Natural 
Panel 
Language Interfaces - Ready for Commercial Success? 
Wolfgang Wahl~tor (chair) 
Department of Computer Seienco 
University of Saarbrfickon 
1} 6600 Saarbrficken 11 
Fed. Rep. of Germany 
Panelists: aalnm G. Carbone\]l, Gary G. Hendri×, Marry Tnnnant 
The goal of this panel is to evaluate three natural language in 
terfaces which were introduced to the commercial market in 
1985 (cf. Carnegie Group 1985, Kamins 1985, Texas Instru 
ments 1985) and to >'elate them to current research in corn 
putational linguistics. Each of the commercial systems selec- 
ted as a starting point for the discussion (see Wahlsler 1986 
for a functional comparision) was developed by a well-known 
scientist with considerable research experience in NL proces- 
sing: Language Craft TM by Carnegie Group (designed under 
tile direction of.\]. Carboncll), NLMenu by Texas Instruments 
(designed untler the direclion of II. Tennant), and Q & A TM 
by Symamec (designed under the direction of G. Hendrix). 
This discussion colnplements a panel at COLING 84- (see 
Sparck Jones et al. 1984), which debated the proposition that 
database query is no longer a paradigmatic problem tor com- 
putational linguistics. 
A commercial success ofNI, imerfaces could have a great value 
to our tield. When thousanct~ of users appreciate the capabili- 
ties of NI, interlaces, computational linguists have something 
to be proud of. The economic interest might push fbrward 
the ne.eded basic research and secure its funding. For the first 
time in the history of computational linguistics tile large user 
comumnities of NL systems sold as mass products provide a 
basis tbr serious empirical research on the dialog behavior of 
real end-users. 
The discussion may inlluenc.e computational linguistics as a 
basic science by tbeussing it to actual needs and by pointing 
out new problems which are typically identified when NL in- 
terfaces are transle.rred out of research laboratories. 
I askcd tile panelists to discuss the following questions in par-- 
ticular: 
I) How successful are the comnmrcial sys|ems? 
- \]\[low many copies have been sold? 
- To what extent are they being used on a regular basis 
by the customers? 
How much money do they save? Are there any cost- 
benefit analyses? 
Are the systems empirically evaluated in a cmnmer- 
cial environment? Which evaluation methodologies are 
used ? 
How well do current users adapt to the limitations of 
the syslems? 
What are the main areas of application? 
2) Wha| are the most important factors for success ofa NL 
interf?tce: 
-- the linguistic and conceptual coverage? 
its robuslness? 
the degree of portability and transportability? 
the qualily of tools lbr adapting the systems to a new 
altplication? 
3) Isn't it too early tbr a commercialization of NL systems? 
4) l)on't you see the danger that tile customers lose interest 
in NL processing forever if their ow:rblown expectations 
cannot be fulfilled by the systems currently available? 
5) Is it just a tremendous amount of engineering to trans- 
ff)rm a research system into a commercial product mr is it 
much more (e.g. new concepts, algorithms, data struc- 
tures?) 
6) How do you see the evolution of tile market for NL in- 
terfaces? What are your market forcast hypotheses? 
7) Should commercial systems be designed for mainframes, 
powerful AI workstations or personal computers? 
8) Is it a market advantage if the system is multilingual 
(e.g. if there exists not only an English version but also 
a French, German, Spanish and Italian one)? 
9) What cm'rent research in.computational linguistics and 
artificial intelligence seems most important tbr improving 
lhe quality of the current products? 
10) Considering the next generation of commercial NL sy- 
stems, how do you assess the role of 
- NL generation 
- cooperative response generation 
speech act recognition 
-- user modeling 
-hybrid representation languages (e.g. KL-TWO, 
KRYPTON) 
- new types of grammars (e.g. FUG, GPSG, LFG, 
TAG) 
- multimedia communication (e.g. luixtures of text, gra- 
phics, gestural input). 
References 

Carnegie Group (1985): The Language Craft TM Manual. Re- 
lease 2.0. Pittsburgh, i A. 

Kamins, S. (1985): Instruclion ManualQ&A. Version 1.0. 
\])oeunnent No.: 1-001. Cupertino: Symantec. 
Sparck Jones, K., Bates, M., Carbonell, J.G., Fliekinger, D., 

McKeown, K.R. (1984.): Natural Language and Data- 
bases, Again. In: Proc. of COLING 84, Stanford, CA., 
p. 182-193 

Texas Instruments (1985): Natural Language Menu User's 
Guide. Digital Systems Group, Dallas, Texas. 

Wahlster, W. (1986): The Role of Natural Language in Ad- 
vanced Knowledge-Based Systems. In: Winter, H. led.): 
Artificial Intelligc,~Pe and Man-Machine Systems. (Lec- 
ture Notes in Control and Information Sciences, VoI. 80), 
Berlin, tIeidelberg , New York, Tokyo: Springer, p. 62-83
