Lexicon quality

A unique lexicon of descriptors for tea has been built up over the years in the tea trade. By its use, the tea-buyer and the tea-taster communicate quality parameters to guide the purchase, blending, and quality control of tea. Some terms in common use include brisk, bright, dull, flat, green, harsh, plain, pungent, quality, and soft.91... 

Since pollution prevention first entered the environmental lexicon in the 1980s, it has become a touchstone of U.S. environmental policy. To succeed, pollution prevention programs need sustained top management support and defined links to the company s core business activities. Equally important, the values represented in the program must fit with the academic institution, company, or individual s existing values. For example, syntheses that minimize wastes are environmentally Mendly and may provide quality improvements and cost and energy savings. Environmental consciousness should be incorporated early into chemical research (Lnnt and Bowen, 1996). 

Unacceptable starting materials used in research must be replaced by those which meet the regulatory quality and safety criteria. Reference to standard volumes of pharmaceutical ingredients and excipients should be made to decide whether a certain ingredient can be used. Standards are described in the pharmacopoeias of relevant countries, the "Handbook of Pharmaceutical Excipients" or its Swiss anchestor the "Katalog pharmazeutischer Hilfsstoffe". Fiedler s "Lexikon der Hilfsstoffe" (Lexicon of Excipients) provides many useful informations about the safety and about applications for excipients. (For references see Annex B to the registration requirements chapter.)... 

Today s chemist, if he wishes to gain an over-all and general view of the literature of chemistry, to learn about a specific class of chemical literature, such as abstract journals, monograph series, formula lexicons, or compilations, or to inquire into the history, scope, or quality of a specific source of chemical information, is fortunate in having at his call a number of basic books in this special field (5, 19,21 2Ji),... 

International Society for Pharmacoeconomics and Outcomes Research, ISPOR Lexicon Pashos, Klein, Wanke, L.A., Eds. ISPOR New Jersey, 1998. Spilker, B. In Quality of Life and Pharmacoeconomics in Clinical Trials 2nd Ed. Spilker, B., Eds. Lippincott-Raven Philadelphia, PA, 1996. 

Now we turn to some more problematic cases. In English, the consonants /dh/, /zh/, /ng/ and /h/ are a little more difficult to deal with. There are clearly exactly four unvoiced fricatives /th/, /f/, /s/ and /sh/ in English and of these, /f/ and /s/ have the voiced equivalents /v/ and /z/, and these are unproblematic. The voiced equivalent of /th/ is /dh/ and while /th/ is a common enough phoneme, /dh/ only occurs in very particular patterns. For a start, the number of minimal pairs that occur between /dh/ and /th/ are few real examples include teeth noun, /1 iy th/ and teeth.verb, /t iy dh/, while most are near minimal pairs such as bath /b ae th/ and bathe / b ey dh/ where the quality of the vowel differs. Even if we except /dh/ as a phoneme on this basis, it occurs in strange lexicon patterns. In an analysis of one dictionary, it occurred in only about 100 words out of a total of 25,000. But if we take token count into consideration, we find in fact that it is one of the most common phonemes of all in that it occurs in some of the most common function words... 

In both the TIMIT and MRPA phoneme inventories, we have only used lower case letters. To some extent this is a purely aesthetic issue as we wish to avoid upper case (which has a somewhat aggressive quality) and mixed case (which can look ugly). There is an additional benefit in that upper case equivalents of the symbols can be used in cases where some emphasis is required, such as stress or other highlighting. In the above, we have chosen though to give simple mnemonic names to each phoneme, so that the first phoneme of tip is named t , the first of ship is named sh and so on. Mnemonic names are usually helpful as we often have to enter words in the lexicon ourselves or perform transcriptions, and this is easier if we can easily recall the symbols to use. These names can sometimes trip us up, especially with vowels, so a little care should be taken. 

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 ... 

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... 

---