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
SNLP2019
Search
Ayana Niwa
September 25, 2019
Research
1
540
SNLP2019
第11回最先端NLP勉強会 発表資料
Ayana Niwa
September 25, 2019
Tweet
Share
More Decks by Ayana Niwa
See All by Ayana Niwa
A Quick Overview to Unlock the Potential of LLMs through Prompt Engineering
ayaniwa
0
160
Learning To Retrieve Prompts for In-Context Learning
ayaniwa
0
1.1k
UnNatural Language Inference
ayaniwa
0
420
Trends in Natural Language Processing at NeurIPS 2019.
ayaniwa
8
4.4k
Other Decks in Research
See All in Research
[輪講] SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features
nk35jk
2
1k
引力・斥力を制御可能なランダム部分集合の確率分布
wasyro
0
240
SSII2025 [SS1] レンズレスカメラ
ssii
PRO
2
1.1k
診断前の病歴テキストを対象としたLLMによるエンティティリンキング精度検証
hagino3000
1
130
AIグラフィックデザインの進化:断片から統合(One Piece)へ / From Fragment to One Piece: A Survey on AI-Driven Graphic Design
shunk031
0
450
Delta Airlines® Customer Care in the U.S.: How to Reach Them Now
bookingcomcustomersupportusa
0
110
cvpaper.challenge 10年の軌跡 / cvpaper.challenge a decade-long journey
gatheluck
3
310
EarthSynth: Generating Informative Earth Observation with Diffusion Models
satai
3
270
近似動的計画入門
mickey_kubo
4
1k
とあるSREの博士「過程」 / A Certain SRE’s Ph.D. Journey
yuukit
10
4.2k
言語モデルの地図:確率分布と情報幾何による類似性の可視化
shimosan
5
1.5k
カスタマーサクセスの視点からAWS Summitの展示を考える~製品開発で活用できる勘所~
masakiokuda
2
190
Featured
See All Featured
Helping Users Find Their Own Way: Creating Modern Search Experiences
danielanewman
29
2.9k
Distributed Sagas: A Protocol for Coordinating Microservices
caitiem20
333
22k
Refactoring Trust on Your Teams (GOTO; Chicago 2020)
rmw
35
3.1k
A better future with KSS
kneath
239
17k
A Modern Web Designer's Workflow
chriscoyier
696
190k
The MySQL Ecosystem @ GitHub 2015
samlambert
251
13k
Testing 201, or: Great Expectations
jmmastey
45
7.7k
Understanding Cognitive Biases in Performance Measurement
bluesmoon
29
1.9k
Navigating Team Friction
lara
189
15k
Docker and Python
trallard
46
3.6k
How to Ace a Technical Interview
jacobian
279
23k
Save Time (by Creating Custom Rails Generators)
garrettdimon
PRO
32
1.6k
Transcript
Probing for Semantic Classes: Diagnosing the Meaning Content of Word
Embeddings Yadollah Yaghoobzadeh, Katharina Kann, Timothy J. Hazen, Eneko Agirre, Hinrich Schutze ACL2019 11 NLP 2019/09/28 <
[email protected]
>
Outline 2 H9P=O8 O8 "!PC
$@K !.S%,'3 F5T&*'$:I P=O8$.S%,'PV ! R<72$PC H/+%)- >#"!.S O8! H9E69NR<72;BJ?Q1L >#".S(rare senses)P=O8APM0 " 4D('% UG
Background 3 # " !Word2VecGrove #"NLPIR* - ELMoBERT
$'+ ) %" full-space , % & L "$( " # &!
Background 4 # "@!Semantic class>*( S-class , "@=6:-&C?4=6
B $8 SEMCAT, HyperLex… =1;+=62%A → 0# .9/ - WIKI-PSE(WIKIpedia-based resource for Probing Semantics in word Embeddings) - "@=1;+sense embedding'7) “Lamb” Food3< Living-thing5< "@ !
Background 5 # Apple Apple Apple
+ Word embeddingsense embedding Arora et al. (2018) Arora et al. (2018) Word embedding sparse coding WIKI-PSEsense embedding Word embedding Sense embedding Sense embedding
WIKI-PSE Resource 6 Wikipedia .!2,$# .! ()
S-class 113 FIGER types%+134"10 person/authorperson/politician… à person -S-class *'.!/.! & - S-class0
WIKI-PSE Resource 7 "$% %2& +:! 343,000-6.-S-class#5'5000;-!% 75,.6,.83
+*1! /4 44,250- -S-class - S-class )0organization (9food @apple@ – food @apple@ – organization
8 Word embedding (word) @apple@ WIKI-PSE
word embedding Sense embedding @apple@ @apple@ - food @apple@ - organization @ word/S-class Uniform sum (unifΣ) !" Weighted sum (wghtΣ) !" # = % " !" #&" # Aggregated word embedding #&" Sense (word/S-class) embeddings word, unifΣ, wghtΣ
Experiments 9 Problem settings SkipGram Structured SkipGram
WIKI-PSE (LR) # (MLP k*)KNN) 1. word % word embedding 2. unifΣ sense embedding " 3. wghtΣ sense embedding ! $'& (" Word embedding
Experiments 10 Probing Task 1S-class Prediction S-class
@apple@ +food ∩ +organization ∩ -event S-class
Experiments 11 MLP > LRKNN 8&!S-class ."7;5
KNN )#, *:3, <(3-/ unifΣ > word > wghtΣ 'rare sense 0$ 97< à Rare sense42%+6 unifΣ Probing Task 1S-class Prediction F11 4
Experiments 12 Probing Task 1S-class Prediction
unifΣ $ #rare sense (13,000) F1! "
Experiments 13 Probing Task 1S-class Prediction * #4(! Recall
) S-class& sense embedding+%3 Rare S-class/"12 %"0, .$- Recall Dominance S-class' Recall
Experiments 14 Probing Task 1S-class Prediction /&)5. !"$' ,(20-
3+1(#41( Personlocation,(20* S-class !"6 20% Recall Recall #4 ,(20 Typicality Recall
Experiments 15 " Probing Task 2Ambiguity
Prediction ! L2 SSKIP$" LR / KNN / MLP ! ! unifΣ > word > wghtΣ KNN à % FREQUENCY(Baseline) # &LR
NLP Application Experiments 16 wghtΣ > word >(=) unifΣ <--
Probing task #!! -the U.S. Attorney’s Office announced Friday → location Common S-classtime Rare S-class location Friday mountain unifΣ ( !+ entity mention MC ), CRMR $)& SUBJ %*' MRPC " S-class
Summary & Conclusion 17 * $2"(3 WIKI-PSE'. 1)/& 0+$2/&
1, 1)/&% # 0+$2- a a e Rare sense e harder NLP Rare sense !# e