The offline lexicon

When we consider the offline lexicon, it is not the ease that a given word is strietly in or not in the lexicon. For instance, we may have entries that are incomplete, in that we know the values for some fields but not others. Most likely there will be missing words also, that is words that we know of but whieh we haven t entered. Not every value in every field will be correct errors may occin or there may be entries which we are not entirely eertain of More formally, we can evaluate om off-line lexieon in terms of coverage, completeness, and quality  

Coverage This is a measure of how many words in the language are in the lexicon. For every 

In this example we have seen how our knowledge of missing words (that is words known to us, but not in the lexicon) can be used to improve coverage. We have also seen that we can use rules to improve the completeness of the lexieon, but in doing so we don t necessarily know if the rule 

For every word in the language, we can assign it to one of three categories  

Completeness Every lexicon entry has a number of fields, but some of these may 

---

If we generate the offline lexicon by the above process we will have a lexicon that is very large, but will have a considerable number of entries whose quality is unknown. In traditional TTS, the normal approach was to only have entries which were carefiilly checked in the lexicon, with the idea that the rules would be used for all other cases. However, it should now be clear that whether we use the rules at run-time to handle words not in the lexicon, or use the rules offline to expand the lexicon will not have any effect on the overall quality all we are doing is changing the place where the rules are used. 

Given this large offline lexicon, we see then that the real debate about rules vs lexicons is not one of quality, but rather one of balance between run-time and off-line resources. If we take the case where we include the entire off-line lexicon in the system lexicon, we will have a system which uses a considerable amount of memory, but where the processing speed is minimal (simply the small amount of time taken to look up a word). If on the other hand we create a system lexicon that is only a small subset of the offline lexicon, this will result in a smaller footprint, but as the pronunciation of absent words will have to generate at run-time, the processing costs... 

---