Semantic Parsing Via Staged Query Graph Generation

Semantic Parsing Via Staged Query Graph Generation

Scott Wen-tau Yih Who is Justin Bieber’s sister? Jazmyn Bieber semantic parsing Knowledge 휆푥. sister_of(justin_bieber, 푥) Base query matching 휆푥. sibling_of(justin_bieber, x) ∧ gender(x, female) Who is Justin Bieber’s sister? Jazmyn Bieber Knowledge semantic parsing Base query 휆푥. sibling_of(justin_bieber, x) ∧ gender(x, female) “What was the date that Minnesota became a state?” “When was the state Minnesota created?” “Minnesota's date it entered the union?” location.dated_location.date_founded Basic idea directly grows staged Addresses Key Challenges 52.5 • Introduction • Background • Graph knowledge base • Query graph 12/26/1999 • Mila Kunis from cvt1 • Meg Griffin • Family Guy cvt2 Lacey Chabert series cvt3 1/31/1999 constraints argmin Meg Griffin topic entity core inferential chain Family Guy cast y x • Introduction • Background • Staged Query Graph Generation (Our Approach) • Link topic entity • Identify core inferential chain • Augment constraints Staged Meg Family Guy (1) Link Topic Entity s1 Family Guy s0 ϕ s2 Meg Griffin Staged Meg Family Guy (2) Identify Core Inferential Chain s3 Family Guy cast y actor x s1 s4 Family Guy Family Guy writer y start x s5 Family Guy genre x Staged Meg Family Guy (3) Augment Constraints s3 Family Guy cast y actor x s6 Meg Griffin Family Guy cast y actor x s7 argmin Meg Griffin Family Guy y x s1 Family Guy s0 ϕ s2 Meg Griffin s3 Family Guy cast y actor x s1 Who first voiceds4 Meg on Family Guy? Family Guy Family Guy writer y start x {cast−actor, s5writer−start, genre} Family Guy genre x • Input is mapped to two 푘-dimensional vectors 300 exp cos(푦 , 푦 ) 푃 푅 푃 = 푅 푃 ′ 푅′ exp cos(푦푅 , 푦푃) 300 ... 푘 푘 ... 푦푃 ∈ R 푦푅 ∈ R max max max ... ... ... 1000 1000 ... 1000 15K 15K 15K ... 15K 15K who voiced meg on 푒 cast−actor <s> w1 w2 wT </s> • Who voiced Family Guy s3 Family Guy cast y actor x s3 cast FamilyGuyFamily Guy cast푦 yactoractor 푦 x푥 s6 • One or more constraint nodes can beMeg added Griffin to 푦 or 푥 • 푦 : Additional property of this event (e.g., character 푦 MegGriffin ) Family Guy cast y actor x • 푥 : Additional property of the answer entity (e.g., gender) s7 argmin Meg Griffin Family Guy y x Who first voiced Meg on Family Guy? s3 Family Guy cast y actor x s4 Family Guy writer y start x Who first voiced Meg on Family Guy? s7 argmin Meg Griffin s3 Family Guy cast y actor x Family Guy y x 푞 =Who first voiced Meg on Family Guy? 푠 = argmin Meg Griffin Family Guy cast y x • Introduction • Background • Staged Query Graph Generation (Our Approach) • Experiments • Data & evaluation metric • Creating training data from Q/A pairs • Results • What character did Natalie Portman play in Star Wars? Padme Amidala • What currency do you use in Costa Rica? Costa Rican colon • What did Obama study in school? political science • What do Michelle Obama do for a living? writer, lawyer • What killed Sammy Davis Jr? throat cancer [Examples from Berant] Relation Matching (Identifying Core Inferential Chain) Pattern Inferential Chain what was <e> known for people.person.profession what kind of government does <e> have location.country.form_of_government what year were the <e> established sports.sports_team.founded what city was <e> born in people.person.place_of_birth what did <e> die from people.deceased_person.cause_of_death who married <e> people.person.spouse_s people.marriage.spouse Reward Function 훾 Avg. F1 (Accuracy) on WebQuestions Test Set 60 52.5 50 44.3 45.3 41.3 39.2 39.9 40 37.5 35.7 33 30 20 10 0 Yao-14 Berant-13 Bao-14 Bordes-14b Berant-14 Yang-14 Yao-15 Wang-14 Yih-15 Method #Entities Covered Ques. Labeled Ent. Freebase API 19,485 98.8% 81.2% Yang & Chang, ACL-15 9,147 99.8% 87.8% 52.5% 48.4% 49.6 52.5 A random sample of 100 incorrectly answered questions directly http://aka.ms/sent2vec http://aka.ms/codalab-webq http://aka.ms/stagg.

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    32 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us