Advances in monolingual and crosslingual automatic disability annotation in Spanish

BMC Bioinformatics

Table 2 Overview of the different approaches used for automatic disability annotation, tested on Spanish data

Main architecture	System	Stacked external embeddings	Level of granularity	Train/dev
Monolingual approaches
BiLSTM-CRF	2018 Shared task Best system [49]	W2V static word + character features (EHR)	Static word	DIANN Spa
	Best published Result [14]	GLOVE static word + character features (general texts)	Static word
	\(FLAIR_{FT}\) [37]	FASTTEXT Static word & subword (general texts)	Contextual character
	\(FLAIR_{Wiki2V}\) [37]	Wikipedia2Vec Static word
	\(FLAIR_{SkipNG\_EHR}\) [43]	SkipNG Static word (EHR)
	\(FLAIR_{LM\_EHR}\) [37]	FLAIR contextual Character (EHR)
Transformer	BETO [44]	Spanish contextual subword	Spanish contextual subword
Crosslingual approaches
BiLSTM-CRF	\(FLAIR_{MUSE}\) [46]	Bilingual static subword	Contextual character	DIANN Spa/Eng
	\(FLAIR_{ME}\) [47]	Static word
	\(FLAIR_{mBERT}\) [45]	Contextual subword
Transformer	\(XLM-R\) [48]	Multilingual contextual subword	Multilingual contextual subword

The upper table presents the experiments with monolingual approaches (training with Spanish data) and the lower table the ones using crosslingual approaches (from English to Spanish and vice versa)

ISSN: 1471-2105