POS tagging: deleted interpolation

[Massimo I]

Hi,
I just wanted to make sure I got this correctly;
In deleted inerpolation, we compare the MLE value of a trigram with those of its respective bigram and unigram. We then pick the highest one and increment its N-gram’s Lambda value by the trigram count. We do this for each trigram in the corpus, and then normalise the Lambda values so that they add up to 1? So each corpus will be matched with its own set of Lambda values, which we use for calculating all the interpolated N-grams for that corpus?

[Simon]

This is well beyond the scope of what we need to know about POS tagging for speech synthesis! We just need to know that

– POS tagging is very accurate for languages where a large corpus of hand-tagged data is available to train the tagger
– a typical method is HMMs+ngrams with the model being trained on that corpus

[Author]

Posts