Search

The development of a system for the repair of lexical errors detected in the process of recognizing inorganic chemical names is described. Repairs are based on the calculation of similarity between strings, using the model and approximate matching algorithm detailed in the previous article of this series. A hierarchical data structure is proposed for the storage of information generated during the repair process. This structure comprises two hierarchically-related abstract data types, to permit the storage of partial and total repairs generated by the process. The great amount of information generated by the system might recommend its use in applications requiring explanatory subsystems to report on the success of recognition of user-input information. Problems in the repair process arising from inorganic grammar structures are also examined, and solutions are proposed for cases such as that of the delimiters between significant parts of the inorganic substance name.

The repair of lexicographic errors occuring during the computer translation of knowledge is performed by approximate string matching algorithms. This article presents a new algorithm that attempts to repair lexical errors in inorganic chemical names. The algorithm allows the repair of an incorrect string containing one or more errors due to deletion, insertion, or transposition of multiple characters. Repair may be carried out on all or only part of the incorrect string, so the algorithm may be used to repair strings containing more than one incorrect morpheme. A model is proposed for the use of this algorithm in the repair of lexical errors in inorganic chemical names. The model attempts to generate a set of possible repair strings free from either syntactic or lexicographic errors. Repair is performed by comparing all or part of the error string with the set of terminal symbols or morphemes included in the grammar. A new model is proposed for the calculation of similarity; it is highly discriminating and allows the repair of parts of an error string.

The treatment of errors generated in the transmission of chemical information from humans to machines is examined. Every type of human-machine communication requires a translation subprocess to deal with the various possible representations of knowledge; the present study was designed to consider errors occurring in such a subsystem. Regardless of the model employed, translation of knowledge is performed over several successive stages; at each stage different types of errors may be detected. A method is proposed to classify these errors according to the stage at which they occur, thus facilitating the generalization of the process. An analysis of errors occurring during the translation of inorganic chemical names is also presented. The particular grammatical features of different nomenclatures would require a study of the errors appearing in the human-machine communication process in modern chemistry-oriented systems, a problem which has only been touched upon in the systems proposed to date. Attention here is centered on lexicographic errors, since that stage of the translation process is completely independent of the model employed and might thus be useful in a more general sense.