Error-Analysis Semantics

In order to determine an error a system needs some kind of description of what is correct. For semantic errors this amounts to a (almost) complete description of the possible states of the world including all things in the world. This task has not been realized as yet which makes it difficult to identify semtic errors in general. However limiting the domain in a sensible way can provide a system with some capabilities for recognizing semantic errors. The presentation of a so called micro world is a step towards this goal.

In the SEMF-project the system tried to recognize not only syntactic errors but also semantic errors. Some restrictions need to be added to realize such an ambitious task. In the example here the input sentences are always decriptions of a so called blocks world. After parsing a sentence syntactically the system then checks if the semantic representation of the sentence matches the actual blocks world at that moment. Since the contents can only refer to the items in the blocks world this seems to work. An application of this method to larger and more complicated worlds seems albeit difficult. A complete representation of a micro world would have to be developed not only with states but possibly also including changes of states (but see Hamburger, Henry in: Appelo, L. ; de Jong, F. (Ed.), 1994, Hamburger, Henry in: Holland, V.M. ; Kaplan, J.D. et al. (Ed.), 1995, who proposes a kitchen scene manipulated by the learner).

Some more basic semantic knowledge such as thematic restrictions may already be covered in the system's lexicon. This approach has been tried in the ReCALL project. The grammar rules were modified in order to include semantic information parallel to the usual syntactic agreement information. This could then be used in order to generate appropriate feedback to a learner if the restrictions were violated. Again these constraints can only be used in very restricted cases and are not a general method of identifying semantic errors.

To summarize almost all linguistic levels have been considered in research on error-analysis. Most has been done on grammar checking, which on the one hand seems to be managable and on the other hand is important for learners. Little work has been done on the presentation of error-messages to learners (but see Heift, Trude, 2001). This second important step has not been considered much maybe because analysing errors is already a very difficult task. As mentioned above, there is some research from the teaching perspective, e.g. Gnutzmann, C. ; Kiffe, M., 1993. Here the question is, which type of error-handling students prefer while speaking in foreign-language classes.