Bidirectional Inference with the Easiest-First Strategy for Tagging Sequence Data

By Yoshimasa Tsuruoka, Junichi Tsujii. 2005

286 citations recorded by the source corpus.

This paper presents a bidirectional inference algorithm for sequence labeling problems such as part-of-speech tagging, named entity recognition and text chunking. The algorithm can enumerate all possible decomposition structures and find the highest probability sequence together with the corresponding decomposition structure in polynomial time. We also present an efficient decoding algorithm based on the easiest-first strategy, which gives comparably good performance to full bidirectional inference with significantly lower computational cost. Experimental results of part-of-speech tagging and text chunking show that the proposed bidirectional inference methods consistently outperform unidirectional inference methods and bidirectional MEMMs give comparable performance to that achieved by state-of-the-art learning algorithms including kernel support vector machines.

Subfields: Artificial Intelligence

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

Recurrent Neural Networks for Classifying Relations in Clinical Notes (2017, neutral): “, from cTakes [19], MetaMap [20], and GeniaTagger [54]) to recognize modifiers of head words in phrases [13] to address the inconsistent annotation issue.”
Guided Learning for Bidirectional Sequence Classification (2007, methodology): “With set E, we use prefixes and suffixes of length up to 9, as in (Toutanova et al., 2003; Tsuruoka and Tsujii, 2005).”
Learning What’s Easy: Fully Differentiable Neural Easy-First Taggers (2017, methodology): “Vanilla easy-first decoders have been used in POS tagging (Tsuruoka and Tsujii, 2005; Ma et al., 2013), dependency parsing (Goldberg and Elhadad, 2010), and coreference resolution (Stoyanov and Eisner, 2012), being related to cyclic dependency networks and guided learning (Toutanova et al.”
Deep learning for extracting protein-protein interactions from biomedical literature (2017, methodology): “We used the chunk tags obtained from Genia Tag-ger for each word (Tsuruoka and Tsujii, 2005).”
Posterior Regularization for Structured Latent Variable Models (2010, methodology): “We used the Tokyo tagger (Tsuruoka and Tsujii, 2005) to POS tag the English tokens, and generated parses using the ﬁrst-order model of McDonald et al. (2005) with projective decoding, trained on sections 2-21 of the Penn treebank with dependencies extracted using the head rules of Yamada and…”
Text Classification using Graph Mining-based Feature Extraction (2010, methodology): “The text was divided into sentences using a regular expression b ased tokenizer and then each sentence was POS tagged using Tsuruoaka and Tsujii’s “g eniatagger”[18], producing a sequence of “word/POS” tokens plus the lemma (stemm ed form) of each word.”
Biomedical text mining and its applications in cancer research (2013, unknown): “GENIATagger [132,133] The GENIA tagger is specifically tuned for biomedical text such as MEDLINE abstracts.”
Segment convolutional neural networks (Seg-CNNs) for classifying relations in clinical notes (2018, methodology): “All participating systems in the 2010 i2b2/VA challenge utilized heavy feature engineering for their machine learning models(13); many also harvested features from existing NLP pipelines such as cTakes,(14) MetaMap,(15) and GeniaTagger.(16) All systems combined lexical, syntactic, and semantic featu”
DCU Terminology Translation System for Medical Query Subtask at WMT14 (2014, methodology): “For the extraction from parallel corpora (cf. Section 2.2), we used Genia tagger (Tsuruoka and Tsujii, 2005) and the Berkeley parser (Petrov and Klein, 2007).”
ScispaCy: Fast and Robust Models for Biomedical Natural Language Processing (2019, methodology): “However, tools which cover more classical natu-ral language processing (NLP) tasks such as the GENIA tagger (Tsuruoka et al., 2005; Tsuruoka and Tsujii, 2005) and phrase structure parsers such as those presented in (McClosky and Charniak, 2008) typically do not make use of new research innovations…”
Improving Gender Classification of Blog Authors (2010, unknown): “We used (Tsuruoka and Tsujii , 2005) as our POS tag-ger.”
Part-of-Speech Tagging from 97% to 100%: Is It Time for Some Linguistics? (2011, background): “Some of the other recent taggers cited earlier have made use of even more higher order features and conjunctive features [2, 10], but in our tagger they seem to provide marginal to negative gains.”
Automatic classification of sentences to support Evidence Based Medicine (2011, methodology): “Also, [12] proposed a probability-based learner inspired by the sequence of abstracts.”
Overview of BioCreative II gene mention recognition (2008, methodology): “Runs 1 and 3 used CRFs with different settings of the Gaussian prior, whereas run 2 used a bidirectional maximum entropy Markov model (BMEMM) [25].”