Topic: Natural Language Processing/Engineering

1. (a) What makes part-of-speech tagging non-trivial? [10 marks]

(b) Explain how labelled training data can be used to estimate the

probabilities needed by a Hidden Markov Model performing part-ofspeech

tagging. [10 marks]

(c) How would you go about performing a quantitative evaluation of a partof-

speech tagger? What method could you use to investigate whether

the tagger was often making the same sorts of tagging mistakes?

[10 marks]

(d) Why might part-of-speech tagging be a useful processing step in a

system that judges whether a document is expressing mostly opinions

or mostly facts? [10 marks]

(e) What are the similarities and the differences between stemming and

lemmatising? [10 marks]