What do language representations really represent?
Research output: Contribution to journal › Journal article › Research › peer-review
Standard
What do language representations really represent? / Bjerva, Johannes; Östling, Robert; Veiga, Maria Han; Tiedemann, Jörg; Augenstein, Isabelle.
In: Computational Linguistics, Vol. 45, No. 2, 2019, p. 381-389.Research output: Contribution to journal › Journal article › Research › peer-review
Harvard
APA
Vancouver
Author
Bibtex
}
RIS
TY - JOUR
T1 - What do language representations really represent?
AU - Bjerva, Johannes
AU - Östling, Robert
AU - Veiga, Maria Han
AU - Tiedemann, Jörg
AU - Augenstein, Isabelle
PY - 2019
Y1 - 2019
N2 - A neural language model trained on a text corpus can be used to induce distributed representations of words, such that similar words end up with similar representations. If the corpus is multilingual, the same model can be used to learn distributed representations of languages, such that similar languages end up with similar representations. We show that this holds even when the multilingual corpus has been translated into English, by picking up the faint signal left by the source languages. However, just as it is a thorny problem to separate semantic from syntactic similarity in word representations, it is not obvious what type of similarity is captured by language representations. We investigate correlations and causal relationships between language representations learned from translations on one hand, and genetic, geographical, and several levels of structural similarity between languages on the other. Of these, structural similarity is found to correlate most strongly with language representation similarity, whereas genetic relationships—a convenient benchmark used for evaluation in previous work—appears to be a confounding factor. Apart from implications about translation effects, we see this more generally as a case where NLP and linguistic typology can interact and benefit one another.
AB - A neural language model trained on a text corpus can be used to induce distributed representations of words, such that similar words end up with similar representations. If the corpus is multilingual, the same model can be used to learn distributed representations of languages, such that similar languages end up with similar representations. We show that this holds even when the multilingual corpus has been translated into English, by picking up the faint signal left by the source languages. However, just as it is a thorny problem to separate semantic from syntactic similarity in word representations, it is not obvious what type of similarity is captured by language representations. We investigate correlations and causal relationships between language representations learned from translations on one hand, and genetic, geographical, and several levels of structural similarity between languages on the other. Of these, structural similarity is found to correlate most strongly with language representation similarity, whereas genetic relationships—a convenient benchmark used for evaluation in previous work—appears to be a confounding factor. Apart from implications about translation effects, we see this more generally as a case where NLP and linguistic typology can interact and benefit one another.
UR - http://www.scopus.com/inward/record.url?scp=85081180389&partnerID=8YFLogxK
U2 - 10.1162/COLIa00351
DO - 10.1162/COLIa00351
M3 - Journal article
AN - SCOPUS:85081180389
VL - 45
SP - 381
EP - 389
JO - Computational Linguistics
JF - Computational Linguistics
SN - 1530-9312
IS - 2
ER -
ID: 239206885