Feature Engineering for Machine Learning in Natural Language Processing

 

Workshop at the Annual Meeting of the Association of Computational Linguistics (ACL 2005)

 

WORKSHOP DATE: June 29, 2005

 

LINK: Call for Papers

 

INTRODUCTION

The ACL 2005 Workshop on Feature Engineering for Machine Learning in Natural Language Processing is an opportunity to explore the various dimensions of feature engineering for problems that are of interest to the ACL community. Feature Engineering encompasses feature design, feature selection, feature induction, studies of feature impact (including feature ablation studies), and related topics. In 2003, there was a NIPS workshop on feature engineering (“Feature Extraction and Feature Selection”), but the focus was not on NLP problems specifically. Also, although the various aspects of feature engineering have been dealt with at times in various ACL forums, until now, to our knowledge, the spotlight has never been shone directly on this topic specifically for NLP and language technology problems. We feel that now is the time to look more closely.

As experience with machine learning for solving natural language processing tasks accumulates in the field, practitioners are finding that feature engineering is as critical as the choice of machine learning algorithm, if not more so. Feature engineering significantly affects the performance of systems and deserves greater attention. Also, in the wake of the shift in our field away from knowledge engineering and of the successes of data-driven and statistical methods, researchers are likely to make further progress by incorporating additional, sometimes familiar, sources of knowledge as features. Feature design may benefit from expert insight even where the relative merits of features must be assessed through empirical techniques from data. Although some experience in the area of feature engineering is to be found in the theoretical machine learning community, the particular demands of natural language processing leave much to be discovered.

In the call for papers, we expressed our intent of bringing together practitioners of NLP, machine learning, information extraction, speech processing, and related fields with the goal of sharing experimental evidence for successful approaches to feature engineering. Judging by the quality and diversity of the submissions received, we believe we have succeeded, and the resulting program should be of great interest to many researchers in the ACL community. We hope that the workshop will contribute to the distillation of best practices and to the discovery of new sources of knowledge and features previously untapped.

We also extend an open invitation to the reader to continue investigation in all aspects of feature engineering for machine learning in NLP, including:
 

• Novel methods for discovering or inducing features, such as mining the web for closed classes, useful for indicator features.
• Comparative studies of different feature selection algorithms for NLP tasks.
• Error analysis tools that help researchers to identify ambiguous cases that could be disambiguated by the addition of features.
• Error analysis of various aspects of feature induction, selection, representation.
• Issues with representation, e.g., strategies for handling hierarchical representations, including decomposing to atomic features or by employing statistical relational learning.
• Techniques used in fields outside NLP that prove useful in NLP.
• The impact of feature selection and feature design on such practical considerations as training time, experimental design, domain independence, and evaluation.
• Analysis of feature engineering and its interaction with specific machine learning methods commonly used in NLP.
• Ensemble methods employing diverse types of features.
• Studies of methods for inducing a feature set, for example by iteratively expanding a base feature set.
• Issues with representing and combining real-valued and categorical features for NLP tasks.

We anticipate that contributions in these areas will move the field of NLP and language technologies forward, with greater system performance and further insight into our own data and perhaps language itself.

We wish to thank all of the researchers who submitted papers to the workshop. Also, thanks go to the entire program committee and those who assisted them in their reviewing responsibilities.

Best regards,

Eric Ringger

Microsoft Research (USA)

 

 

WORKSHOP SCHEDULE