@inproceedings{dubossarsky-etal-2019-time,
title = "Time-Out: Temporal Referencing for Robust Modeling of Lexical Semantic Change",
author = "Dubossarsky, Haim and
Hengchen, Simon and
Tahmasebi, Nina and
Schlechtweg, Dominik",
booktitle = "Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics",
month = jul,
year = "2019",
address = "Florence, Italy",
publisher = "Association for Computational Linguistics",
url = "https://www.aclweb.org/anthology/P19-1044",
doi = "10.18653/v1/P19-1044",
pages = "457--470",
abstract = "State-of-the-art models of lexical semantic change detection suffer from noise stemming from vector space alignment. We have empirically tested the Temporal Referencing method for lexical semantic change and show that, by avoiding alignment, it is less affected by this noise. We show that, trained on a diachronic corpus, the skip-gram with negative sampling architecture with temporal referencing outperforms alignment models on a synthetic task as well as a manual testset. We introduce a principled way to simulate lexical semantic change and systematically control for possible biases.",
}
1. What is it?
They proposed Temporal Referencing (avoid alignment), a method for training a model for lexical semantic change.
2. What is amazing compared to previous works?
Their method does not need alignments and achieve high performance
3. Where is the key to technologies and techniques?
Temporal Referencing
They train their model with all time span
Each target word w in time t is replaced w_t, so they can obtain a single vector space.
4. How did evaluate it?
4.1 Noise Detection
Dubossarsky et al. (2017) proposed an evaluation method
train models with diachronic corpus
train models with shuffled corpus
compute average cosine distance in each model
check the difference in average cosine distance between model_genuine and model_shuffled.
The high average cosine distance shows the better model.
Their method (TR) achieved higher performance than state-of-the-art model (AL, alignment)
avoid alignment
product more stable context vector by increased volume of data
Also their model obtain more constant cumulative progression of cosine distances over time.
4.2 Nearest neighbors
Their model (TR) can capture the effect of digital computer showing.
a1da4
commented
4 years ago
0. Paper
@inproceedings{dubossarsky-etal-2019-time, title = "Time-Out: Temporal Referencing for Robust Modeling of Lexical Semantic Change", author = "Dubossarsky, Haim and Hengchen, Simon and Tahmasebi, Nina and Schlechtweg, Dominik", booktitle = "Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics", month = jul, year = "2019", address = "Florence, Italy", publisher = "Association for Computational Linguistics", url = "https://www.aclweb.org/anthology/P19-1044", doi = "10.18653/v1/P19-1044", pages = "457--470", abstract = "State-of-the-art models of lexical semantic change detection suffer from noise stemming from vector space alignment. We have empirically tested the Temporal Referencing method for lexical semantic change and show that, by avoiding alignment, it is less affected by this noise. We show that, trained on a diachronic corpus, the skip-gram with negative sampling architecture with temporal referencing outperforms alignment models on a synthetic task as well as a manual testset. We introduce a principled way to simulate lexical semantic change and systematically control for possible biases.", }
1. What is it?
They proposed Temporal Referencing (avoid alignment), a method for training a model for lexical semantic change.
2. What is amazing compared to previous works?
Their method does not need alignments and achieve high performance
3. Where is the key to technologies and techniques?
Temporal Referencing
They train their model with all time span Each target word w in time t is replaced w_t, so they can obtain a single vector space.
4. How did evaluate it?
4.1 Noise Detection
Dubossarsky et al. (2017) proposed an evaluation method
Their method (TR) achieved higher performance than state-of-the-art model (AL, alignment)
Also their model obtain more constant cumulative progression of cosine distances over time.
4.2 Nearest neighbors
Their model (TR) can capture the effect of digital computer showing.
5. Is there a discussion?
6. Which paper should read next?
Dubossarsky et al. (2017)