@inproceedings{kim-etal-2014-temporal,
title = "Temporal Analysis of Language through Neural Language Models",
author = "Kim, Yoon and
Chiu, Yi-I and
Hanaki, Kentaro and
Hegde, Darshan and
Petrov, Slav",
booktitle = "Proceedings of the {ACL} 2014 Workshop on Language Technologies and Computational Social Science",
month = jun,
year = "2014",
address = "Baltimore, MD, USA",
publisher = "Association for Computational Linguistics",
url = "https://www.aclweb.org/anthology/W14-2517",
doi = "10.3115/v1/W14-2517",
pages = "61--65",
}
1. What is it?
They proposed a method using Neural Language Model into semantic change.
2. What is amazing compared to previous works?
They used word2vec trained each year and initialized next one by past embedding.
3. Where is the key to technologies and techniques?
Learning time series word embeddings without any alignment
a word embeddings in time t is initialized by the previous year t-1's word embedding.
they iterate over epochs until convergence, which is defined as the average angular change.
4. How did evaluate it?
Their model can detect the semantic change of words automatically(by cosine similarity).
a1da4
commented
4 years ago
0. Paper
@inproceedings{kim-etal-2014-temporal, title = "Temporal Analysis of Language through Neural Language Models", author = "Kim, Yoon and Chiu, Yi-I and Hanaki, Kentaro and Hegde, Darshan and Petrov, Slav", booktitle = "Proceedings of the {ACL} 2014 Workshop on Language Technologies and Computational Social Science", month = jun, year = "2014", address = "Baltimore, MD, USA", publisher = "Association for Computational Linguistics", url = "https://www.aclweb.org/anthology/W14-2517", doi = "10.3115/v1/W14-2517", pages = "61--65", }
1. What is it?
They proposed a method using Neural Language Model into semantic change.
2. What is amazing compared to previous works?
They used word2vec trained each year and initialized next one by past embedding.
3. Where is the key to technologies and techniques?
Learning time series word embeddings without any alignment
a word embeddings in time t is initialized by the previous year t-1's word embedding. they iterate over epochs until convergence, which is defined as the average angular change.
4. How did evaluate it?
Their model can detect the semantic change of words automatically(by cosine similarity).
5. Is there a discussion?
6. Which paper should read next?
Previous works by statistical way