Professional Documents
Culture Documents
1
1 Introduction
2.1 Phonetic and phonological literature is replete with the discussion of [h] in different
languages from a variety of points of view- its production, its perception, its phonological
representation in terms of distinctive features, its asymmetrical behaviour in phonological
1
The IPA symbol is enclosed in [ ] or / / bracket to suggest, as is the usual practice, its status as a phone or a
phoneme, respectively.
2
contrast patterns, its role in diachrony and its sociolinguistic significance among others. We
take up some of these below with the purpose of showing the significance of the present
investigation to the questions raised above.
2.2 The phonetics of [h]
The phonetics of [h] is of interest on two main grounds. The first is its defintion. The IPA
includes it in the chart for pulmonic consonants as a voiceless glottal fricative. Ladefoged
(1982) argues that [h] is not a glottal sound, and defines it as a voiceless counterpart of the
following sound. Keating (1988) shows that this feature of [h] may be present in some
utterance-initial cases, but elsewhere it must be assumed to be underspecified for a feature, as
in those positions its formants are found to be interpolated between two vowels. Keating
shows [h] to be an exemplar case of underspecification of segment-internal features. On these
grounds, Ladefoged (1990) further argues that [h] is not a fricative as it has very little
friction at the glottis. The vocal cords are apart and any turbulent airflow that there might be
is due to "cavity friction" rather than local friction at a particular point. [h] has no more
friction at the glottis than [f] or any other voiceless sound with a comparable airflow. (1990:
24). Ladefoged proposes that [h] must be treated as a voiceless approximant. On similar
grounds, [] should be treated as breathy voiced approximant. Catford (1990: 25), on the
contrary, suggests that ...[h] and [] stay where they are as glottal fricativeshonorary
glottal fricatives, if you like!.. My reason for wanting to keep [h] and [] in the chart is
perhaps more phonological and practical than strictly phonetic. There are so many languages
in which one or both of these sounds functions as a syllabic margin
Thus, as far as the label for the sounds repesented as [h] and [] is concerned, we are in a
conflicting situation. They should either be changed to approximants or kept as glottal
fricatives.
The second ground for interest in the phonetics of [h] and and [] is the use of the
distinctive feature(s) for them, based on their phonetic properties. The earliest attempt at
assigning a distinctive feature to them was made by Chomsky and Halle (1968). They gave
an aerodynamic account of the sounds, with specific reference to the aspirates in Hindi, and
assigned them the feature Heightened Subglottal Air-pressure (HSAP), which means an
accompanying oral constriction and a momentary lowering of air-pressure. Ohala (1972)
argued against the consistent presence of an accompanying oral constriction and a momentary
lowering of air-pressure in the production of aspirated consonants. Kim (1970) in a study of
Korean sounds proposed the feature spread glottis to characterize aspirates. The proposal was
3
taken up, as recounted in Ridouane et al. (2011), by Halle & Stevens, who assigned the
feature [+spread glottis, -constricted glottis] to aspirated consonants. They also used the
features [stiff vocal cords] and [slack vocal cords] to distinguish voiceless and voiced
aspirated plosives, respectively. Voiceless aspirated plosives, e.g. [p p], were proposed to be
[+stiff vocal cords, -slack vocal cords], and voiced aspirated plosives, e.g. [b b], to be [-stiff
vocal cords, +slack vocal cords]. The evidence for all these features to be used for aspirates is
not quite strong, specifically for the latter set of features, namely, [stiff vocal cords] and
[slack vocal cords]. The present data reveal that [b] (sic), in accord with the Halle-Stevens
proposal, is correctly considered to have slack vocal cords and that [p] has spread vocal
cords; however there is no evidence that (nor is it necessary to assume that) [ph] has stiff
vocal cords and [bh] (sic) has spread vocal cords. However, more experimental data rather
than mere speculation are needed to settle this point. (Ohala & Ohala 1972: 45). In later
investigations into the articulatory properties of aspiration, Dixit (1982, 1987) and Davis
(1994) claimed that aspiration should be defined in terms of a period of turbulent noise that
accompanies [+spread glottis]. Ridouane et al. (2011) further support this claim:
aspiration noise, i.e. aperiodic energy in the second and higher formants is a necessary
component of the definition of [+spread glottis].
Lisker & Abramson (1964) proposed the articulatory and acoustic feature "Voice Onset
Time (VOT) for aspirated consonants. The feature has been found to be useful for voiceless
aspirated stops, but not for voiced aspirated stops.
An early study of aspiration, Dixit and MacNeilage (1974), investigated the articulation of
/p/, /ph/, /b/, /b/, and/h/ in Hindi and came to the following conclusions: ... (1) the opening
gesture of the glottis is a necessary condition for the production of the glottal fricative and
unaspirated plosives, both voiced and unvoiced; (2) unaspirated plosives, both voiced and
unvoiced, may be produced with or without opening gesture of the glottis; (3) regardless of
aspiration, voiced plosives are produced with uninterrupted vocal-fold vibrations, whereas
unvoiced plosives are not; (4) glottal fricative is produced without interruption of vocal-fold
vibrations; (5) duration and extent of glottal opening do not seem to be systematically related;
(6) there is no apparent direct correlation between the degree of glottal opening at the time of
oral release and the degree of aspiration; (7) the initiation of opening and closing gestures of
the glottis, in relation to oral closure and release, respectively, is variously timed for various
4
plosives. The study showed clearly that the feature [+spread glottis] was common to
aspirated and unaspirated plosives as well as fricatives.
Vaux (1998) notes that voiceless fricatives, in particular, but not voiced fricatives, must
be specified as [+spread glottis]. This general use of the feature [+spread glottis] is unable to
specifically distinguish aspirated from unaspirated plosives. The feature applies correctly to
aspirated consonants, but not exhaustively, as it also applies to voiceless fricatives. There is
thus need for another feature that, in addition to [+spread glottis], distinguishes aspiration
from other consonants produced with open glottis.
In this context, it is interesting to note that Sanskrit phonetics dealt with the issue of
distinguishing the different types of aspiration. For example, the phonetic treatise Rk
Prtishkhya appended to the Rgveda used the term mahprNa meaning great
breathforce or extra breathforce. The common term used for breath is also shw sa. Voiced
aspirates were saghosh mahprNa voiced extra-breathforce (see Allen (1953), Busetto
(2003) for modern treatments of Sanskrit phonetics). Given the fact that the feature spread
glottis applies non-exhaustively to aspiration, the feature extra breathforce can be added as
a distinctive feature to exhaustively apply to aspiration ([ph], [b] and [h] or []). The term
has been in use in the British approach to the study of aspiration (see e.g. Roach 1983). The
present proposal is in consonance with Clements (2005), who, following Chomsky and Halle
(1968), argues for a focus on articulatory features rather than acoustic features on the grounds
(see also Vaux 2005) that an articulatory feature may have multiple acoustic manifestations.
5
1998 for a detailed account) is the theory of the explanation of historical change in Indo-
European languages from the reconstructed Proto-Indo-European (PIE). In brief, the theory
posits three phonemes h, h and h in PIE, also known as the "neutral" laryngeal, the "a-
coloring" laryngeal and the "o-coloring" laryngeal, respectively. Whereas the neutral
laryngeal of PIE remained a laryngeal consonant in Indo-European languages, the other two
laryngeals merged with neighbouring vowels, yielding vowels that lacked in regular
correspondence with the vowels of PIE. The theory was first proposed by Saussure (1879)
and elaborated much later in Mller (1970 [1911]). The different variants of PIE /h/ are called
laryngeals as they are assumed to have been produced in the area of the larynx.
6
phoneme // as [] or [e] in the context of following [h]. The context is a little more
complex- the following /h/ is not followed by any vowel other than //. Thus the words
/bhn/ sister and /bhta:/ flow-3PER-PRES-SING are pronounced as [bhn] ~
[bhn]~ [behen] and [bhta] or [behta:]. The word /bha:na/ excuse, however, is
pronounced as [bha:na], not *[bha:na:]. A fronted schwa is a sociolinguistic variable of
Standard Hindi speech, as noted in Pandey (1991). It has been found to be on the rise in the
speech of speakers of eastern standard Hindi (see also Mohan (2009).
3.2 Method
Four female subjects, all native speakers of Hindi, were recruited for the experiment. They all
belonged to the age group 22- 25 years, and were graduate students pursuing M. A.
Linguistics studies at Jawaharlal Nehru University. Four of the students were born and
brought up in Delhi and four hailed from the State of Bihar. The two groups were selected in
order to compare the relation of /h/ with the surrounding vowels, noted for the difference
between the two varieties (e.g. Shapiro 2003). All subjects reported normal hearing. They
offered to be part of the experiment without receiving any payment.
The participants were asked to read three stories.The stories were first given to them in
typed form and then presented on a 14 inch computer screen. They were advised to read them
naturally at a normal tempo. Each story contained /h/ in different environments in words. The
data were collected through digital recording in a sound proof recording studio using Edirole-
07 high quality digital recorder with a sampling frequency of 44 KHz 16 bit wave form. The
microphone of the recorder was kept at an equal distance of 5 cm from the mouth of each
speaker in order to get controlled data. The acoustic data were edited and analyzed using the
techniques in Praat speech software.
7
3.2 Factors in the selection of words
The following factors determined the selection of the words.
a. /h/ is clearly audible in the word-initial position and as onset in a stressed syllable, e.g.
b. /h/ is weakly audible in the onset position in unstressed syllables and word-finally,
e.g. /bhn/ (stressed on the first syllable) sister, /h/ he.
c. Hindi has a general process of schwa deletion in unstressed syllables (Pandey 1990),
e.g. /kmla:/ [ kmla:] (a name), /la:pta:/ [ la:pta:] lost, but not /kbi:/
[kbi:](a sport). It has been found by the present author that although the schwa
following a /h/ may be heard to be deleted, in speech synthesis based on trained data, the
deletion of the schwa leads to problems2. Alongwith the schwa, /h/ too gets dropped. The
acoustic features of /h/ in relation to the presence of flanking vowels and a following
consonant after the deletion of the following schwa need examining.
e. Standard Hindi shows the process of schwa fronting before a /h/ followed by a
consonant (in turn optionally followed by a schwa, also fronted), e.g. /hna:/ [hna
:]
jewelry and /hr/[hr] city, as discussed in section 2.4.
Taking the above considerations into account, a total number of 35 words including
content words (nouns, verbs, adjectives) and grammatical words (pronouns, auxiliary verbs)
were recorded. Out of these, 28 words were chosen for analysis as they were found to be
representative of all the contexts. Thus, the total number of tokens analyzed for the data were
22x8 =176.
There are two clear manifestations of /h/ in the data- as a fricative and as an approximant.
The approximant realization of /h/ is in turn of two types- one dividable with flanking
vowels, the other merged with the vowels. The difference between the fricativized
approximant and the merged approximant was determined visually from the spectrograms
and oscillograms.
Murthy says (personal communication) that the Hindi speech synthesis programme carried out by her and
2
colleagues, statistical parametric synthesizers are used in order to take care of the issue relating to the
difficulty in slicing off /h/ as well as some other issues based on the context. See
http://www.iitm.ac.in/donlab/website_files/research/Speech/TTS/contents/main.html
8
3.3 The data
The data consisted of the words listed in the contexts below, taking into account the factors
affecting the behaviour of /h/.
9
/trh/ [trh] as, like
/h/ [h] place
/kh/ [keh] or [kh] say-IMP
/h/ [eh] or [h] 3PER-SING
4. Discussion
4.1 As mentioned above, the data for experimental analysis in the present paper are from
speakers from two regional sources represented by Delhi and Bihar. For each region, both
male and female speakers were selected. The discussion below revolves around the two
variants of /h/- voiceless aspirated fricative [h] and voiced approximant []. The discussion
is organized around the following points:
Fricative /h/
Approximant []
Intervocalic stressed
Intervocalic unstressed
Following schwa deletion
Before obstruents
Before sonorants
In the end, the the gender and regional differences are also discussed.
In all instances, where /h/ occurs word-initially (examples in 2-1), it has the properties of a
voiceless aspirated fricative. Look at the following graphs for word-initial fricative [h].
10
Graph-1: Oscillogram and spectrogram of word-initial /h/ in /ha:thi/
The oscillograms of /h/ compared to the following vowel are clearly indicative of the fricative
nature of /h/ in this context/
A fricative, or rather a fricativized, [h] is also found foot-initially (examples in (2-2b).
Hindi feet are moraic trochees (see Pandey 1989, Hayes 1995), which means the first syllable
in a foot in Hindi is stressed. In this position, however, there are instances of both a
fricativized as well as an approximant /h/, as can be seen in Graph-3 and Graph-4 below,
both productions of a Delhi female speaker.
11
Graph-3: Foot- initial fricativized /h/
4.3 Approximant []
/h/ is ambiguous in realization foot initially. However, in all other contexts, including
unstressed intervocalic positions and coda positions, /h/ is realized predictably as an
approximant. The segment is perceived as voiced. An attempt at slicing /h/ in these positions
failed. In all instances, /h/ was inseparable from the surrounding vowels. This feature was
attested in cases where the following schwa is perceived as deleted, discussed in section 4.4
in detail.
As extracting /h/ in the above contexts was not possible, a strong case was found for
characterizing the phenomenon as segment merger. The features of /h/ merged with the
features of the neigbouring vowels in the following way: the aspiration of /h/ was overlaid on
the neighbouring vowels, and the formants of the neighbouring vowels were extended to /h/.
A relevant question that arose in the course of the investigation was, in the intervocalic
12
position which of the vowels had greater influence on /h/- the preceding or the following? In
order to find an answer to the question, the VhV sequence was sliced into three equal parts.
For objective results, the mdeial 75% portion was selected, with the beginning and ending
12.5% of the sequence kept out for measurement. It was possible to do so in PRAAT,
following the slicing devices available.
The acoustic consequences of the merger of aspiration in [h] with the surrounding vowels
can be read from the following tables:
Intervocalic (Unstressed) [h] formant figures
Without schwa fronted, the average figures show F1 and F2 to be closer to either the
following or the preceding vowel. The variation is gender-based, as will be seen in section
4.6 below.
The abbreviations used in the tables and graphs below have the following full forms:
BF= Bihar female, DF= Delhi Female, BM= Bihar male, DM= Delhi male,
AVG= Average, TAVG= Total average
3350
3100
2850
2600
2350 AVG.DF
2100
1850 AVG.BF
1600
1350 AVG.DM
1100 AVG.BM
850
600
350
100
F1 F2 F3 F1 F2 F3 F1 F2 F3
13
3350
3100
2850
2600
2350 AVG.DF
2100
1850 AVG.BF
1600
AVG.DM
1350
1100 AVG.BM
850
600
350
100
F1 F2 F3 F1 F2 F3 F1 F2 F3
Graph-6: Average of six words in Set 2(b)ii with identical flanking vowels around /h/
Schwa Deletion
In the context of schwa deletion noted in Hindi phonological literature (Ohala 1974,
Pandey 1990, Mohan 2014, among others), it came to be observed that in fact the following
schwa is merged with [h]. The tripartite division of the sequence V1-[h]-V2 shows that V2 is
not deleted, but in fact merged with []. In this context, [] is inseparable from the schwa
following it. Two contexts were selected for the consonant following the putative schwa
deletion- obstruents and sonorants. The figures are being produced below for both. Graph-7
presents the figures for the schwa preceding obstruents
3500.00
3250.00
3000.00
2750.00
2500.00
2250.00 TAVGDF
2000.00
1750.00 TAVGBF
1500.00 TAVGDM
1250.00
1000.00 TAVGBM
750.00
500.00
250.00
0.00
F1 F1 F3 F1 F2 F3 F1 F2 F3
Graph-7: Total Average of two words in (2)-b)-iii- /a:hte:/ [a:hte] and /khta:/ [khta:]
14
As the following schwa in these cases is merged with the preceding [[]], the merger can
only be noted. The IPA convention of transcription allows for representing the merger of []
with a vowel- [e]. The correct representation in these cases is either as [ee]or as [e]. In either
case, the second segment is an approximant without a place of articulation, rather than a
fricative. The segment takes its place of articulation from the surrounding vowels. The above
graph shows that in both its formants, the ranges in the first and the last part are closer to each
other than to the middle part, which is the approximant []
The figures for + sonorant sequences are not significantly different from the figures for
+obstruent sequences, as can be seen in the graph below. These figures show that the
approximant [] takes its features from an adjacent vowel and that there is no strict
directionality involved in the process of the merger of /h/.
3500.00
3000.00
2500.00
TAVGDF
2000.00
TAVGBF
1500.00
TAVGDM
1000.00
TAVGBM
500.00
0.00
F1 F1 F3 F1 F2 F3 F1 F2 F3
Graph-8: Total Average of 2 words in (2)-b)-iii, /khna:/ [kena:] and /phle:/ [pele:]
The approximant character of [] is attested for the word-final position occurrences, too.
Look at Graph-9 for its formant figures and the spectrograms and oscillograms in Graph-10
of the word /kh/ from set (2)-3) above.
15
3450.00
3200.00
2950.00
2700.00
2450.00 AVGDM
2200.00
1950.00 AVGBF
1700.00
1450.00 AVGDM
1200.00
AVGBM
950.00
700.00
450.00
200.00
F1 F2 F3 F1 F2 F3
Graph-10: oscillogram and spectrogram of /kh/ from a Delhi female and Delhi Male
16
medially. In this position, /h/ can also be pronounced as a fricative. Look at the table and the
spectrograms and oscillograms of a word in two productions.
4000
3750
3500
3250
3000
2750 AVGDF
2500
2250 AVGBF
2000
1750
1500 AVGDM
1250
1000
750 AVGBM
500
250
0
F1 F2 F3 F1 F2 F3 F1 F2 F3
17
It is interesting to observe that with regard to the acoustic features of /h/, there is no
significant regional difference found in the speech of the informants in spite of considerable
geographical distance between the varieties represented here. The only difference that was
noted was that the speakers of the eastern variety showed a greater tendency for a fricativized
/h/ in stressed syllable-initial positions than the speakers of the Delhi variety.
V1 H V2
2)-b-i
F1 F2 F3 F1 F2 F3 F1 F2 F3
/bhut/
AVG.DF 483 1119 2752 775 1255 3132 751 1499 2800
AVG.BF 518 1115 2671 662 1455 2778 639 1539 2959
AVG.DM 498 911 2441 502 897 2517 424 1156 2432
AVG.BM 535 881 2773 561 881 2785 493 1139 2676
2)-b-ii
AVG.DF 685 2232 3134 728 2308 3098 674 2103 2968
AVG.BF 562 2011 2968 639 2136 3189 583 1981 2879
AVG.DM 503 1822 2612 553 1909 2645 479 1688 2593
AVG.BM 510 1722 2666 550 1671 2711 543 1685 2673
2)-a- [suha:na]
AVGDF 488 1223 3467 839 1576 3161 722 1494 2101
AVGBF 442 1312 3232 701 1460 3021 752 1625 2338
AVGDM 466 931 2496 758 1484 2788 651 1308 2356
AVGBM 431 1101 2598 609 1097 2410 632 1241 2297
2)-a-
[nha:na:]
AVGDF 771 1848 2994 1165 1898 3283 752 1680 2344
AVGBF 785 1722 2197 1039 1687 2664 822 1716 2336
AVGDM 579 1517 2619 738 1321 2493 672 1386 2482
18
AVGBM 457 1646 2315 676 1354 2340 639 1295 2291
2)-a-[sahe:li]
AVGDF 546 1930 2486 370 2479 3045 437 2153 2844
AVGBF 534 2025 2778 423 2247 2952 466 2250 2984
AVGDM 465 1793 2563 478 1890 2564 418 1968 2562
AVGBM 479 1749 2626 450 1921 2641 414 1938 2648
2)-b-iii-before obstruents-
[a:hte:]
TAVGDF 522 2210 2836 582 2281 2876 533 2146 2818
TAVGBF 597 2034 2791 652 2087 2978 560 1974 2971
TAVGDM 481 1780 2451 543 1739 2476 462 1692 2629
TAVGBM 474 1539 2265 513 1760 2536 477 1653 2799
2)-b-iii-before sonorant-
[kena:]
TAVGDF 561 2274 2933 612 2335 2948 530 2143 2783
TAVGBF 610 1906 2572 631 2022 2786 570 2063 3073
TAVGDM 556 1845 2557 616 1848 2572 575 1834 2624
TAVGBM 474 1539 2265 513 1760 2536 477 1653 2799
Table-1: Formant values of intervocalic /h/ and the flanking vowels
The issue of gender difference in the formant realizations of /h/, however, needs further
examination.
4. 7 Conclusions
The main aspects of the acoustic properties of /h/ in Hindi can be stated as the following.
a. The schwa following /h/ is merged with it rather than deleted, unlike after other
consonants. This finding has implications for speech synthesis in Hindi. Schwa deletion does
not apply in this context.
b. The apparently contrasting views about /h/ being a voiceless glottal fricative and a
glottal approximant, as discussed in the beginning, has a resolution in the present
investigation of the consonant in Hindi. It is realized both as a voiceless glottal fricative as
well as an approximant. The fricative allophone is realized intially in words and occasionally
in stressed syllables and the approximant allophone is realized unequivocally syllable-finally
and intervocalically before an unstressed vowel.
On the whole, there is no directionality involved in the process of the merger of /h/ and an
adjacent vowel in Hindi, yielding []. There is evidence, however, about there being some
gender difference in this regard, with females showing rightward merger and males showing
leftward merger with adjacent vowel
19
5 Phonology of /h/
5.1 The fact that /h/ has both a fricative or fricativized and approximant realizations can be
explained by a combination of both phonetic and phonological facts. Phoneticaly, we have
seen that the fricative [h] is realized uniformaly in the speech of eastern and western standard
Hindi speakers word-initially, when it is not preceded by any other sound.
Aspiration is also realized with voiceless stops, as in the word /ha:thi:/. In some articulations,
especially in the speech of eastern standard Hindi, it is also realized in the stress-initial
position, as in [suha:na:]. This can be explained as owing to greater force of articulation of
5.2.1 Laryngeal segments are noted for restricted contrasts among them (e.g. Halle & Stevens
1971, Steriade 1997, Kehrein & Golston 2004). The following laryngeal contrasts are
reported in the literature:
(i). Voiceless glottal fricative /h/ versus voiceless glottal stop //- (common):
E. g. Khasi (Rabel 1961), Kurux (Ekka 1985), Mizo (Fanai 1992)
(ii). Voiced glottal fricative // versus Voiceless glottal stop //- (common):
E.g. Kuvi (Israel 1975), Limbu (van Driem 1987)
(iii). Voiceless glottal fricative /h/ versus Voiced glottal fricative //- (rare):
E.g. Lam (Sachnine1982, Ladefoged & Maddieson 1996), Khezha (Kapfo
2005)
(iv). Voiceless glottal stop // versus Voiced glottal stop / /- (doubtful):
E.g. Jingpho (Halle & Stevens 1971)
20
Of the four possible contrasts, (i) and (ii) are common, (iii) is rare, and (iv) is doubtful.
With regard to the contrast in (iv), Halle & Stevens (1971: 208f) argue that a distinction in
tone in Jingpho (Dai & Diehl 2003), a Tibeto-Burman language spoken in Myanmar, India
and China, is caused by voicing in adjacent consonants. However, on an alternative analysis,
the relation between laryngeal voicing and tone could be seen the other way round, that is, the
tones are causing the difference in laryngeal voicing. Unequivocal evidence for the contrast
between // and // is missing.
Languages which contrast between /h/ and // are rare. The following languages have
been noted (in all five, to the best of my knowledge) in the literature to have a /h/ versus //
contrast: Lam, (Ladefoged & Maddieson 1996), Zulu (Traill et al. 1987), SiSwati (Bradshaw
1999), Musey (Shryock 1995), Wu (Cao & Maddieson 1992) and Khezha (Kapfo 2005, see
also http://www.ciil-ebooks.net/html/khezha/phonology.htm). The last one, Khezha is spoken
in the State of Nagaland in northeast India. The syllable in Khezha is open. The glottal
fricative phonemes in Khezha are found to occur word initially and medially, as all other
consonants. A minimal and a sub-minimal pair for the sounds are given below:
Given the rarity of contrast between a voiceless and a voiced glottal fricative in world
languages, and the greater realization of the glottal fricative as an approximant, how do we
characterize the glottal fricative/ approximant consonant? In languages that have only the
approximant realization [], it should well be treated as a glottal approximant phoneme with
the feature combination [+voice, +spread glottis]. In languages such as Hindi and Turkish,
that have both the fricative and approximant realizations, the phoneme may be treated as a
fricative with the feature combination [+spread glottis, -voice] that also has an allophonic
approximant realization having the feature combination [+spread glottis, +voice].
There are two important questions that arise at this point and that need to be addressed.
a) How is it that a contrast between a voiceless and a voiced glottal fricative is so rare in
languages?
b) How is it that languages that lack in laryngeal voiceless and voiced fricative contrast
have them as allophones?
21
There appears to be a puzzling situation here. To put it briefly, although languages lack in
both voiceless and voiced glottal fricative phonemes, they have them both as allophones.
An answer to this puzzle lies in my view in the explanations for the structure of
phonological inventories. These are of two types. One explains why a laryngeal contrast is
hard to come by and the other predicts why laryngeal allophones emerge. These are
complementary in nature. We will discuss the first type first.
There have been attempts at explaining how segmental inventories are determined in terms of
the presence of absence of contrasts. Two of them are discussed below.
i u
e o
a
i u
e o
a
22
Chart-2: A typical 7 vowel system
The vowel inventories in the above charts show a symmetrical distribution of vowels in
terms of height and front-back parameters. Such symmetrical patterns are also a subject of
phonetic investigation and explanation (e.g. Boersma 1998, De Boer 2001, and Schwartz et
al. 1997). The phonetic explanation rests largely on showing how the perceptual distance
between vowels matches the model of vowel systems as minimizing their potential energy.
The theory of Sufficient Contrast gives a coherent account of symmetrical vowel patterns.
However it hasnt been found to be equally applicable to asymmetrical vowel patterns as well
as consonant systems in general. The distribution of contrast patterns of laryngeal consonants,
however, that we have discussed above, appears to be predictable by the theory. It is not
difficult to see that the distinction between a voiceless glottal fricative and a voiced glottal
fricative is not so easily perceptible. As we saw above for Hindi, following a vowel the
distinction between a fricativized [h] and an approximant [] is blurred. Graph-3 and Graph-
4 are reproduced below for the sake of easy reference:
23
Graph-4: Foot- initial approximant []
In order to maintain a difference between the two, the potential articulatory energy must
be maximized, as it is in the case of vowels that are articulated close to each other, such as [
u].
We are now left with the question addressed above: How is it that languages that lack in
laryngeal voiceless and voiced fricative contrast have them as allophones?
24
Clements in a number of studies (e.g. 2001, 2003) drawing from earlier work (e.g. Martinet
1955) and in consonance with a lot of contemporary work on featural economy argues for a
general principle of representational economy according to which representational elements
are specified in a given language only to the extent that they are needed in order to express
generalizations about the phonological system (Clements 2001).
The theory of representational economy departs from earlier versions of feature theory,
namely, full specification (Chomsky and Halle 1968), and underspecification theories (e.g.
Archangeli ) in assuming that only active features are specified. Features should be
lexically distinctive, phonologically involved in alternations, and phonetically pronounceable.
They are thus activated at different levels: lexical, phonological and phonetic. Features can be
ranked on a Feature Accessibility Scale (e.g. Calabrese 1988), given the frequency of their
activation. The feature [sonorant ] thus ranks highest on the scale, because it is contrasts
segments at the lexical level in all languages, whereas the feature [constricted glottis] ranks
lower on the scale, since fewer languages have glottal stops.
Distinctive features that are required in some languages may not be relevant in other
languages. When they are not relevant, it is assumed that they lie dormant and are not
activated in the language. For example, Tamil does not have aspirated consonants either as
phonemes or as allophones. We assume that the feature [spread glottis] is not activated at
any stage in the phonology of Tamil.
25
Clements proposes a way of giving a quantitative measure of economy in terms of the
Economy Index. This is simply the ratio of speech sounds in a system over the smallest
number of features required to characterize them. Thus, given a system with S speech sounds
characterized by F features, its economy index, E, is given by expression (8):
(8) E=S/F
The higher the value of E, the greater the economy. (Clements 2003: 289).
The measure of economy is calculated by counting the number of segments present and by
dividing it with the number of features involved (see Clements 2003: 290 for details).
Clements illustrates the working of the principle with the help of three consonant systems-
Hawaiian, French and Nepali- that show consonant inventories of different sizes, but involve
feature economy in their patterning
a. Hawaiian: 8 consonants
p k
m n
w l
h
b. French: 18 consonants
p t k
b d g
f s
v z
m n
l
c. Nepali: 27 consonants
26
p t ts k
p t ts k
b d dz g
b d dz g
s
m n ()
l,r
Hawaiian is selected for using two places of articulation, labial and non-labial, with
three manners of articulation yielding the maximum number of 6 (2X3) supralaryngeal
consonants. In French voicing is employed with stops and fricatives to yield 12 (2X6)
obstruents. In Nepali, five places of articulation and four manners of articulations are
used fully to give 20 (5X4) stops.
Notice that not all features are utilized in these systems. For example, although there
is /h/ in Hawaiian, but there is no aspirated stop, or there is only one nasal consonant in
it. Similarly, French lacks in palatal stops, and Nepali excludes aspirated sonorants,
although theoretically these sounds were possible in the system.
Now let us see how the Economy measure applies to the three systems. Following the
algorithm described in Clements (2001), the list of features involved in the three systems
is given in (..) below. Features that are checked are distinctive and that are left blank are
non-distinctive in the systems. Thus [dorsal] is non-distinctive in Hawaiian and [spread
glottis] is non-distinctive in French.
Distinctive features in Hawaiian, French and Nepali consonants
27
[posterior]
[apical]
[lateral] 5 7
10
total
The Economy measure yields the following ranking among the three systems:
Hawaiian 3 5 1.60
French 7 18 2.57
Nepali 10 27 2.70
the voiced/ voiceless contrast in the stop series- /p p b b/ etc. As they are already activated
for stops, the feature economy predicts that the combination will apply to other entities as
well to yield more segments. [] thus results from the combination of [+voice] to the
laryngeal node. Tha latter is an abstract category which functions in phonological processes.
28
argued that the phoneme is realized in a non-neutralizing context, such as the word-initial
position, as a fricative and should thus be treated as such. /h/ also has, however, an
approximant realization in neutralizing contexts such as the intervocalic and the word-final
positions. Both the lack of a laryngeal contrast as well as the emergence of laryngeal non-
contrastive segments in world languages are explained by independent but consistent lines of
inquiry, namely Maximal Dispersion, Prosodic Licensing, Licensing by Cue and Feature
Economy. The paper also shows that phonological and phonetic inventories are governed by
both articulatory and perceptual constraints (e.g. Ohala 1980, 1981, Westbury and Keating
1985, Kingston 1990, Hayes et al. 1994, Kingston & Diehl 1995) and abstract phonological
and cognitive constraints, such as Feature Economy discussed above.
An evidence for the argument that phonological inventories are governed by abstract
phonological and cognitive properties in the grammar is forthcoming from the fact that not
only a voiced glottal fricative emerges with the combination of the activated features [+voice]
and [spread glottis], but that other laryngeal segments that function under the laryngeal node
may also emerge as allophones. This indeed is the case in some languages. Cases of the
glottal stop realized from /h/ abound in languages, as, for instance in the Indic languages
Korku, Mandeali and Zo (Pandey 2014: 142). A more interesting case is of the Nigerian
language in which /h/ is realized additionally as [] voiced pharyngeal fricative and []:
How can the approximant realization of the combination of the features [+voice] and
[spread glottis] be explained? The way the combination works can be described thus. /h/ is
devoid of an articulatory gesture in not having a place node, but is associated to the laryngeal
node for phonological functioning. When it takes the feature [+voice], it does so as
assimilation to the neighbouring adjacent vowel(s). In effect, it also takes the formant values
of the adjacent vowel(s). This has the effect of a merger of the vowel features to the glottal
fricative as illustrated in the following diagram:
[-voice] V place
+voice
[h]
29
As /h/ is merged with the adjacent vowel, it acquires the properties of an approximant.
An evidence of the merger account of /h/ with the [+voice] and place features of the adjacent
vowel in Hindi comes from difficulty or rather impossibility of separating the fricative
component of /h/ from the following vowel in the appparent cases of schwa deletion in Hindi.
In cases of an lexically intervocalic /h/, attempts at slicing the following vowel from the
preceding [] fail. The resultant sound is still a breathy vowel [] or [e]. If the slicing is
considerable, one gets to hear a glottal stop in stead, as discussed in section above. This is
not the case with the word-initial fricative /h/. The fricative can be sliced off leaving the
following vowel as is.
It appears that most cases of approximant [] arise from a merger of voice and the vowel
place features with the fricative.
References
CHOMSKY, N & M. HALLE 1968. Sound Pattern of English. NY: Harper and Row.
30
CLEMENTS, Geoge N. (2009). The role of features in phonological inventories. In E. Raimy
and C. E. Cairns (eds.) Contemporary views on architecture and representations in
phonology. Massachusetts: MIT Press. 19-68.
DAI, Qingxia, & DIEHL, Lon. (2003). Jingpho. In: Thurgood, Graham, and LaPolla, Randy
J. (eds.) The Sino-Tibetan Languages. London: Routledge. 401-408.
DE BOER, B. (2001). The Origin of Vowel Systems. Oxford University Press, Oxford.
DIXIT, R. P. & J. MACNEILAGE (1974). Glottal dynamics during Hindi bilabial plosives
and the glottal fricative. MM10. Acoust. Soc. Am. 55, S80.
http://dx.doi.org/10.1121/1.1919977
DIXIT, R. P. (1963). The segmental phonemes of contemporary Hindi. Unpublished M.A.
thesis. Austin, University of Texas.
EKKA, F. (1985). Kurux phonetic reader. Mysore: Central Institute of Indian
Languages.
FANAI, L. T. (1992). Some aspects of the lexical phonology of Mizo and English: An
autosegmental approach. Ph.D. dissertation. Hyderabad: CIEFL.
HALLE, Morris & Kenneth N. STEVENS, (1971). A note on laryngeal features. Quarterly
Progress Report. Cambridge MA: MIT, Research Laboratory of Electronics. 101.
198-213.
HALLE, Morris & Kenneth STEVENS (1971). A note on laryngeal features. Quarterly
Progress Report 101 198-192. (Research Laboratory of Electronics, MIT)
HAYES, Bruce, Robert KIRCHNER & Donca STERIADE (eds.) (2004). Phonetically Based
Phonology. Cambridge: Cambridge University Press.
HINSKENS, Frans & Jeroen van de WEIJER (2003). Patterns of segmental modification in
consonant inventories : a cross-linguistic study. Linguistics 41. 10411084.
ISRAEL, M. (1979). A grammar of the Kuvi language. Trivandrum: DLA.
ITO, Junko (1989). A Prosodic Theory of Epenthesis. Natural Language and Linguistic
Theory 7. 217-259
KAPFO, Kedutso (2005). The Ethnology of the Khezhas and the Khezha Grammar. Mysore
: Central Institute of Indian Languages.
KEATING, P. (1988). Underspecification in phonetics. Phonology 5. 275-292.
KEHREIN, Wolgang & Chris GOLSTON. (2004). A prosodic theory of laryngeal contrasts.
Phonology 21. 325357.
31
KELLOGG, S. H. (1893). A Grammar of the Hindi Language. 2nd ed. London: Routledge &
Kegan Paul.
KENSTOWICZ, Michael. (1994). Phonology in Generative Grammar. Oxford: Blackwell.
KIM, C-W. (1970). A theory of aspiration. Phonetica 21. 107116.
KINGSTON, John (1985). The Phonetics and Phonology of the Timing of Oral and Glottal
Events, University of California, Berkeley Ph.D.Dissertation.
KINGSTON, John & Randy DIEHL (1995). "Intermediate properties in the perception of
distinctive feature values," in Bruce Connell and Amalia Arvaniti (ed.) Papers in
Laboratory Phonology 4, Cambridge University Press.
LADAFOGED, P. (2001) A Course in Phonetics, 4th ed. New York: Harcourt College
Publishers.
LADEFOGED, P. (1982). A Course in Phonetics. San Diego: Harcourt Brace Jovanovich.
LADEFOGED, P. (1990). Some proposals concerning glottal consonants. Journal of the
International Phonetic Association 20. 33-40.
LADEFOGED, Peter & Ian MADDIESON (1996). The Sounds of the World's Languages.
London: Blackwell.
LAHIRI, Aditi (ed.) (2000). Analogy, Leveling, Markdedness. Priinciples of Change in
Phonology and Morphology. Berlin: Mouton de Gruyter.
LINDBLOM, Bjrn (1986). Phonetic universals in vowel systems. In John J. Ohala & Jeri J.
Jaeger (eds.) (1986). Experimental phonology. Orlando: Academic Press. 1344.
32
LOMBARDI, Linda (1995). Laryngeal neutralization and syllable wellformedness. NLLT 13.
3974.
MIELKE , Jeff. (2002). Turkish /h/ Deletion: Evidence for the Interplay of Speech Perception
and Phonology. ZAS Papers in Linguistics 28. 55-7.
MOHAN, Shailendra (2009.) A note on schwa fronting in Hindi. Indian Linguistics 70 223-
225.
MLLER, Hermann (1970) [1911]). Vergleichendes indogermanisch-semitisches
Wrterbuch. Gttingen: Vandenhoek & Ruprecht.
MURTHY, Hema A. Text-to-Speech Synthesis for Hindi and Indian Languages.
http://www.iitm.ac.in/donlab/website_files/research/Speech/TTS/contents/main.html
OHALA, John J. (1980). Introduction to the symposium on phonetic universals in
phonological systems and their explanation. In Proceedings of the 9th International
Congress of Phonetic Sciences. Vol. 3. University of Copenhagen: Institute of
Phonetics. 181185.
OHALA, John J. (1981). The listener as a source of sound change. In Carrie Masek, Roberta
Hendrick and Mary Frances Miller (eds.) Papers from the parasession on language
and behavior. Chicago, Il: Chicago Linguistic Society. 178-203.
33
University Press. 90-116
STERIADE, Donca. (2009). The Phonology of Perceptibility Effects: The P-Map and Its
Consequences for Constraint Organization. In Kristin Hanson and Sharon Inkelas
(eds.), The Nature of the Word: Studies in Honor of Paul Kiparsky. Cambridge,
Massachusetts: The MIT Press. Pp.151- 180.
VAUX, B. (1998). The laryngeal specifications of fricatives. Linguistic Inquiry 29. 497511.
VAUX, Bert (2005). The role of features in a symbolic theory of phonology: comments on
34
Clements.
http://people.ds.cam.ac.uk/bv230/li6/comments%20on%20clements%2011-7-05.pdf
WELLS, John. (1982). Accents of English 1: An Introduction. Cambridge University Press.
WESTBURY, John R. & Patricia KEATING (1986). On the naturalness of stop consonant
voicing. Journal of Linguistics 22. 145-166
35