Early work in computational linguistics did not emphasize distinctions between theoretical and applied efforts. The division came into existence naturally, with the growth of the field, and was caused by a mixture of scientific, logistical and sociological factors. Detailed intellectual history of computational linguistics is, of course, still waiting for its chronicler.1 But an important early event contributing to the split between the theoretical and applied NLP must be mentioned. The 1966 ALPAC report criticized the practical achievements of the early machine translation efforts in the US. It was perceived by the public (and the funding agencies) as a death knell to the earliest - and widely publicized - attempt by computational linguists to build natural language processing systems that were actually deployed.
As a result of the report, U.S. Government funding for applied NLP shrank, but the real needs for NLP, needless to say, remained strong. CL and NLP practitioners at the time largely concentrated either on theoretical work, work on techniques and methods, or addressed less "glamorous" tasks, such as computer-assisted translation or spelling correction. The latter kind of work was often carried out at industrial and other non-academic research sites and might have been perceived by the academics as less than worthy of presentation at the computational linguistics community meetings. While the industrial researchers were less vulnerable to the "publish or perish" paradigm, they, naturally, also wanted to disseminate the results of their work. It is but a small logical step from here to the idea of creating a new venue dedicated to presenting and publishing such work.
By the early 1980s, the computational linguistics community started to regain its popularity in the wide world after a fifteen-year hiatus brought about by the ALPAC report. Research on MT rebounded in the US, and new areas, notably, natural language interaction with databases, became popular. In view of the above, it is not surprising that the first ACL Conference on Applied Natural Language Processing (popularly known as ANLP or even "Applied ACL") was convened. Of the eight organizers of the conference, only one was at the time affiliated with a university.
At first, ANLP conferences were held at irregular intervals, presumably, on the "as needed" basis. Gradually, the cycle has become more stable. The topics and kinds of material presented and discussed at these conferences have evolved over the years, with the changes in both the accepted research paradigms (e.g., corpus-based NLP) and the acceptable levels of output quality. It is significant, for example, that in early days of NLP 100% quality of output was accepted as a given and when a system (not surprisingly) failed to attain it, human intervention was assumed to bridge the quality gap. Since the early 1990s, the burden of bridging this gap has been moved on to the consumer of NLP, with good uses for "crummy"2 NLP methods sought and extolled. This more practical attitude to computational linguistics belongs squarely in the ANLP purview, and the recent conferences reflect this state of affairs. In general, though, the boundary between material that fits the ANLP conference and contributions that should belong to its big brother is difficult to demarcate (see, e.g., Ralph Grishman's Program Chair's Preface to the Proceedings of ANLP-97). Time will tell whether the two conferences will further diverge or whether the distinction will for a period of time lose its significance. With the currently high and rising societal and commercial demand for NLP applications, it seems that the number of deployed NLP systems will continue to grow in the years to come, and with them, the need for describing and discussing them. In short, the immediate future of ANLP conferences seems to be bright.
1. The 1989 ACL Presidential Address by Jerry Hobbs can, however, be considered a good start.
2. The term is due to Ken Church and Ed Hovy.