AUTOLING : AN AUTOMATED LINGUISTIC FIELDWORKER 
by Sheldon KLEIN 
+COmputer Sciences Department 
University of Wisconsin 
MADISON-WISCONSIN 53706 
U. S. A. 
Sheldon Klein, Barbara G. Davis, William Fabens, Robert G. Herriot, Bill J. 
Katke, Michael A. Kuppin and Alicia E. Towster. 
ABSTRACT 
A system intended to act as a linguistic fieldworker via teletype 
'interaction with a linguistically unsophisticated informant has been designed 
and ls being progra,~med in extende~GOL for the Burroughs5500 and8500 
computers. 
The system consists of the three major analytic components ; a program 
for performing morphological analyses and deriving a segmentation algorithm for 
sentences in any language ; a syntactic learning program that formulates context 
free and context sensitive phrase structure rules (monolingual learning component 
to be added later) ; and a machine translation program that learns to translate 
in both directions between the query language (English) and the language of the 
llve informant via bi-lingual transformations. 
The informant may be viewed as a fourth component, and is assumed 
to be able to read and write English in standard graphemics, and to be able to 
read and writeh~on-English language in a phonemic notation. 
-I- 
The analytic methods used by the AUTOLIN@ are heuristic rather tha 
algorithmic, and hence do not guarantee complete success. ~he various components 
make use of protocols derived~from the experience of live fieldworkers . The 
representations of these strategies are relatively separeted from the analytic 
mechanisms of the program and, accordingly, may be deleted, altered or incre- 
mented at the discretion of the system designers~ 
As a partial illustration of the system's operation, assume that 
the analystic process is at an intermediate stage, i.e. some morphological, 
syntactic and translation rules have been posited. Under the control of a moni- 
tor program, the system would test a nexly formulated syntactic rule by : 
i- Generating (via usage of the new rule) a form implied by the 
grammar butnever elicited from the live informant. 
2- Translating the Hypothetical form into the query language 
(English). 
3- Asking the informant via a teletype query to translate the 
En~glish output of step 2. 
If the informant's reply is equivalent to the form derived from 
step 2, the newly posited rule is maintened in the tenative grammar (although 
subject to later check). If the informant's reply deviates from the output of 
step 2, the new rule is treated as suspect and subjected to further verification 
procedures. If thedeviating reply contains new morphological material, the 
monitor program instructs the morphological analytic component to update its 
analysis. The machine translation learning component would also recheck and 
update its rules. 
The program as a whole continues its interaction with the informant 
indefini~.ly, using the just mentioned strategies as well as others. At any 
given time the system designers may interrupt the field work process and ask the 
machine to list its corpus and current rules. 
-2- 
