An analysis of Ill-formed input in natural language queries

Abstract

The most common errors were missing conjunc­ tions; 18.31 per cent of the queries lacked a needed con­ junction. In almost all of these, the English conjunction and was needed to connect a list of words in a scries. The next most serious problem was punctuation errors; these were cither incorrect, missing or misplaced punctuation. Most involved the comma or quotation mark combined with other punctuation markj. The queries containing punctuation errors totaled 6.81 per cent. The third most common type of ill-formed input was spelling errors; 2.34 per cent of the queries had a spelling error. Since some of the spelling errors had been corrected when the data was entered for the computer analysis, it is almost certain Uiat the original queries contained more spelling errors. We did not find any co-occurrence violations such as subjccl/vcrb disagreement or use of ellipsis. Slightly less than one third of the queries examined contained some form of error. This rate is comparable to that found in an earlier study (Eastman and McLean, 1981). However, at least one earlier study found almost no instances of ill-formed input (Fineman, 1983).