Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
文献紹介:参照訳を必要としない単語分散表現による異言語間類似度を用いた訳文の自動評価
Search
Taichi Aida
June 12, 2019
Technology
160
0
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
文献紹介:参照訳を必要としない単語分散表現による異言語間類似度を用いた訳文の自動評価
Taichi Aida
June 12, 2019
More Decks by Taichi Aida
See All by Taichi Aida
意味を表すベクトル表現を用いたテキスト分析
a1da4
0
140
スウェーデン滞在報告
a1da4
0
35
PhD Defence: Considering Temporal and Contextual Information for Lexical Semantic Change Detection
a1da4
1
310
文献紹介:A Multidimensional Framework for Evaluating Lexical Semantic Change with Social Science Applications
a1da4
1
400
YANS2024:目指せ国際会議!「ネットワーキングの極意(国際会議編)」
a1da4
0
330
言語処理学会30周年記念事業留学支援交流会@YANS2024:「学生のための短期留学」
a1da4
1
460
新入生向けチュートリアル:文献のサーベイv2
a1da4
18
12k
文献紹介:Isotropic Representation Can Improve Zero-Shot Cross-Lingual Transfer on Multilingual Language Models
a1da4
0
250
文献紹介:WhitenedCSE: Whitening-based Contrastive Learning of Sentence Embeddings
a1da4
1
400
Other Decks in Technology
See All in Technology
Keeping applications secure by evolving OAuth 2.0 and OpenID Connect
ahus1
PRO
1
160
誤解だらけの開発生産性 / Myths and Misconceptions about Developer Productivity
i35_267
1
180
Kaggleで成長するために意識したこと
prgckwb
1
220
オブザーバビリティ、本当に活用できてる? 〜API連携×生成AIで成熟度を自動評価〜
dmmsre
1
2.9k
Text-to-SQLをAgentCoreで実現し、生成されるSQLの精度を定量的に評価する
yakumo
2
810
Baseline対応のDOMの型定義を作った
uhyo
3
730
Control Planeで育てるBtoB SaaSの認証基盤 - SRE NEXT 2026
pokohide
1
2.2k
Gen3R: 3D Scene Generation Meets Feed-Forward Reconstruction
spatial_ai_network
0
110
「最後に責任を取るのはチーム」— 人間のPRレビューを最小化してアップデートしたメンタルモデル
jnishime_dresscode
0
340
プロダクトだけじゃない、社内プロセスにおける自動化・省力化ノススメ
kakehashi
PRO
1
3.5k
LLM/Agent評価:トップ営業の発言を「正解」にする 〜暗黙的正解による評価を営業資産に変える〜
takkuhiro
1
190
プライバシー保護の理論と実践
lycorptech_jp
PRO
1
250
Featured
See All Featured
Evolution of real-time – Irina Nazarova, EuRuKo, 2024
irinanazarova
9
1.4k
Building Experiences: Design Systems, User Experience, and Full Site Editing
marktimemedia
0
550
What does AI have to do with Human Rights?
axbom
PRO
1
2.2k
Between Models and Reality
mayunak
4
370
The Cult of Friendly URLs
andyhume
79
6.9k
Designing for humans not robots
tammielis
254
26k
Technical Leadership for Architectural Decision Making
baasie
3
440
Save Time (by Creating Custom Rails Generators)
garrettdimon
PRO
32
3.7k
How GitHub (no longer) Works
holman
316
150k
The Organizational Zoo: Understanding Human Behavior Agility Through Metaphoric Constructive Conversations (based on the works of Arthur Shelley, Ph.D)
kimpetersen
PRO
0
380
Let's Do A Bunch of Simple Stuff to Make Websites Faster
chriscoyier
508
140k
Principles of Awesome APIs and How to Build Them.
keavy
128
18k
Transcript
จݙհʢʣ ࢀর༁Λඞཁͱ͠ͳ͍୯ޠࢄදݱʹΑΔ ҟݴޠؒྨࣅΛ༻͍ͨ༁จͷࣗಈධՁ ૬ాɹଠҰ Ԭٕज़Պֶେֶ ࣗવݴޠॲཧݚڀࣨ
LITERATURE ➤ ౻, ӽલ୩, ߥ. ࢀর༁Λඞཁͱ͠ͳ͍୯ޠࢄදݱʹΑΔҟݴޠؒྨࣅΛ༻͍ ͨ༁จͷࣗಈධՁ. ిࢠใ௨৴ֶձ. 2018.
ABSTRACT ➤ ࢀর༁Λ༻͍ͨ༁ͷධՁख๏͕ଘࡏ ➤ Ϣʔβ͕ػց༁Λར༻͢Δࡍࢀর༁Λ༻͍ͳ͍ ➤ QEɿࢀর༁ͷΘΓʹେنͳର༁ίʔύεΛ༻͍Δ ➤ ԤभҎ֎ͷݴޠͰର༁ίʔύε͕͍͠ ➤
ࢀর༁ɺର༁ίʔύεΛ༻͍ͳ͍ධՁํ๏ΛఏҊ
INTRODUCTION ➤ ݱঢ়ͷػց༁ඞͣਖ਼͍͠༁Λग़ྗ͢ΔͱݶΒͳ͍ ➤ খઆͳͲͷந͕ߴ͍จॻͰਖ਼͘͠༁ͤͳ͍ࣄ͕ଟ͍ ➤ ࢀর༁ɺର༁ίʔύεΛ༻͍ͯग़ྗΛධՁ ➤ ࢀর༁ɿ༁จͷ࣭ྔʹґଘ ➤
ର༁ίʔύεɿେنͰ͋Δલఏ ➤ ୯ޠͷҟݴޠؒྨࣅ͔ΒධՁ͢Δख๏ΛఏҊ
PROPOSAL 1. ୯ޠࢄදݱΛֶश 2. ༁ߦྻͰϚοϐϯά 3. ҟݴޠؒྨࣅͷܭࢉ 4. ྨࣅͷग़ྗ
PROPOSALʼ ୯ޠࢄදݱΛֶश ➤ WikipediaͷσʔλͰֶश ➤ ӳޠɿ1.3GB ➤ ຊޠɿ850MB ➤ ࣍ͷurl͔Βμϯϩʔυ
https://dumps.wikimedia.org ➤ ݴޠ͝ͱʹهࣄͷ͕ҟͳΔ ➤ ӳޠ൛ͷํ͕ଟ͔ͬͨ
PROPOSALʼ ༁ߦྻͰϚοϐϯά ➤ ಘΒΕͨࢄදݱΛҟݴޠؒͰൺֱ͍ͨ͠ ➤ Word2VecͰɺҟݴޠؒʹ͓͍ͯ୯ޠؒͷ͕ؔྨࣅ ➤ ϕΫτϧҟͳΔͨΊɺྨࣅܭࢉͰ͖ͳ͍ ➤ ઢܗมɿ༁ߦྻW
➤ ୯ޠϖΞ(xi , zi )Λ࠷খೋ๏Ͱۙࣅ͢Δ
PROPOSALʼ ҟݴޠؒྨࣅͷܭࢉ1 ➤ ΞϥΠϝϯτ ➤ จؒͷྨࣅΛܭࢉ͢Δࡍɺෆཁͳ୯ޠؒͷܭࢉϊΠζ ➤ શͯͷ୯ޠؒͰΞϥΠϝϯτείΞΛܭࢉ ➤ ҎԼͷΑ͏ʹͯ͠୯ޠϕΫτϧؒͷίαΠϯྨࣅ
di Λઃఆ ➤ DICEʢଜΒͷॏΈ͖DICEΛ࠾༻ʣ ➤ ୯ޠؒͷڞىใ f ͔Βܭࢉ tɿᮢʢҙʣ
PROPOSALʼ ҟݴޠؒྨࣅͷܭࢉ2 ➤ EMDɿEarth Mover’s Distance ➤ ྨࣅը૾ݕࡧʹ༻͍ΒΕΔख๏ ➤ ؒͷڑΛ࠷దԽ͢Δࡍɺ༌ૹΛͱʹఆٛ
➤ ֤P , QͦΕͧΕಛྔͱॏΈ͔ΒͳΔγάωνϟͷू߹ ➤ pi ͔Βqj ʹ༌ૹ͢Δ߹ ➤ dij ɿ2ؒͷڑ ➤ fij ɿ༌ૹ͢Δՙྔ ➤ ࣄྔWORKΛ࠷খԽ Ҿ༻ɿ[1]
PROPOSALʼ ҟݴޠؒྨࣅͷܭࢉ3 ➤ EMDɿEarth Mover’s Distance ➤ ҎԼ4ͭͷ੍݅ ➤ ಘΒΕͨ࠷దղ
f*ij Λ༻͍ͯɺ P , QؒͷڑΛܭࢉ Ҿ༻ɿ[1]
PROPOSALʼ ྨࣅͷग़ྗ ➤ ࠓճͷ݅ʹ͓͍ͯɺdijɿҟݴޠؒͷ୯ޠͷྨࣅ ➤ ಘΒΕͨEMDΑΓɺจؒͷྨࣅҎԼͷΑ͏ʹදͤΔ
EXPERIMENT ➤ σʔλ ➤ ʮThe Old Capitalʯ ➤ ݪจɿ߁ʮݹʯ ➤
಄100จΛ༻ ➤ ༁ʢӳˠʣ ➤ Google༁ ➤ Microsoft Translator ➤ Excite༁ ➤ ༁จͷධՁʢਓखʣ ➤ 8ਓͷཧܥେֶӃੜ ➤ 5ஈ֊ͰධՁ ➤ શһͷฏۉΛ༁จͷͱͨ͠
EXPERIMENT ➤ σʔλ ➤ ʮThe Old Capitalʯ ➤ ݪจɿ߁ʮݹʯ ➤
಄100จΛ༻ ➤ ༁ʢӳˠʣ ➤ Google༁ ➤ Microsoft Translator ➤ Excite༁ ➤ ༁จͷධՁʢਓखʣ ➤ 8ਓͷཧܥେֶӃੜ ➤ 5ஈ֊ͰධՁ ➤ શһͷฏۉΛ༁จͷͱͨ͠ ਓखධՁͷॱҐ༁ 1→༁ 2→༁ 3
EXPERIMENT ➤ ධՁํ๏ ɹ3ͭͷ༁จʹରͯ͠ɺྨࣅॱʹॱҐΛ࡞ ×100จ ➤ ਖ਼ղʢਓखͱͷશҰகʣ ➤ έϯυʔϧͷॱҐ૬ؔ ➤
༁จͷॱҐͷେখؔͷ૬ؔ ➤ ൺֱର ➤ ఏҊख๏ ➤ ॏΈʴΞϥΠϝϯτ ➤ ॏΈͳ͠ ➤ ΞϥΠϝϯτͳ͠ ➤ ࣗಈධՁई ➤ METEOR ➤ RIBES
RESULT ➤ ఏҊख๏METEORΛ্ճΓɺRIBES ʹഭΔ݁Ռ ➤ ୯ޠͷΞϥΠϝϯτ͕େ͖͘࡞༻͍ͯͨ͠
DISCUSSION ➤ ఏҊख๏ͷᮢʹର͢ΔҰகɺॱҐ૬ؔ ➤ ᮢΛ্͛Δ΄ͲҰக্͕Δ ➤ ᮢ0.73, 1.46ͰϐʔΫΛܴ͑Δ͕ɺޙऀҰக͕গͳ͍ ➤ ᮢ͕ߴ͘ͳΔ΄Ͳܭࢉʹඞཁͳ୯ޠϖΞഉআ͞ΕΔ
DISCUSSION ➤ RIBESਓखͱࣅͨॱͰධՁ͍ͯ͠Δ͕ɺఏҊख๏ҟͳΔ ➤ ఏҊख๏༁ௐͷ༁จΛߴ͘ධՁ͕ͪ͠
CONCLUSION ➤ ࢀর༁ର༁ίʔύε༻͍ͳ͍ධՁख๏ΛఏҊ ➤ ҟݴޠؒͷࢄදݱΛઢܗม͢Δ༁ߦྻ ➤ ྨࣅը૾ݕࡧʹ༻͍ΒΕΔEMDͰจؒྨࣅΛࢉग़ ➤ METEORΛ্ճΓɺRIBESʹഭΔ݁Ռͱͳͬͨ ➤
ఏҊख๏ͷੑೳʹ୯ޠͷΞϥΠϝϯτ͕ޮ͍͍ͯͨ ➤ ࢺใΛߟྀ͠ɺ୯ޠͷΞϥΠϝϯτΛվળ͍ͨ͠
REFERENCES 1. aidiary. Earth Mover’s Distance(EMD). ਓೳʹؔ͢Δஅ. 2012. http://aidiary.hatenablog.com/entry/20120804/1344058475