Main Article Content
Gevorderde akoestiese korrelate van Afrikaanse klemtoon
Abstract
Opsomming: Hierdie studie is 'n vervolg op Wissing (2007). Waar genoemde artikel die tradisionele klemtoonkenmerke, grondtoon, intensiteit, duur en vokaalkwaliteit ondersoek, fokus die huidige artikel op 'n aantal minder bekende, en ook onbekende parameters. Die minder bekendes is ontleen aan die invloedryke studie van Sluijter en Van Heuven (1996a). Sommige van die onbekendes is geformuleer op grond van dié wat genoemde outeurs ingevoer het; ander is afgelei van meer algemene beskrywings van die fisiologiese proses wat plaasvind tydens fonasie, en, meer spesifiek, beklemtoning. 'n Voorbeeld van eersgenoemde groep is die parameter F0-gradiënt, wat 'n meetbare operasionalisering is van hulle algemene verwysing na F0-beweging. 'n Voorbeeld van die tweede groep is 'n uitbreiding hiervan na intensiteit en die twee vokaalformante. 'n Ander groep parameters wat dusver nog nie aandag gekry het nie, is die begin- en eindpunte van hierdie bewegings. Met die oog op 'n optimale vergelyking van resultate is die empiriese metode van Wissing (2007) ook in hierdie studie so presies moontlik gevolg.
Van die belangrikste resultate sluit in die ondersteuning van een van Sluijter en Van Heuven 1996a)se hoofbevindinge, naamlik dat 'n onderskeid gemaak moet word tussen linguistiese klemtoon en sinsaksent. Hierdie twee kontekste is as sodanig in die huidige studie geformaliseer en gebruik. Laasgenoemde dien ter bevordering van klemtoon: wanneer 'n beklemtoonde sillabe bowendien in sinsaksentposisie staan, word die metinge ten opsigte van meeste parameters betekenisvol groter of hoër in beklemtoonde vokale se gevalle. Genoemde outeurs het bevind dat die intensiteit in 'n beklemtoonde vokaalspektrum tussen 2 kHz en 5 kHz 'n beter aanduider van klemtoon is as die intensiteit van die totale spektrum(dus van 0 kHz tot 5 kHz). Dit word ten opsigte van die Afrikaanse /A/-vokaal gestaaf. Die resultate weerspreek in 'n opsig die bevinding vir Nederlands deur Sluijter en Van Heuven(1996a) dat F0-beweging aksentverlenend is, en slegs in sinaksentkonteks as sodanig optree. Sommige van die nuwere parameters is in die besonder kragtig om die fonologiese proses, vokaalreduksie, mee te beskryf. Onbeklemtoonde /A/ toon 'n baie duidelike neiging in die rigting van die schwa-vokaal (/@/), veral soos gesien kan word aan die beduidende verskille wat gevind is tussen beklemtoonde vokale en onbeklemtoonde vokale by die parameters wat verwys na die verloop van die formante (m.n. F1) – dit is die gradiënt, maar, meer nog, die eindpunt van F1.
Harmonisiteit-tot-Geraas se verhouding, 'n parameter waarop toevallig afgekom is, blyk'n interessante een te wees wanneer beklemtoonde vokale en onbeklemtoonde vokale met mekaar vergelyk word. Hoewel die verskille nie so groot soos die meeste van die ander parameters is nie, is dit nogtans in byna al die kontekste waarin dit gemeet is van beduidende belang. Tot 'n geringer mate
geld hierdie bevinding ook die bandwydtes van die eerste drie vokaalformante.
'n Gekombineerde stel parameters, naamlik van die huidige ondersoek met die tradisionele ses van Wissing (2007), wat dan 'n stel bestaande uit 24 parameters tot gevolg het, lei tot 'n belangrike verbetering in die mate van sukses by die onderskeid tussen beklemtoonde en onbeklemtoonde vokale.
Abstract: In Wissing (2007) an investigation was conducted into the traditional stress parameters
of Afrikaans (fundamental frequency, intensity, duration and vowel quality). In the present study the focus is on a number of less well-known, as well as some totally new parameters. The former set of parameters was taken over from the influential work of Sluijter and Van Heuven (1996a) on Dutch stress. The rest are deducted from general descriptions of the process of phonation, in particular during the stressing of syllables/vowels. An example of the former group of parameters is F0 gradient. Sluijter and Van Heuven (1996a) describe this in terms of F0 movement. The latter refers to parameters such as the starting and end points of such F0 gradient, as well as those of intensity and vowel formant moves.
Results reported by Sluijter and Van Heuven (1996a) in the case of Dutch are contradicted by the current finding that accent-lending F0 movement is restricted to sentence accent contexts. Important results include the support of one of the main findings of Sluijter and Van Heuven (1996a), namely the necessity of taking into account the specific word and sentence contexts when analysing stress parameters. Linguistic stress has to be distinguished from sentence accent. A specific property of accent is that it normally enhances the manifestation of stress. This is true for most of the acoustic correlates under scrutiny in this study. Sluijter and Van Heuven (1996a) found that BdF_2, that is intensity located in the spectral region of 2 kHz to 5 kHz , proved to be a better indicator of stress than overall intensity (i.e. the whole spectrum, from 0 Hz to 5000 Hz). This finding of theirs was supported in the case of Afrikaans /A/. Of the new parameters, the gradient of F1, and in particular the end point of F1, seem to be powerful instruments in describing vowel reduction as a phonological process. As a result, unstressed /A/ tends to shift in the direction of schwa (/@/). This is evident from the F1 readings (cf. Wissing, 2007), but even more so, from the gradient and end point of F1.
Harmonicity-to-Noise Ratio is a parameter we stumbled on by accident. Although this is not one of the most powerful parameters in comparing stressed with unstressed vowels, in most contexts the difference found in such comparison turned out to be of practical significance. The same goes, albeit to a much lesser degree, for the bandwidths of the first three vowel formants.
When all of the parameters considered in both studies on the acoustic correlates of stress in Afrikaans were pooled in one comprehensive cluster — consisting of 24 parameters in all — the success rate was lifted considerably. It is recommended that, in the realm of speech technology, such a total cluster should be used for detection of stressed vowels. This is particularly the case with grapheme-to-phoneme converters.
Southern African Linguistics and Applied Language Studies 2007, 25(4): 607–624
Van die belangrikste resultate sluit in die ondersteuning van een van Sluijter en Van Heuven 1996a)se hoofbevindinge, naamlik dat 'n onderskeid gemaak moet word tussen linguistiese klemtoon en sinsaksent. Hierdie twee kontekste is as sodanig in die huidige studie geformaliseer en gebruik. Laasgenoemde dien ter bevordering van klemtoon: wanneer 'n beklemtoonde sillabe bowendien in sinsaksentposisie staan, word die metinge ten opsigte van meeste parameters betekenisvol groter of hoër in beklemtoonde vokale se gevalle. Genoemde outeurs het bevind dat die intensiteit in 'n beklemtoonde vokaalspektrum tussen 2 kHz en 5 kHz 'n beter aanduider van klemtoon is as die intensiteit van die totale spektrum(dus van 0 kHz tot 5 kHz). Dit word ten opsigte van die Afrikaanse /A/-vokaal gestaaf. Die resultate weerspreek in 'n opsig die bevinding vir Nederlands deur Sluijter en Van Heuven(1996a) dat F0-beweging aksentverlenend is, en slegs in sinaksentkonteks as sodanig optree. Sommige van die nuwere parameters is in die besonder kragtig om die fonologiese proses, vokaalreduksie, mee te beskryf. Onbeklemtoonde /A/ toon 'n baie duidelike neiging in die rigting van die schwa-vokaal (/@/), veral soos gesien kan word aan die beduidende verskille wat gevind is tussen beklemtoonde vokale en onbeklemtoonde vokale by die parameters wat verwys na die verloop van die formante (m.n. F1) – dit is die gradiënt, maar, meer nog, die eindpunt van F1.
Harmonisiteit-tot-Geraas se verhouding, 'n parameter waarop toevallig afgekom is, blyk'n interessante een te wees wanneer beklemtoonde vokale en onbeklemtoonde vokale met mekaar vergelyk word. Hoewel die verskille nie so groot soos die meeste van die ander parameters is nie, is dit nogtans in byna al die kontekste waarin dit gemeet is van beduidende belang. Tot 'n geringer mate
geld hierdie bevinding ook die bandwydtes van die eerste drie vokaalformante.
'n Gekombineerde stel parameters, naamlik van die huidige ondersoek met die tradisionele ses van Wissing (2007), wat dan 'n stel bestaande uit 24 parameters tot gevolg het, lei tot 'n belangrike verbetering in die mate van sukses by die onderskeid tussen beklemtoonde en onbeklemtoonde vokale.
Abstract: In Wissing (2007) an investigation was conducted into the traditional stress parameters
of Afrikaans (fundamental frequency, intensity, duration and vowel quality). In the present study the focus is on a number of less well-known, as well as some totally new parameters. The former set of parameters was taken over from the influential work of Sluijter and Van Heuven (1996a) on Dutch stress. The rest are deducted from general descriptions of the process of phonation, in particular during the stressing of syllables/vowels. An example of the former group of parameters is F0 gradient. Sluijter and Van Heuven (1996a) describe this in terms of F0 movement. The latter refers to parameters such as the starting and end points of such F0 gradient, as well as those of intensity and vowel formant moves.
Results reported by Sluijter and Van Heuven (1996a) in the case of Dutch are contradicted by the current finding that accent-lending F0 movement is restricted to sentence accent contexts. Important results include the support of one of the main findings of Sluijter and Van Heuven (1996a), namely the necessity of taking into account the specific word and sentence contexts when analysing stress parameters. Linguistic stress has to be distinguished from sentence accent. A specific property of accent is that it normally enhances the manifestation of stress. This is true for most of the acoustic correlates under scrutiny in this study. Sluijter and Van Heuven (1996a) found that BdF_2, that is intensity located in the spectral region of 2 kHz to 5 kHz , proved to be a better indicator of stress than overall intensity (i.e. the whole spectrum, from 0 Hz to 5000 Hz). This finding of theirs was supported in the case of Afrikaans /A/. Of the new parameters, the gradient of F1, and in particular the end point of F1, seem to be powerful instruments in describing vowel reduction as a phonological process. As a result, unstressed /A/ tends to shift in the direction of schwa (/@/). This is evident from the F1 readings (cf. Wissing, 2007), but even more so, from the gradient and end point of F1.
Harmonicity-to-Noise Ratio is a parameter we stumbled on by accident. Although this is not one of the most powerful parameters in comparing stressed with unstressed vowels, in most contexts the difference found in such comparison turned out to be of practical significance. The same goes, albeit to a much lesser degree, for the bandwidths of the first three vowel formants.
When all of the parameters considered in both studies on the acoustic correlates of stress in Afrikaans were pooled in one comprehensive cluster — consisting of 24 parameters in all — the success rate was lifted considerably. It is recommended that, in the realm of speech technology, such a total cluster should be used for detection of stressed vowels. This is particularly the case with grapheme-to-phoneme converters.
Southern African Linguistics and Applied Language Studies 2007, 25(4): 607–624
