Tristan McLeay wrote: > I think the term was popularised around the time I made the CXS chart. Looking at the chart it's obvious not everyone agrees on stress marking :-) Personally, I think comma for secondary stress is by far the most logical of the three options. It's very counterintuitive to have a "heavier" character to mark secondary stress than to mark primary stress ("heavier" = "takes more ink to draw"). Mind you, I've never understood why a retroflex trill is not shaded impossible. If it's retroflex then the tip of your tongue points backwards, and you can't trill like that. *** Changing topic quite a bit, but still talking about phonetics, I thought people might like to see how I went about explaining phonetics in my research assignment last year. You see, I find the process of how people choose to explain things and why to be quite fascinating. Therefore I assume that many other people find it interesting, too (it's definitely one reason why lots of scientists enjoy reading good popular science books). In this case, the relevant background is that I, a non-linguist, needed to explain phonetics and phonology to a lecturer, also a non-linguist, with the emphasis being on the dimensions within which the data exists (very important in DMKD). In practical terms, consider this a review of the fundamentals and an opportunity to pester me for fun if there's some obscure and pedantic error :-) 3 Phonetics and Phonology The applications of data mining to phonetics and phonology cannot be discussed without a brief overview of those fields. Phonetics is a study of speech sounds (phones) that emphasises their means of articulation and corresponding acoustic properties, while phonology is a more abstract study of speech sounds as they are utilised by language. 3.1 Phonetics Consonants and vowels are categorised along different sets of dimensions, and can be underestood from either an articulatory or an acoustic perspective. The characters of the International Phonetic Alphabet correspond to distinct phones, but with limited resolution. For obvious reasons, there is a tradeoff between the accuracy and resolution of acoustic analysis and the size of the database that it is viable to constuct (see later), and since improvement of these qualities is an ongoing research area, we can expect the phonetic analysis database of the future to be larger and more exact than what is possible today, increasing the applicability of data mining. 3.1.1 Consonants Consonants generally have much more complex acoustic properties than vowels because many of them involve the production of turbulent air flow. The author has little knowledge about acoustic analysis of consonants, but papers such as de Mareuil (1999) contain keywords that might aid an interesting reader in learning more. The articulatory dimensions along which a consonant can vary have multiple continuous and discrete components, but the standard IPA transcriptions capture a simplified description that suffices for most purposes (Catford 2001). For example the description of 's' as an unvoiced aveolar fricative refers to three discrete-valued dimensions ('unvoiced' meaning that the vocal chords are not vibrated, 'aveolar' meaning that the sound is produced near the ridge behind the upper teeth, and 'fricative' meaning that turbulence is created by forcing air through a narrow channel) but this description discretises some dimensions that are really continuous (e.g. the distance between the tongue and the teeth) and ignores other dimensions altogether (e.g. the part of the tongue used to make the sound: the distinction between apico- lamino- and dorso- articulations). 3.1.2 Vowels Vowels can be described using either acoustic dimensions or articulatory ones, corresponding to low-level or high-level descriptions of the vowel respectively. Again, IPA symbols offer only a rough approximation, but a higher resolution is attainable in which continuous-valued dimensions are indeed treated as continuous. The articulatory description of a vowel includes two major dimensions and several minor dimensions (Catford 2001). The major dimensions concern the horizontal and vertical position of the tongue, distinguishing between front vowels (as in 'head'), back vowels (as in 'hoard'), closed vowels (as in 'hoot') and open vowels (as in 'heart'). A map of these dimensions is often referred to as the vowel diagram, and is well illustrated with respect to several dialects in Mannell: Vowels. The most important of the minor dimensions is the extent to which the lips are rounded (roundedness), another is the extent to which air passes through the nose as well as the mouth (nasalisation) and there are more besides. The continuous nature of the major dimensions is very important - which is why the vowel diagram exists as a standard illustration - while the continuous nature of the minor dimensions is less important, and they are conventionally described in either/or terms (the lips are either rounded or not; the vowel is either nasalised or not). The low-level (acoustic) dimensions of vowels are defined by formant frequencies, which are the frequencies in the acoustic spectrum of the sound that are amplified by a particular configuration of the mputh and throat (Catford 2001). Formants are numbered in ascending order of frequency, each one being a continuous-valued dimension. For example the first formant of an 'ee'-like sound is around 250 Hz, while that of an 'ah'-like sound is around 750 Hz. Likewise the second formants are around 2300 and 1400 Hz respectively. The number of formants that must be recorded depends depends upon the range of contrasts that may affect the vowel, for example all else being equal the primary dimensions of a vowel can be described using only the first two formants but if lip rounding etc are not constant then two formants are insufficient. 3.1.3 Temporal Dimensions The dimensions discussed above are defined in abstract articulatory or acoustic spaces. The temporal dimension is also important, as it allows us to capture information about the length for which a phone is sustained, the way that its articulation is influenced by its neighbours, and the frequency at which different phones appear together. The importance of such informaiton is well established and studied. For one thing, languages make direct use of it. The words 'hut' and 'heart', for example, are distinguised only by the duration of the 'ah' sound. Diphthongs and affricates are examples where neighbouring phones are best understood as a unit - in the 'eye' diphthong an 'ah' sound glides toweards an 'ee' sound,, and in the 'ch' affricate a 't' sound gives way to a 'sh' sound (Catford 2001). For another thing, people have a tendency to modify a sound in order to assist the flow of the word, for example the 'h' in 'huge' is objectively much closer to the 'ch' in German 'ich' than the 'h' in 'hunt'. The related phenomenon of assimilation means that neighbouring phones often acquire the same point of articulation, for example if there was a word 'sumda' then it would quite likely mutate to 'sunda' or 'sumba' (and the word 'import' is derived from Latin 'in' + 'port'). For these reasons the fact that phonetic data exists in a spatiotemporal context is certainly important. 3.2 Phonology A phone is a sound in language viewed from a low-level perspective, while a phoneme is a sound viewed from a higher-level perspective. The mechanics of spoken languages are not concerned with details of articulation, but only with the fact that units of sound can be put together to make words (Catford 2001). For example the difference between the 'h' in 'huge' and 'hunt' was mentioned above, yet from the point of view of how English works they are nevertheless the same unit. For this reason, the articulations of 'h' are said to be different /allphones/ of the same /phoneme/. Similarly the distinct 'oo' sounds in 'food' and 'fool' are allophones of another phoneme, the latter being modified by the following 'l'. It should be apparent that the 'ah' sounds in 'heart' and 'hut' are distinct phonemes, whereas the 'oo' sounds in 'feud' and 'refute' are the same phoneme, because English doesn't care that one is longer than the other in that case (vowels are typically shortened in English when followed by an unvoiced consnant). Diphthongs and affricates etc are single phonemes, because they are single units in language. 3.3. Sound Change It is well established that the dominant pronunciation of words changes over time, and that this process eventually leads to new dialects and even languages. A driving force behind this process is the tendency people have to imitate the speech of people who are socially important to them (peers and role models). Children consistently adapt the accent of their peers and not the accent of their parents. This process explains why pronunciation changes are sometimes detectable from one generation to the next. Sound change and similar processes lead eventually to language diversion, for example Italian and Spanish are both descendents of Latin with many similarities remaining in their phonology, vocabulary and grammar. Whether data mining can be used to form hypotheses about which languages are related to which will be discussed later. 3.4 Size and Resolution of Phonetic Databases <snip> References ---------- J. C. Catford: /A Practical Introduction to Phonetics: Second Edition (2001). Oxford University Press. R. Mannell: /The Vowels of Australian English and Other English Dialects/. Website: www.ling.mq.edu.au/units/ling210-901/phonetics/ ausenglish/auseng_vowels.html (part of linguistics website at Macquarie University). Accessed 2003. P. B. de Mareuil, C. Corredor-Ardoy & M. Adda-Decker: /Multi-Lingual Automated Phoneme Clustering/ (1999). ICPh599 San Francisco. Search Internet for title to find this third reference. Adrian.