A Game Interface to Study Semantic Grounding in Text-Based Models

Timothee Mickus; Mathieu Constant; Denis Paperno

doi:10.1109/CoG52621.2021.9619149

A Game Interface to Study Semantic Grounding in Text-Based Models

Timothee Mickus, Mathieu Constant, Denis Paperno

ILS - Language, logic and information

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

Abstract

Can language models learn grounded representations from text distribution alone? This question is both central and recurrent in natural language processing; authors generally agree that grounding requires more than textual distribution. We propose to experimentally test this claim: if any two words have different meanings and yet cannot be distinguished from distribution alone, then grounding is out of the reach of textbased models. To that end, we present early work on an online game for the collection of human judgments on the distributional similarity of word pairs in five languages. We further report early results of our data collection campaign.

Original language	English
Title of host publication	2021 IEEE Conference on Games (CoG)
Publisher	IEEE
Number of pages	5
ISBN (Electronic)	978-1-6654-3886-5
ISBN (Print)	978-1-6654-4608-2
DOIs	https://doi.org/10.1109/CoG52621.2021.9619149
Publication status	Published - 2021
Event	2021 IEEE Conference on Games - Duration: 17 Aug 2021 → 20 Aug 2021 https://ieee-cog.org/2021/

Conference

Conference	2021 IEEE Conference on Games
Abbreviated title	CoG
Period	17/08/21 → 20/08/21
Internet address	https://ieee-cog.org/2021/

Access to Document

10.1109/CoG52621.2021.9619149

paper_268Final published version, 376 KBLicence: Taverne

Cite this

@inproceedings{a086efeac467408f9893f51d8eb08be4,

title = "A Game Interface to Study Semantic Grounding in Text-Based Models",

abstract = "Can language models learn grounded representations from text distribution alone? This question is both central and recurrent in natural language processing; authors generally agree that grounding requires more than textual distribution. We propose to experimentally test this claim: if any two words have different meanings and yet cannot be distinguished from distribution alone, then grounding is out of the reach of textbased models. To that end, we present early work on an online game for the collection of human judgments on the distributional similarity of word pairs in five languages. We further report early results of our data collection campaign.",

author = "Timothee Mickus and Mathieu Constant and Denis Paperno",

year = "2021",

doi = "10.1109/CoG52621.2021.9619149",

language = "English",

isbn = "978-1-6654-4608-2",

booktitle = "2021 IEEE Conference on Games (CoG)",

publisher = "IEEE",

address = "United States",

note = "2021 IEEE Conference on Games , CoG ; Conference date: 17-08-2021 Through 20-08-2021",

url = "https://ieee-cog.org/2021/",

}

TY - GEN

T1 - A Game Interface to Study Semantic Grounding in Text-Based Models

AU - Mickus, Timothee

AU - Constant, Mathieu

AU - Paperno, Denis

PY - 2021

Y1 - 2021

N2 - Can language models learn grounded representations from text distribution alone? This question is both central and recurrent in natural language processing; authors generally agree that grounding requires more than textual distribution. We propose to experimentally test this claim: if any two words have different meanings and yet cannot be distinguished from distribution alone, then grounding is out of the reach of textbased models. To that end, we present early work on an online game for the collection of human judgments on the distributional similarity of word pairs in five languages. We further report early results of our data collection campaign.

AB - Can language models learn grounded representations from text distribution alone? This question is both central and recurrent in natural language processing; authors generally agree that grounding requires more than textual distribution. We propose to experimentally test this claim: if any two words have different meanings and yet cannot be distinguished from distribution alone, then grounding is out of the reach of textbased models. To that end, we present early work on an online game for the collection of human judgments on the distributional similarity of word pairs in five languages. We further report early results of our data collection campaign.

U2 - 10.1109/CoG52621.2021.9619149

DO - 10.1109/CoG52621.2021.9619149

M3 - Conference contribution

SN - 978-1-6654-4608-2

BT - 2021 IEEE Conference on Games (CoG)

PB - IEEE

T2 - 2021 IEEE Conference on Games

Y2 - 17 August 2021 through 20 August 2021

ER -

A Game Interface to Study Semantic Grounding in Text-Based Models

Abstract

Conference

Access to Document

Fingerprint

Cite this