Grammar or style checkers check for syntactic incorrectness, such as subject-verb agreement and capitalization of proper names. Syntactic errors are typically found in surface linguistic forms and are easy to detect by following a list of pedagogical conventions. For example, “He study hard” would be marked as ungrammatical, because this sentence does not comply to subject-verb agreement. While sentences with obvious grammatical errors can be difficult to read, they do not hinder proper interpretation in general.
In contrast, our VTAS ambiguity detector checks for semantic and pragmatic incorrectness. We do not aim to detect superficial grammatical errors; instead, we target structures that are perfectly grammatical but are susceptible to ambiguous interpretations. Occurring at a deeper cognitive level, semantic ambiguities can be more disruptive in understanding and result in more costly errors. Despite the fact that ambiguities are extremely difficult to detect, our application tackles some representative ambiguous structures in its initial release, and its detection accuracy significantly improves as the application is trained on more data.