Constraint Grammar As a Hand-Crafted Transformer

Anssi Yli-Jyrä

Constraint Grammar As a Hand-Crafted Transformer

Julkaisun otsikon käännös: Rajoitekielioppi käsintehtynä Transformer-verkkona

Tutkimustuotos: Konferenssimateriaalit › Konferenssiesitys › vertaisarvioitu

Abstrakti

The differences between the rule-based NLP such as CG and the deep neural networks, such as the Transformer (Vaswani et al., 2017) are so striking that it is really hard to see any relevant conceptual links between them. However, this paper sketches a thought experiment that assumes an equivalent input-output behaviour by both systems and aligns certain structural aspects of the computation behind a practical Constraint Grammar with the computation structure of Transformer. Based on this scene, several findings are presented that state some functional similarities in the computation graphs of the systems.

Julkaisun otsikon käännös	Rajoitekielioppi käsintehtynä Transformer-verkkona
Alkuperäiskieli	englanti
Tila	Julkaistu - 2019
OKM-julkaisutyyppi	Ei sovellu
Tapahtuma	NoDaLiDa 2019 workshop on Constraint Grammar - Methods, Tools, and Applications - University of Turku, Turku, Suomi Kesto: 30 syysk. 2019 → 30 syysk. 2019 https://visl.sdu.dk/nodalida2019.html

Työpaja

Työpaja	NoDaLiDa 2019 workshop on Constraint Grammar - Methods, Tools, and Applications
Lyhennettä	CG 2019
Maa/Alue	Suomi
Kaupunki	Turku
Ajanjakso	30/09/2019 → 30/09/2019
www-osoite	https://visl.sdu.dk/nodalida2019.html

Tieteenalat

113 Tietojenkäsittely- ja informaatiotieteet
6121 Kielitieteet

Pääsy asiakirjaan

https://visl.sdu.dk/pdf/CG-workshop2019_paper_9.pdfLisenssi: CC BY

1 Päättynyt

ADEQSYNTAX: A Usable Finite-State Model for Adequate Syntactic Complexity
Yli-Jyrä, A. (Projektinjohtaja)
Suomen tietokirjailijat
01/09/2013 → 30/04/2019
Projekti: Tutkimusprojekti

Siteeraa tätä

@conference{a9472e075e334f82a71f215b27898b0b,

title = "Constraint Grammar As a Hand-Crafted Transformer",

abstract = "The differences between the rule-based NLP such as CG and the deep neural networks, such as the Transformer (Vaswani et al., 2017) are so striking that it is really hard to see any relevant conceptual links between them. However, this paper sketches a thought experiment that assumes an equivalent input-output behaviour by both systems and aligns certain structural aspects of the computation behind a practical Constraint Grammar with the computation structure of Transformer. Based on this scene, several findings are presented that state some functional similarities in the computation graphs of the systems.",

keywords = "113 Computer and information sciences, 6121 Languages",

author = "Anssi Yli-Jyr{\"a}",

year = "2019",

language = "English",

note = "NoDaLiDa 2019 workshop on Constraint Grammar - Methods, Tools, and Applications ; Conference date: 30-09-2019 Through 30-09-2019",

url = "https://visl.sdu.dk/nodalida2019.html",

}

TY - CONF

T1 - Constraint Grammar As a Hand-Crafted Transformer

AU - Yli-Jyrä, Anssi

PY - 2019

Y1 - 2019

N2 - The differences between the rule-based NLP such as CG and the deep neural networks, such as the Transformer (Vaswani et al., 2017) are so striking that it is really hard to see any relevant conceptual links between them. However, this paper sketches a thought experiment that assumes an equivalent input-output behaviour by both systems and aligns certain structural aspects of the computation behind a practical Constraint Grammar with the computation structure of Transformer. Based on this scene, several findings are presented that state some functional similarities in the computation graphs of the systems.

AB - The differences between the rule-based NLP such as CG and the deep neural networks, such as the Transformer (Vaswani et al., 2017) are so striking that it is really hard to see any relevant conceptual links between them. However, this paper sketches a thought experiment that assumes an equivalent input-output behaviour by both systems and aligns certain structural aspects of the computation behind a practical Constraint Grammar with the computation structure of Transformer. Based on this scene, several findings are presented that state some functional similarities in the computation graphs of the systems.

KW - 113 Computer and information sciences

KW - 6121 Languages

M3 - Paper

T2 - NoDaLiDa 2019 workshop on Constraint Grammar - Methods, Tools, and Applications

Y2 - 30 September 2019 through 30 September 2019

ER -

Constraint Grammar As a Hand-Crafted Transformer

Abstrakti

Työpaja

Tieteenalat

Pääsy asiakirjaan

Projektit

ADEQSYNTAX: A Usable Finite-State Model for Adequate Syntactic Complexity

Siteeraa tätä