Part-of-speech tagging

A second interest in POS-tagging is that this technique can be used to assign any type of labels to the tokens, not just parts-of-speech. Hence we could use a tagging algorithm for general disambiguation purposes. 

POS-tagging has been a problem of concern in NLP for some time. There it is often used as a first required step before syntactic parsing is performed, but it is also used in other applications and seen as a problem in its own right. Many approaches have been tried but probably the most 

Chapter 5. Text Decoding Finding the words from the text 

Rather than compute the probability that a token generates a tag P(/ t), an HMM is a generative model which computes P t l), that is, the probability of seeing a token given the tag. We can calculate one from the other by concatenating the HMM models to form sequences and then apply Bayes rule  

We will consider the full issue of how to train an HMM in Section 15.1.8. For now, let us simply assume that we can calculate the transition probabilities and observation probabilities by simply counting occurrences in a labelled database. To see how the tagging operates consider the issue of resolving the classic POS homograph record. This can be a norm or a verb, and a trained HMM would tell us for instance  

Rgure 5.1 Example transition probabilities from a simpUlied four-state POS tagger. The probabilities exiting a state always add up to 1. 

Rgure 5.2 Example observation probabilities for three POS tags, showing the probability of observing a token given the state. 

---

Pipelined models Quite often the signal-to-signal model is implemented as a pipelined model were the process is seen as one of passing representations from one module to the next. Each module performs one specific task such as part-of-speech tagging, or pause insertion and so on. No explicit distinction is made between analysis and synthesis tasks. These systems are often highly modular, such that each module s job is defined as reading one type of information and producing another. Often the modules are not explicitly linked so that different theories and techniques can co-exist in the same overall system. 

Brill, E. Transformation-based error-driven learning and natural language processing A case study in part of speech tagging. Computational Linguistics (1995). 

Reichel, U. Improving Data Driven Part-of-Speech Tagging by Morphologic Knowledge Induction. Proc. Advances in Speech Technology AST (2005). 

Reichel, U. Improving data driven part-of-speech tagging by morphologic knowledge induction. In Proceedings of Advances in Speech Technology (2005). 

---