SLIDE 1
Ryan McDonald
Google NLU team Google Linguistics team Universal Dependencies Group
META-Forum 2016
Europe & the Internet
Language of top 10M pages (W3Tech 2015)
14% 32% 54%
~600M internet users
(~80% internet penetration)
- ther
Multilingual Europe As A Challenge for Language Technology Ryan - - PowerPoint PPT Presentation
Multilingual Europe As A Challenge for Language Technology Ryan - - PowerPoint PPT Presentation
Multilingual Europe As A Challenge for Language Technology Ryan McDonald Google NLU team Google Linguistics team Universal Dependencies Group Europe & the Internet 54% 32% 14% other ~600M internet users (~80% internet penetration)
SLIDE 2
SLIDE 3
META-Forum 2016
Google: Mobile vs. Desktop
SLIDE 4
META-Forum 2016
Search & Language Technology
Ten links
SLIDE 5
META-Forum 2016
Search & Language Technology
SLIDE 6
META-Forum 2016
Mobile & Language Technology
Generation & Text-to-speech Speech recognition Natural Language Understanding
SLIDE 7
META-Forum 2016
Mobile & Language Technology
Order a pizza Get weather Q & A Predictive info
SLIDE 8
META-Forum 2016
Mobile & Language Technology
Mobile is the future Language technologies key to mobile experience Users demand native language support Europe: large market dozens of languages
SLIDE 9
META-Forum 2016
Language Technology Productionization
NLU system
English version Internationalize
SLIDE 10
META-Forum 2016
Baked-in multilingualism in NLU
NLU system
Multilingual systems
SLIDE 11
META-Forum 2016
End-to-end NN
Pros Simple & flexible Multilingual by nature? High accuracy** Cons Hard to interpret Need a lot of data**
SLIDE 12
META-Forum 2016
Similarity / Paraphrase / NLI
Parikh et al. 2016
Multilingual?? Units/words? Word order? Structural bias?
SLIDE 13
META-Forum 2016
Intermediate Representations
Useful (discrete) abstractions for i18n NLU?
NLU Abstraction layer
Ο πληθυσμός του Καναδά είναι 35,000,000 (The population of Canada is 35M)
SLIDE 14
NLU focused awards, May 2016
Morphosyntax
Ο Ιωάννης είδε τους γονείς του, όταν πήγε στην Αθήνα.
John saw family his when went to Athens
Ήταν ευτυχής να τον δουν.
were happy to him see δουν:
number: plural person: third tense: subjunctive/future
Ήταν:
number: ? person: third tense: past
πήγε:
number: singular person: third tense: past
Missing pronoun realization
- 1. Syntax to identify relevant verbs
- 2. Morphology to piece together
gender/number
NLP / Analysis
SLIDE 15
NLU focused awards, May 2016
Morphosyntax
Generation / TTS
population(Καναδάς, 35.000.000) Ο πληθυσμός <masc-gen-sing-prep> <masc-gen-sing-Entity> <sing-cop> <pop>
+
nom-fem-plur
Ο πληθυσμός του Καναδά είναι 35.000.000 = Ο πληθυσμός του Καναδά είναι τριάντα πέντε εκατομμύρια =
SLIDE 16
META-Forum 2016
Universal Dependencies (UD)
(Nivre et al. 2016)
✤ Content-head reigns supreme for dependencies ✤ UPOS + Morphology + lemma surface analysis
~26 European languages covered in v1.3
SLIDE 17
META-Forum 2016
Analysis: SyntaxNet
(Andor et al. 2016)
UD
Chen & Manning 14 Weiss et al. 15
SLIDE 18
META-Forum 2016
Morphosyntactic Analysis @ Google
SLIDE 19
META-Forum 2016
Generation: Entity Lexicons
Καναδάς: Gender=Masc, Case=Nom, Number=Sing Καναδά: Gender=Masc, Case=Acc, Number=Sing Καναδά: Gender=Masc, Case=Gen, Number=Sing
Ο Καναδάς έχει πολλά δέντρα.
Gender=Masc Case=Nom Number=Sing
Ο πληθυσμός του Καναδά είναι 35M.
Gender=Masc Case=Gen Number=Sing
Είδα τον Καναδά με το τρένο.
Gender=Masc Case=Acc Number=Sing Inflectional table Edit-distance Unsup-morpher etc.
SLIDE 20
META-Forum 2016
Summary
NLU + = Multilingualism in NLU from ground-up End-to-end, Morphosyntax, more?
SLIDE 21