In this paper, we present results of our ex- periments with ASR for a highly inflected Dravidian language, Telugu. First, we pro- pose a new metric for evaluating ASR per- formance for inflectional languages (Inflec- tional Word Error Rate IWER) which takes into account whether the incorrectly recog- nized word corresponds to the same lexi- con lemma or not. We also present results achieved by applying a novel method – er- rgrams – to ASR lattice. With respect to confidence scores, the method tries to learn typical error patterns, which are then used for lattice correction, and applied just be- fore standard lattice rescoring. Our confi- dence measures are based on word posteri- ors and were improved by applying antimod- els trained on anti-examples generated by the standard N-gram lan...

Click Here to View the Paper In XML Format