Roadmaps and Resource Maps: Coordinating Language Resources for Cooperative Technology Development and Evaluation

Advances in speech technologies increase demand for linguistic data in more languages with more sophisticated annotation. In speech recognition research, corpora of dozens of broadcast hours, hundreds of conversations and tens of millions of words of text are replaced by thousands of broadcast hours, tens of thousands of conversations and billions words of text. The scope, scale and schedule of language resource development continues to evolve and the availability of high quality language remains a central issue for speech researchers. Although advancements in desktop computing allow researchers to create corpora in their offices, most cannot maintain the scale of corpus development their research requires. Instead they seek to find, re-annotate, reuse and re-distribute existing corpora. International data centers support grass-roots corpus creation by providing infrastructure, tools and best practices. However, such data centers must assume a more central role in creating and coordinating the creation of language recourse and must collaborate with their peer organizations to identify and address the needs of research communities and commercial developers. Initiative such as the ENABLER (European National Activities for Basic Language Resources) Project recognize and have created technology roadmaps to begin to address this need. The present paper will describe language resource activities in the United States focusing on those that coordinate effort across multiple areas of inquiry.