Limitations of Co-Training for Natural Language Learning from Large Datasets

By David R. Pierce, Claire Cardie. 2001

204 citations recorded by the source corpus.

How researchers see this paper

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

In Natural Language Processing

Citations in the corpus

Semi-supervised sequence tagging with bidirectional language models (2017, methodology): “Other semisupervised learning methods for structured prediction problems include co-training (Blum and Mitchell, 1998; Pierce and Cardie, 2001), expectation maximization (Nigam et al.”
A Survey on Multi-view Learning (2013, methodology): “novel method has caught the attention of many researchers and has been widely applied in the ﬁeld of natural language processing (Craven et al., 2000; Mu¨ller et al., 2002; Phillips and Riloﬀ, 2002). Pierce and Cardie (2001) studied the learning behavior of co-training and showed that given a small ”
Semi-Supervised Regression with Co-Training (2005, background): “Although the requirement of sufficient and redundant views is quite strict, the co-training paradigm has already been used in many domains such as statistical parsing and noun phrase identification[Hwa et al., 2003][Pierce and Cardie, 2001][Sarkar, 2001][Steedmanet al., 2003].”
Extractive Summarization Using Supervised and Semi-Supervised Learning (2008, background): “A number of successful studies emerged thereafter for other natural language processing tasks, such as text classification (Denis and Gilleron, 2003), noun phrase chunking (Pierce and Cardie, 2001), parsing (Sarkar, 2001) and reference or relation resolution (Muller et al., 2001; Li et al., 2004).”
Example Selection for Bootstrapping Statistical Parsers (2003, methodology): “We show that incorporating the utility criterion into the selection method results in better parsers for both frameworks.”
Co-Training and Expansion: Towards Bridging Theory and Practice (2004, background): “named entity classification [3], text classification [10, 5] , natural language processing [13], large scale document classification [12], and visual detect ors [8].”
A Framework for Learning Predictive Structures from Multiple Tasks and Unlabeled Data (2005, methodology): “However, it was pointed out by Pierce and Cardie (2001) that this method may degrade the classification performance when the assumptions of the method are not satisfied (that is when noise is introduced into the labels through non-perfect classification).”
Semi-Supervised Learning for Natural Language (2005, background): “, 2001; Abney, 2002) and empirical analysis (Nigam and Ghani, 2000; Pierce and Cardie, 2001) have been done on co-training.”
Bootstrapping Coreference Classifiers with Multiple Machine Learning Algorithms (2003, methodology): “Since the B&M co-training algorithm is sensitive not only to the views employed but also to other input parame- ters such as the pool size and the growth size (Pierce and Cardie, 2001), we evaluate the algorithm under different parameter settings, as described below.”
Semi-Supervised Active Learning for Sequence Labeling (2009, methodology): “Pierce and Cardie (2001) showed that the quality of the automatically labeled training data is crucial for co-training to perform well because too many tagging errors prevent a highperforming model from being learned.”
A High-Performance Semi-Supervised Learning Method for Text Chunking (2005, background): “For example, co-training (Blum and Mitchell, 1998) automatically bootstraps labels, and such labels are not necessarily reliable (Pierce and Cardie, 2001).”
Analyzing Co-training Style Algorithms (2007, background): “Note that in some empirical study of the natural language processing community, it has been found that sometimes the performances of the two learners can degrade if the co-training process is run to convergence [14].”