೿ • μΠϨΫτʹܭࢉػ͕ղऍՄೳͳ࣌ؒΛग़ྗ͢ΔɺQuestion Answeringతʹղ͘ɺͳͲ
  13. 23 ࣌ؒ৘ใදݱͷۙ೥ͷݚڀ • BERTΛ༻͍ͨݻ༗දݱநग़Ξϓϩʔν (2019) [5] • σʔληοτͷछྨʹΑͬͯɺBERTͱϧʔϧϕʔεͷ༏ྼ͕มΘΔ݁Ռ • seq2seqλεΫͱͯ͠ɺTIMEX3λάΛؚΉςΩετΛ௚઀ੜ੒͢ΔΞϓϩʔν

    (2021) [6] • ϧʔϧϕʔεͷ݁ՌΛWeak SupervisionʹΑΓFine-tuningͷσʔλʹར༻ • BERT౳ͷNER (Token Classification) ͱൺֱͯ͠ɺଟ͘ͷσʔληοτͰߴਫ਼౓ • Huggingfaceʹͯར༻ՄೳͳϞσϧ͕഑෍͞Ε͍ͯΔ • https://github.com/satya77/Transformer_Temporal_Tagger
  14. 24 ͦ΋ͦ΋࣌ؒ৘ใදݱ͸ࣗવݴޠॲཧʹͱͬͯԿͳͷ͔ʁ • ਺஋తͳଆ໘ • Ճࢉݮࢉͱ͍ͬͨૢ࡞͕Մೳͳɺ఺΍ͦͷൣғͱͯ͠ͷ਺஋දݱ • 1࣌ؒ + 30෼

    = 1.5࣌ؒ = 90෼ • ࡢ೔ͷ໌ޙ೔ = ໌ޙ೔ͷࡢ೔ = ໌೔ • ݴޠతͳଆ໘ • ਓؒੜ׆Λओ࣠ͱͨ͠ײ֮తදݱ • ேɾனɾ൩ɿओ؍తͳ1೔ͷ۠෼ͷදݱ • ઌ೔ɿ͋ΔҰఆظؒ಺ͷաڈͷ1೔Λࢦ͢දݱ • 25࣌ɿཌ1࣌ΛಉҰͷ೔ͷԆ௕ઢ্Ͱ͋Δଊ͑ͨͱ͖ͷදݱ • Ұສ೥ͱೋઍ೥લɿ͸Δ͔ੲͰ͋Δ͜ͱΛද͢ތுදݱ
  15. 25 ࣌ؒ৘ใදݱ͸ ৗࣝ Commonsense ͱߟ͑ΒΕΔ • Numerical Commonsense [7] •

    ࣌ؒ৘ใදݱ͸ɺ਺஋ʹؔ͢ΔৗࣝͷҰ෦ • ۩ମతͳλεΫ • ؚҙؔ܎ೝࣝ (Recognizing Textual Entailment: RTE) • ৗࣝਪ࿦ (Commonsense Reasoning) • ϚεΫ͞Εͨ୯ޠͷ༧ଌ (Masked-word-prediction) • σʔληοτ • NumerSense [8] • Ұ෦ʹӈਤͷΑ͏ͳ࣌ؒʹؔ͢Δ࣭໰͕͋Δ ※ ja-timexͰ͸೥ྸ͸࣌ؒ৘ใදݱͱͯ͠ ͸ѻ͍ͬͯͳ͍ͷͰɺݫີʹ͸͜Ε͸ɹ ࣌ؒ৘ใදݱʹؔ͢ΔλεΫͰ͸ͳ͍ https://inklab.usc.edu/NumerSense/
  16. 26 ݱࡏͷେن໛ݴޠϞσϧʹΑΔจੜ੒͸࣌ؒΛͲ͏ଊ͍͑ͯΔ͔ʁ • Tnrasformerϕʔεͷେن໛ݴޠϞσϧʹΑͬͯɺੜ੒͞ΕΔจॻͷࣗવ͞͸֨ஈʹ্͕ͬͨ • ҰํͰɺจͱͯ͠ͷهड़಺༰ͷҰ؏ੑ΍ৗࣝͷѻ͍͸·ͩ·ͩ • ࢼ͠ʹGPT-3Ͱ࣌ؒʹؔ͢ΔςΩετΛੜ੒ͯ͠ΈΔ • ʮ9࣌ͷ1࣌ؒޙ͸10࣌Ͱ͢ɻ9࣌ͷ2࣌ؒޙ͸ʯʹଓ͘จΛੜ੒ͤ͞Δ

    • Ϟσϧ͸rinna/japanese-gpt-1bɺύϥϝʔλ͸do_sample=True, top_p=0.95 9࣌ͷ1࣌ؒޙ͸10࣌Ͱ͢ɻ9࣌ͷ2࣌ؒޙ͸1࣌Ͱ͢ɻ͜͜Ͱݴ͏1࣌ͱ͸9࣌00෼ͷ͜ͱΛࢦ͠·͢ɻͦͯ͠ 9࣌ͷ1࣌ؒޙ͸10࣌Ͱ͢ɻ9࣌ͷ2࣌ؒޙ͸ɺ10͔࣌ͩΒɺԿ΋ແ͠Ͱྑ͍ɻԿ΋ແ͍͔Βग़དྷΔΜͰ͢ɻͦΕʹࠓɺ 9࣌ͷ1࣌ؒޙ͸10࣌Ͱ͢ɻ9࣌ͷ2࣌ؒޙ͸ɺ͓னͷ12࣌Ͱ͢ɻ2͔࣌Β6࣌·Ͱ͕ɺࢲֶ͕ͨͪߍʹ͍Δ࣌ؒͰ͢ 9࣌ͷ1࣌ؒޙ͸10࣌Ͱ͢ɻ9࣌ͷ2࣌ؒޙ͸11࣌Ͱ͢ɻ18࣌ͷ1࣌ؒޙ͸20࣌Ͱ͢ɻ͜ΕΒͷ1೔ͷதʹ͸ 9࣌ͷ1࣌ؒޙ͸10࣌Ͱ͢ɻ9࣌ͷ2࣌ؒޙ͸18࣌Ͱ͢ɻͳͷͰɺͦΜͳʹ͋ͬͱ͍͏ؒʹ͸ಈ͖·ͤΜɻ͔ͩΒ஗͔Ε 9࣌ͷ1࣌ؒޙ͸10࣌Ͱ͢ɻ9࣌ͷ2࣌ؒޙ͸10࣌ͷ5෼ޙͰ͢ɻͭ·Γ10࣌͸ʮ໿5࣌50෼ʯ 9࣌ͷ1࣌ؒޙ͸10࣌Ͱ͢ɻ9࣌ͷ2࣌ؒޙ͸11࣌ɺ12࣌͸͓னɻ25࣌24෼͸2࣌ؒޙɺͦͯ͠ɺ౔༵೔ͷ 9࣌ͷ1࣌ؒޙ͸10࣌Ͱ͢ɻ9࣌ͷ2࣌ؒޙ͸11࣌Ͱ͢ɻ9͔࣌Β11࣌·Ͱͷ2࣌ؒͷฏۉͷ࣌ܭͷ਑ͱͷ࣌ؒ
  17. 27 ͋ΒͨΊͯ࣌ؒ৘ใදݱநग़ͷ͜Ε͔Β • ϧʔϧϕʔεղੳث͸ࠓޙ΋ར༻͞Εଓ͚ΔͩΖ͏ • ࢖͍΍͢͞΍ϝϯςφϯεͷ༰қ͞ͱ͍͏ར༻࣌ͷ؍఺͔Β • Weak SupervisionͰͷֶशσʔλ࡞੒ͱ͍͏ɺΑΓߴਫ਼౓ͳख๏ʹܨ͕Δେ͖ͳ໾໨ •

    จ຺ΛߟྀͰ͖ΔTransformerϕʔεͷݻ༗දݱநग़͸ݱ࣮త • ӳޠΛର৅ͱͨ͠ݚڀͰ͸ϧʔϧϕʔεͱൺֱͯ͠΋ߴਫ਼౓ • ja-timexͷ݁ՌΛ࢖ͬͨWeak SupervisionతΞϓϩʔνͰja-timexͷڧԽ൛Λ࡞ͬͯΈ͍ͨ • ධՁ༻ͷ೔ຊޠσʔληοτ΋࡞ΒͶ͹…… • େن໛ݴޠϞσϧࣗମ͕࣌ؒ৘ใදݱΛਖ਼͘͠ѻ͑Δͷ͸·ͩ·ͩઌ͔΋ • ςΩετੜ੒΍ਪ࿦ϨϕϧͰ͖ͪΜͱͨ݁͠Ռ͕ฦͬͯ͘Δͷ͸ظ଴Ͱ͖ͳ͍ (?) • End2Endతʹ࣌ؒ৘ใΛ͍͍ײ͡ʹѻͬͯ͘ΕΔͷ͸ɺ·ͩ·ͩઌ͔΋
  18. 28 ·ͱΊ • ja-timexͷ঺հ • ࣌ؒ৘ใදݱΛ؆୯ʹநग़/ن֨ԽͰ͖Δϧʔϧϕʔεͷղੳث • େྔͷਖ਼نදݱΛهड़͢Δ͜ͱͰ࣮ݱ͍ͯ͠Δ • จ຺Λߟྀͨ࣌ؒ͠৘ใදݱͷநग़ʹ͸ɺϧʔϧϕʔεͰ͸ݶք͕͋Δ

    • ϧʔϧϕʔεղੳثͷ͜Ε͔Β • େن໛ݴޠϞσϧʹΑΔվྑ͕ൃද͞Ε࢝Ίͨ • ϧʔϧϕʔεղੳثͷ݁ՌΛֶशσʔλʹར༻͢ΔྲྀΕ • ࠓޙTransformerϕʔεͷݻ༗දݱநग़ثʹஔ͖׵Θ͍͖ͬͯͦ͏ • ࣌ؒ৘ใදݱ͸ɺৗࣝΛѻ͏λεΫͷҰ෦ͱଊ͑Δ͜ͱ͕Ͱ͖Δ • େن໛ݴޠϞσϧ͕࣌ؒ৘ใදݱΛݩʹਖ਼͘͠ਪ࿦͢Δͷ͸·ͩ·ͩ೉ͦ͠͏
  19. 29 [1] খ੢ޫ, ઙݪਖ਼޾, & લ઒تٱ༤. (2013). ʰݱ୅೔ຊޠॻ͖ݴ༿ۉߧίʔύεʱ ʹର͢Δ࣌ؒ৘ใΞϊςʔγϣϯ. ࣗવݴޠॲཧ,

    20(2), 201-221. [2] Sun, W., Rumshisky, A., & Uzuner, O. (2013). Temporal reasoning over clinical text: the state of the art. Journal of the American Medical Informatics Association, 20(5), 814-819. [3] Alfattni, G., Peek, N., & Nenadic, G. (2020). Extraction of temporal relations from clinical free text: A systematic review of current approaches. Journal of Biomedical Informatics, 108, 103488. [4] ੒ᖒࠀຑ (2014)ʮࣗવݴޠॲཧʹ͓͚Δ਺ྔදݱͷऔΓѻ͍ʯ౦๺େֶେֶӃ म࢜࿦จ [5] Chen, S., Wang, G., & Karlsson, B. (2019). Exploring word representations on time expression recognition. Technical report, Microsoft Research Asia. [6] Almasian, S., Aumiller, D., & Gertz, M. (2021). BERT got a Date: Introducing Transformers to Temporal Tagging. arXiv preprint arXiv:2109.14927. [7] Narisawa, K., Watanabe, Y., Mizuno, J., Okazaki, N., & Inui, K. (2013, August). Is a 204 cm man tall or small? acquisition of numerical common sense from the web. In Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) (pp. 382-391). [8] Lin, B. Y., Lee, S., Khanna, R., & Ren, X. (2020). Birds have four legs?! numersense: Probing numerical commonsense knowledge of pre-trained language models. Proceedings of EMNLP Reference