Reliable tagging of Temporal Expressions (TEs, e.g., Book a table at L’Osteria for Sunday evening) is a central requirement for Voice Assistants (VAs). However, there is a dearth of resources and systems for the VA domain, since publicly-available temporal taggers are trained only on substantially different domains, such as news and clinical text.
Since the cost of annotating large datasets is prohibitive, we investigate the trade-off between in-domain data and performance in DA-Time, a hybrid temporal tagger for the English VA domain which combines a neural architecture for robust TE recognition, with a parser-based TE normalizer. We find that transfer learning goes a long way even with as little as 25 in-domain sentences: DA-Time performs at the state of the art on the news domain, and substantially outperforms it on the VA domain.
%0 Conference Paper
%1 alam21:_new_domain_major_effor
%A Alam, Touhidul
%A Zarcone, Alessandra
%A Padó, Sebastian
%B Proceedings of IWCS
%C Online
%D 2021
%K conference myown
%P 144--154
%T New Domain, Major Effort? How Much Data is Necessary to Adapt a Temporal Tagger To the Voice Assistant Domain
%U https://iwcs2021.github.io/proceedings/iwcs/pdf/2021.iwcs-1.14.pdf
%X Reliable tagging of Temporal Expressions (TEs, e.g., Book a table at L’Osteria for Sunday evening) is a central requirement for Voice Assistants (VAs). However, there is a dearth of resources and systems for the VA domain, since publicly-available temporal taggers are trained only on substantially different domains, such as news and clinical text.
Since the cost of annotating large datasets is prohibitive, we investigate the trade-off between in-domain data and performance in DA-Time, a hybrid temporal tagger for the English VA domain which combines a neural architecture for robust TE recognition, with a parser-based TE normalizer. We find that transfer learning goes a long way even with as little as 25 in-domain sentences: DA-Time performs at the state of the art on the news domain, and substantially outperforms it on the VA domain.
@inproceedings{alam21:_new_domain_major_effor,
abstract = {Reliable tagging of Temporal Expressions (TEs, e.g., Book a table at L’Osteria for Sunday evening) is a central requirement for Voice Assistants (VAs). However, there is a dearth of resources and systems for the VA domain, since publicly-available temporal taggers are trained only on substantially different domains, such as news and clinical text.
Since the cost of annotating large datasets is prohibitive, we investigate the trade-off between in-domain data and performance in DA-Time, a hybrid temporal tagger for the English VA domain which combines a neural architecture for robust TE recognition, with a parser-based TE normalizer. We find that transfer learning goes a long way even with as little as 25 in-domain sentences: DA-Time performs at the state of the art on the news domain, and substantially outperforms it on the VA domain.},
added-at = {2021-04-17T21:09:02.000+0200},
address = {Online},
author = {Alam, Touhidul and Zarcone, Alessandra and Padó, Sebastian},
biburl = {https://puma.ub.uni-stuttgart.de/bibtex/23370f1e6a50186bf152c32cae388eaa8/sp},
booktitle = {Proceedings of IWCS},
interhash = {640672c5b09f9640eafc620ec084ebfe},
intrahash = {3370f1e6a50186bf152c32cae388eaa8},
keywords = {conference myown},
pages = {144--154},
timestamp = {2021-08-11T11:57:32.000+0200},
title = {New Domain, Major Effort? {H}ow Much Data is Necessary to Adapt a Temporal Tagger To the Voice Assistant Domain},
url = {https://iwcs2021.github.io/proceedings/iwcs/pdf/2021.iwcs-1.14.pdf},
year = 2021
}
