An accurate grammar analyzer that works effectively even with error-ridden sentences
input by learners, based on a context-free probabilistic statistical POST (part-of-speech
tagged) parser, for a template-automation-based computer-assisted language learning
system. For any keyed-in sentence, the parser finds a closest correct sentence
to the keyed-in sentence from among the embedded template paths exploiting a highest
similarity value, and generates a grammar tree for the correct sentence where some
ambiguous words are preassigned by expert language teachers. The system marks the
errors under the leaves of the grammar tree by identifying the differences between
the keyed-in sentence and the grammar tree of the correct sentence as errors committed
by learners. By identifying most frequently recurring grammatical errors of each
student, the system sets up a learner's model, providing a unique level of contingent
remediation most appropriate to each learner involved.