The Blurring Effect of Sonorants and Vocalic Alternations in Gaelic

1
The Blurring Effect of Sonorants andVocalic
Alternations in Gaelic

• Roy Becker
• University College Dublin
• 2003

2
Main hypothesisVowel quality is perceptually
blurred by the presence of a following sonorant

• Why ???
• Co-articulatory effects unique to sonorants alter
  the resonations, and hence the auditory image, of
  the vowel (this is well-known)
• Sonorants have backward masking effects on
  preceding vowels (really?)

• So what ???
• If accurate perception (faithfulness) is
  important in language, then in case of
  substantial blurring, a faithfulness-oriented
  grammar should provide repair mechanisms against
  such blurring.
• Claim this happened inEarly Modern Gaelic.

3
The blurring effect of sonorants (1) Vowel
perception cues for vowel quality
(a) Auditory representation of the acoustic
resonation patterns (formants) of the vowel
kernel, eventually a function of the size and
shape of the cavities of the vocal tract (e.g.
Johnson 199764).
(what if the cavities are altered ?)
4
The blurring effect of sonorants (1) Vowel
perception cues for vowel quality
(b) Vowel-specific co-articulatory effects over
adjacent consonants, i.e. the acoustics of the
consonant and/or the transition serve as a cue
for the quality of the vowel, e.g. Liberman et
al. (1967).Notice the spectrographic differences
between the corresponding zs and ls in the
following spectrograms.
(what if such effects are absent, e.g. in the
case of secondary articulation ?)
5
The blurring effect of sonorants (2) Regressive
co-articulatory colouring effects of sonorants

• Nasals vowel nasalisation - anticipatory
  lowering of the velum (regressive nasalisation)
  resulting in altered resonations due to coupling
  of the oral and nasal cavities (e.g. Wright
  1986).
• Coronal Liquids vowel darkening - anticipatory
  retraction and lowering of the dorsum, a typical
  gesture that facilitates both tongue-tip
  vibration (trill), retroflection (approximant)
  and lateral aperture during central constriction
  (lateral). This anticipatory gesture is
  particularly common for coda liquids, and results
  in altered resonations of the oral cavity (e.g.
  Sproat Fujimura 1993).
• (c) Dorsal Rhotics transitional quality - trills
  and approximants always involve precise (and
  hence gradual, non-ballistic) positioning of the
  articulator. When the articulator is the dorsum,
  a vowel-rhotic sequence, like diphthongs, is
  essentially transitional, but (unlike diphthongs)
  lacking a phonemic status.
• For the general a-typicality of pre-sonorant
  vowel quality as a result of regressive
  co-articulatory effects see e.g. Guenther et al.
  (1999).

6
The blurring effect of sonorants (2) Regressive
co-articulatory colouring effects of sonorants -
illustrations
7
The blurring effect of sonorants (3) Backward
masking effect on a preceding vowel
(a) Introduction to backward masking Auditory
backward masking interference of a consecutive
sound (masker) in the perceptual processing of
the auditory cues of the preceding sound
(target), e.g. Massaro (1972,1973). The
experimental correlate of masking is degraded
performance in target detection/identification
tasks as a function of masker introduction. Simple
illustration with tone recognition Three tones
to remember low, mid high Which tone is this
one? this one? and this one? Now listen to a
tone followed by random noise. Which tone is this
one?
8
The blurring effect of sonorants (3) Backward
masking effect on a preceding vowel

• (b) Properties enhancing backward masking
• Loudness of the masker (softness of the target),
  e.g. Repp (1975).Compare
• Length of the masker (shortness of target ), e.g.
  Repp (1975).Compare
• Immediate adjacency of the masker to the target,
  e.g. Homick et al. (1969).Compare
• Categorical auditory similarity of masker and
  target, e.g. LoebHolding (1975), unless, of
  course, the target and the masker are identical,
  e.g. Repp (1975).Compare
• Inventory and auditory density of possible
  targets, e.g. Dorman et al. (1977).

9
The blurring effect of sonorants (3) Backward
masking effect on a preceding vowel

• (c) Sonorants optimally backward-mask vowels,
  because
• Sonorants are louder than obstruents, e.g.
  Fletcher (195384-86)
• Sonorants (like fricatives) are auditorily
  continuant they do not involve initial silence
  (like stops or affricates).
• Sonorants are categorically similar to vowels
  the auditory image of both is formant-based, e.g.
  Tarnóczy (1948), Boersma (199818).
• But sonorants are typically short, compared to
  obstruents.
• Yet what if the sonorants are lengthened (e.g.
  geminates)?
• Hypothesis
• Compared to other sounds (obstruents), sonorants
  (nasals, liquids, glides and vowels), have
  significantly greater potential for backward
  masking immediately preceding vowels, especially
  when the sonorant is lengthened and the target
  vowel is short.

10
Phonetic features of Early Modern Gaelic
straightforward reconstruction according to
shared features ofmodern spoken dialects of
Irish and Scottish Gaelic (1) Consonants

• Notice
• Secondary articulation slender (palatalised)
  vs. broad (velo-uvularised).
• Multiple contrasts among coronal sonorants.
• Aspiration of voiceless stops.

11
Phonetic features of Early Modern Gaelic(1)
Consonants

• Robust secondary articulation Following
  shortening of certain diphthongs by clipping
  their initial part in Middle Gaelic, allophonic
  co-articulation becomes phonemic secondary
  articulation (McManus 1994).A consonant is
  either palatalised or velo-uvularised (as in e.g.
  Russian), and co-articulatory effects of a
  following vowel are blocked, e.g. Ní Chiosáin
  Padgett (2001). Hence, the acoustics of the
  consonant and the transition do not cue to the
  quality of the following vowel.Notice the long
  transition during the vowels after secondarily
  articulated l

12
Phonetic features of Early Modern Gaelic(1)
Consonants
(b) Lax/tense contrast among coronal sonorants
Lax (historically singleton) coronal sonorants
involve unmarked articulation, while their tense
(historically geminate) correlates involve large
surface contact (distributed) and peripheral
placement of the primary articulator, e.g.
Quiggin (1906), Mhac an Fhailigh (1968). Examples
with the laterals are
The time-costliness of the marked gestures of the
tense coronal sonorants make them inherently
longer than their lax counterparts, while the
labial and dorsal nasals are presumably in
between, duration-wise.
13
Phonetic features of Early Modern Gaelic(1)
Consonants

• (c) Aspiration of voiceless stops Voiceless
  stops are both pre- and post-aspirated, with
  partial devoicing effects on adjacent vowels and
  sonorants, e.g. Ní Chasaide (1999). The devoiced
  portion of a sonorant is merely a mellow
  voiceless fricative, with very limited
  intensity.Compareblaosc 'shell'
  p?l?e?s?k vs. pléasc 'explosion p??l??e?s?k
  borb 'fierce' p????b? vs. corp 'body'
  k????p??seilg 'hunt' ?e?l?? vs. dailc
  'squat' t??l??c?

14
Phonetic features of Early Modern Gaelic(1)
Consonants

• (c) Backward-masking potential of consonants in
  Gaelic Following the principles of
  backward-masking, emphasising maskers duration,
  intensity and categorical similarity to the
  target (slide no. 8 above), Gaelic consonants
  have backward-masking potential over preceding
  vowels, according to the following hierarchy
• Tense coronal sonorants n???, n???, l???, l???,
  r?
• Non-coronal nasals and glides m?, m?, ?, ?, j,
  w
• Lax coronal sonorants n?, n?, l?, l?, r?, r?
• Devoiced lax coronal sonorants n??, n??, l??,
  l??, r??, r??
• Obstruents (voiced fricatives gt voiceless
  fricatives gt voiced stops gt voiceless stops)

15
Phonetic features of Early Modern
Gaelic straightforward reconstruction according
to shared features ofmodern spoken dialects of
Irish and Scottish Gaelic (2) Vowels
(a) Scope - stressed short vowels Quantity was
(and still is) contrastive, e.g. caise
stream ?k?a.?? vs. cáise of cheese
?k?a?.?? ciste fund ?k?i?.t???? vs. císte
cake ?k?i??.t???? As long vowels are rather
immune to blurring, our concern is with short
vowels. Furthermore, as quality contrast among
short vowels was (and still is, in most modern
dialects) limited only to word-initially stressed
syllable (e.g. McManus 1994), and as blurring of
non-contrastive sounds is essentially
ineffective, our focus is on word-initial,
stressed short vowels only.
16
Phonetic features of Early Modern Gaelic(2)
Vowels

• (b) Features of stressed short vowels
• Grammatical (phonemic) contrast is limited to a
  ternary height parameter (high, mid low), e.g.
  Ní Chiosáin (1991).
• The degree of frontness is determined by the
  secondary articulation of the surrounding
  consonants. Lip-rounding is determined by a
  combination of vowel height and both primary and
  secondary articulation of the surrounding
  consonants. Both features are therefore
  non-contrastive.
• Nevertheless, there is great auditory diversity,
  spanning most of the vocalic space (e.g.
  Sommerfelt 1922 mentions 20 different qualities
  of short vowels).
• Examples

17
Vocalic Alternations in Contemporary Irish

• Lengthening of a stressed short vowel before a
  non-lax sonorant in codasingular
  pluralcam fraud ?k???m?
  ?k??m? cama ?k??.m?? ?k???.m??
  faill 'opportunity ?f???l??
  ?f??l?? failleannaí ?f??.l???.n??i?
  ?f???.l???.n??i?
• but not if the sonorant is lax or if the vowel
  is long or if the vowel is unstressed
• fail 'hick-up' f??l? faileannaí
  ?f??.l??.n??i?
• fáil 'receiving ?f???l?
  fáileachaí ?f???.l??.xi?
• gradam award ?gr??.d???m?
  ?gr??.d????m?
• capaill 'horses' ?k??.p???l??
  ?k??.p????l??
• 2. Lengthening before a lax coronal sonorant
  followed by a coronal other than t?
• ard 'high' ?????d? ????d?
• art 'on you' ??????t??
• 3. Epenthesis between a lax coronal sonorant and
  a non-coronal other than a voiceless stop, after
  a short voweltairbh bulls ?t???.r??v?
  ?t??????v? tairg offer (v.) ?t???.r???
  ?t???????
• táirg produce ?t???????
• toirc boars' ?t???????c?
• For evidence of the productiveness of these
  lengthening and epenthesis alternations see
  Becker (200368-75).

18
Vocalic Alternations in Early Modern Gaelic
The emergence of half-long vowels, e.g. McManus
(1994)
A stressed short vowel becomes half-long before
any sonorant in coda unless the sonorant is lax
and is word final or followed by a
(pre-aspirated) voiceless stop. E.g. (a)
faill 'opportunity ?f???l?? cam fraud
?k???m? (b) tairbh bulls ?t???a???v? ard
high ?????d?? According to Greene (1952), the
second environment gave rise to epenthesis of a
copy vowel into the cluster, rather than
lengthening. At least perceptually this copy
vowel did not introduce another syllable, as is
known to be the case of modern instances of this
epenthesis in Scottish Gaelic (e.g. Watson
1994) tairbh bulls ?t???ar?av? ard high
??r??d??. In studies of backward masking, both
target lengthening and recurrence (copying) are
known to enhance target recognition. Claim
Half-long vowels / copy epenthesis were repair
mechanisms used to overcome the crucial blurring
effect of sonorants in Gaelic.
19
Vocalic Alternations in Early Modern Gaelic The
emergence of half-long vowels / copy-epenthesis

• This lengthening / epenthesis process did not
  affect
• Long vowels, which are immune to blurring due to
  their inherent length.
• Unstressed (non-contrastive) short vowels,
  because blurring is essentially ineffective when
  the target is non-contrastive.
• Short vowels in open or half-open syllables.
  The split of syllabic domain reduces regressive
  co-articulatory effects, and possibly also
  reduces auditory association of the target and
  the masker.
• Short vowels followed by a devoiced sonorant.
  Auditorily, the sonorant becomes a soft
  fricative, and loses most of its masking
  potential.
• Short vowels in monosyllabic words with a lax
  sonorant as a simple coda, e.g. col prohibition
  ?k??l?. Gaelic is a stress-timing as well as
  stress-initial language. It is plausible that
  such short monosyllabic words are automatically
  lengthened in order to increase the period of
  time until the next stressed syllable (the
  initial syllable of the following word). In
  Donegal Irish, the only modern dialect which has
  syllable timing (Ní Chasaide, pc.), a lax
  sonorant in such words is devoiced (e.g. Ní
  Chasaide 1979) ?k??l???. Compare ngael of
  Gaels ??e?l? vs. geal clear ?cal??? sról
  satin ?s?r???l? vs. scoil school
  ?s?k?l???

20
The Blurring Effect - Experimental
Validation(1) Hypotheses

• General hypothesis Perception of vowel quality
  is degraded by the presence of a following
  sonorant, as a combination of regressive
  co-articulatory colouring effects unique to
  sonorants and backward masking. The case of
  regressive colouring is well established, and the
  current study focuses on the effect of
  backward-masking on vowel perception.
• Further operational hypotheses concerning
  backward masking
• In an identification task, a pre-sonorant vowel
  should be identified less adequately than a
  pre-obstruent vowel, all else being equal (no
  co-articulatory effects, identical pitch,
  acoustics, duration and intensity).
• Lengthening a pre-sonorant vowel should improve
  its perception.
• Devoicing a sonorant should improve the
  perception of a preceding vowel.
• Lengthening the sonorant should degrade the
  perception of a preceding vowel (not studied
  here).
• With particular reference to Irish,
  co-articulatory effects of vowels on adjacent
  consonants should be neutralised, and correct
  vowel identification is determined by height
  identification.

21
The Blurring Effect - Experimental
Validation(2) Stimuli

• The stimuli are nonsense monosyllables of the
  form dVC. They were recorded and constructed
  using PRAAT, as follows
• A reference monosyllable d? was recorded with
  level pitch p and vowel kernel amplitude a. The
  onset together with the first two voicing cycles
  of the vowel were extracted to yield a constant
  onset and transition (all signal truncations here
  and elsewhere were performed at zero-crossings in
  the steep rising of a voicing cycle).
• The vowels ? e? a ? o? ? were recorded with p
  in environment d_. Their kernels were then
  normalised to a. The high and low vowels were
  then slightly attenuated and amplified,
  respectively, as best compromise between natural
  relative amplitude and experimental
  comparability. Then kernels with c-1, c and
  c1voicing cycles were extracted, for high, mid
  and low vowels respectively, again as best
  compromise between natural relative duration and
  experimental comparability.
• The codas p t k b d g f s ? v z ? m n ? l r ?
  m?? n?? ??? l?? r?? ??? were recorded with
  roughly identical duration d in environment d?_
  with p. ? kernels were equalized to a, thus
  obtaining representative relative intensities
  amongst the various codas. Codas were extracted
  together with the last two cycles of the
  preceding vowel.
• All combinations of common onset vowel coda
  were concatenated to yield naturally sounding,
  co-articulation-free d?V?C tokens.

22
The Blurring Effect - Experimental
Validation(2) Stimuli
23
The Blurring Effect - Experimental Validation(3)
Method

• Subjects About 40 first-year students of
  linguistics, most of whom speakers of
  Hiberno-English (importantly no native speaker
  of Irish), with a few weeks experience in
  phonetic transcription.
• Task Passive (multiple choice) phonetic
  transcription of the vowels in the stimuli.
• Procedure An experiment recording consisted of a
  sequence of n blocks of 8 stimuli-tokens, in
  random order (adjacent tokens never contained the
  same vowel). Dummy tokens padded the beginning
  and end of the recording. Between-block interval
  was 7.2s and between-token interval was 2.8s. A
  brief order announcement preceded each block and
  token. The digital master-recording was rendered
  onto analog tapes. The subjects, seated in
  separate booths, listened to the tapes via
  earphones at a comfortable level and could
  neither stop nor rewind the tapes. Upon hearing a
  stimulus, the subject, faced with the IPA symbols
  ? ? a ? ? ? ?, circled the one matching the
  vowel she heard, in the appropriate cell in the
  provided answering sheet. No performance feedback
  was given at any stage.
• Evaluation A given stimulus transcription was
  evaluated as correct if the height of the
  transcribed vowel matched the height of the vowel
  contained in the stimulus. Height mismatch or
  blank cell were evaluated as incorrect. Subjects
  with extremely good performance and subjects with
  near (or below) random performance were dropped
  in order to avoid ceiling and random-noise
  effects, respectively.
• Why was ? included? Perception-grammar Anyway,
  it was evaluated exactly like ?.

24
The Blurring Effect - Experimental Validation(4)
Experiment 1 Sonorants vs. fricatives vs. stops
statistic parameters

• Stimuli ?,e?,a,?,o?,?xp,t,k,b,d,g,f,s,?,v,z,?,
  m,n,?,l,r,?,no-coda6 dummies 120 stimuli.
• Auditory classes stop (p,t,k),(b,d,g),
  fricative (f,s,?),(v,z, ?), sonorant
  (m,n,?),(l,r,?)
• Removal of side-effects ?-data was dropped
  due to a clear ceiling effect (almost all
  subjects had near-perfect recognition results for
  this vowel, probably because of the substantial
  experience subjects had with this vowel in their
  phonetics lab sessions. It is also likely that
  the double centre-of-gravity effect of high front
  vowels (F0F1, F2F3F4), contribute to the
  relatively stable auditory image of this vowel
  and immune it against masking (e.g. Stevens
  1998266-268).As for the rest of the data, 3 out
  of the 36 subjects were dropped due to ceiling
  effect (above 85 correct). 14 subjects were
  dropped due to random-noise effect (below 45
  correct). No other sounds or subjects
  systematically twist the results.
• Statistics after removal of side effects
• Subjects 19, Tokens 95 (including 5 no-coda
  tokens)
• Answers per coda 95, per auditory class
  570, per auditory sub-class 285 per class per
  subject 30, per sub-class per subject 15
• Correct answers for no-coda stimuli (can be taken
  as default vowel recognition for this subject
  population at these experiment conditions) 56
  (58.9).

25
The Blurring Effect - Experimental Validation(4)
Experiment 1 Sonorants vs. fricatives vs. stops
- results
Significance tests across coda classesAssuming
default recognition probability per token is
0.589, cumulative z-scores are
Stops p0.448
Fricatives p0.289 Sonorants
plt0.000001voiceless p0.57, voiced p0.38
voiceless p0.57, voiced p0.18 nasal
plt0.002, oral plt0.0001 Significance tests across
subjects (no default recognition probability
assumed)One-tailed matched-pair t-test
cumulative p-scores (on correct answers per coda
class)
Both tests indicate that the results are
significant both at auditory class and sub-class
level, and confirm the hypothesis that vowel
recognition is degraded by the presence of a
following sonorant vowels are backward-masked
(only) by sonorants.
26
The Blurring Effect - Experimental Validation(5)
Experiment 2 the effects of vowel lengthening
and syllabic split

• Additional stimuli This experiment used the
  sonorant-coda stimuli of ex. 1 (?-tokens were
  dropped), together with two parallel sets. For
  each d?V?C token two parallel tokens were
  constructed (a) A token with a longer vowel (the
  same vowel kernel was used, but the middle
  voicing cycle was reduplicated as needed). These
  longer vowels had c1, c2 and c4 voicing cycles
  for high, mid low vowels respectively. (b) A
  token in which the sonorant is slightly shortened
  and released into ? (The sonorant was recorded
  in environment d?_? with p and equalised to a
  as in ex.1. Then the final portion of the
  sonorant together with the release into the
  following ? were truncated and concatenated to
  the d?V?C token in the middle of the coda. A
  constant unstressed final ? was concatenated to
  all tokens at the end).Thus d?V?C, d?V??C,
  d?V?C? triplets were obtained,
  e.g.Admittedly, the d?V?C? did not sound very
  natural, and this might affect the results.The
  stimuli were randomized as in ex.1
• Parameters
• Stimuli e?,a,?,o?,?xm,n,?,l,r,?x 3
  conditions 6 dummies 96 stimuli
• Removal of side effects 2 out of the 36 subjects
  were dropped due to ceiling effect (above 85
  correct). 17 other subjects were dropped due to
  random-noise effect (below 45 correct).
• Answers per token 17, per sonorant 255, per
  condition 510, per condition per sonorant
  85, per condition per subject 30

27
The Blurring Effect - Experimental Validation(5)
Experiment 2 vowel lengthening and syllabic
split - results
Significance tests across token
tripletsOne-tailed matched-pair t-test
cumulative probabilities (on correct answers per
tokens)dV?C/dVC? 0.198 dV?C/dVC
0.049 dVC?/dVC 0.221Significance tests
across subjectsTwo-tailed matched-pair t-test
cumulative probabilities (on correct answers per
condition) dV?C/dVC? 0.165 dV?C/dVC
0.006 dVC?/dVC 0.192 The tests indicate
that the results are significant for the
dV?C/dVC condition, while the tendency for
the dVC?/dVC condition is not significant.
Therefore, the results confirm the hypothesis
that lengthening enhances vowel recognition.
Hence, vowel lengthening is an appropriate repair
mechanism for overcoming backward masking.
28
The Blurring Effect - Experimental Validation(6)
Experiment 3 the effect of sonorant devoicing

• Stimuli This experiment used the sonorant-coda
  stimuli of ex. 1, together with a parallel set of
  stimuli, containing the corresponding d?V?C??
  (the coda sonorant in such tokens has the same
  duration and its latter half is voiceless).
  E.g.The stimuli were randomised as in the
  previous experiments.
• Parameters
• Stimuli e?,a,?,o?,?x(m,n,?,l,r,?),(m??,n??,???
  ,l??,r??,???) 4 dummies 64 stimuli
• Removal of side effects 2 out of the 36 subjects
  were dropped due to ceiling effect (above 85
  correct). 17 other subjects were dropped due to
  random-noise effect (below 45 correct).
• Answers per token 15, per sonorant 225, per
  voicing condition 450 per condition per
  sonorant 75, per condition per subject 30

29
The Blurring Effect - Experimental Validation(6)
Experiment 3 the effect of sonorant devoicing -
results
Significance tests (One-tailed matched-pair
t-test cumulative probabilities ) Across token
pairs (on total correct answers per parallel
tokens) 0.038Across subjects (on subjects
correct answers per voicing condition) 0.030Both
tests indicate that the results are nearly
significant, supporting the hypothesis that
devoicing reduces the backward-masking effect of
a sonorant on a preceding vowel and thus
improving vowel recognition.
30
The Blurring Effect - Experimental Validation(7)
Commentary on the experiments

• The lexical status of tokens was not taken into
  account. Familiarity with a lexical item (e.g
  d?k dan compared to d?g, dar) enhances
  the recognition of a corresponding stimulus, e.g.
  (WalleyFlege 1999). This could have been
  prevented, by e.g. using an any ungrammatical
  onset (x,? etc.). Nevertheless, lexical status
  of tokens had no effect on the variables under
  question, and at least the high recognition
  results for dV?-tokens indicate that lexical
  status had only marginal effect here, if at all
  (in English, ? is an ungrammatical coda after
  short vowels, and an extremely rare coda
  otherwise).
• The effect of coda consonant on vowel duration
  was not taken into account. In English (e.g.
  PetersonLehiste 1960), vowel duration is a
  function of the following coda hierarchy
  stop-vls ltlt stop-vcd lt fric-vls lt son ltlt fric-vcd
  lt ?.Nevertheless, the vowel in the stimuli
  were always ungrammatically extra-short,
  biasing similarly against all coda classes except
  for voiceless stops. Finally, the predicted
  strong duration bias against dV?-tokens is
  unattested in the results.
• The absence of co-articulatory effects
  characteristic of post-vocalic sonorants makes
  the sonorant-final stimuli more unnatural than
  obstruent-final stimuli. Unfortunately, removing
  co-articulation is essential for testing the
  backward masking hypothesis, because degraded
  recognition performance can always be attributed
  to the presence of co-articulatory effects.
  However, in ex. 3, the absence of such effects is
  equally unnatural, while the results are
  nevertheless significant.
• A series of similar pilot experiments was
  conducted a year earlier, producing practically
  the same results (Becker 2003). While those
  earlier experiments lacked consistent and careful
  design, the repetition of the results indicate
  that the findings are reliable.

31
References
Becker, R. (2003) The Blurring Effect of
Sonorants and Vocalic Alternations in Irish. MA
Thesis. Dublin University College
Dublin. Boersma, P. (1998) Functional Phonology.
The Hague Holland Academic Graphics. Dorman, M.,
D. Kewley-Port, S. Brady M. Turvey (1977)
'Vowel recognition, inferences from studies of
forward and backward masking'. Quarterly Journal
of Experimental Psychology 29483-497. Fletcher,
H. (1953) Speech and Hearing in Communication.
Lancaster, PA Lancaster Press Inc. Greene, D.
(1952) 'Middle Quantity in Irish'. Ériu
18108-112. Guenter, J., J. Lewis M. Urban
(1999) 'A perceptual study of vowels before /r/',
in Proceedings of the Fourtienth International
Congress of Phonetic Sciences. Homick, J., L.
Elfner G. Boothe (1969) 'Auditory temporal
masking and the perception of order'. Journal of
the Acoustical Society of America
45712-718. Johnson, K. (1997) Acoustic
Auditory Phonetics. Cambridge, MA Blackwell
Publishers. Liberman, A., F. Cooper, D.
Shankweiler M. Studdert-Kennedy (1967)
Perception of the speech code. Psychological
Review 74431-461. Loeb, M. D. Holding (1975)
'Backward interference by tones or noise in pitch
perception as a function of practice'. Perception
and Psychophysics 18205-208. Massaro, D. (1972)
'Perceptual images, processing time and
perceptual units in auditory perception'.
Psychological Review 79124-145. Massaro, D.
(1973) 'A comparison of forward versus backward
recognition masking'. Journal of Experimental
Psychology 100434-436. McManus, D. (1994)
'Classical Modern Gaelic', in Stáir na Gaeilge
(eds. McKone, K., D. McManus, C. Ó Háinle, N.
Williams L. Bratnach), 336-446. Maynooth
Coláiste Phádraig. (in Irish). Mhac an Fhailigh,
E. (1968) The Irish of Erris, Co. Mayo. Dublin
Dublin Institute for Advanced Studies. Ní
Chasaide, A. (1979) 'Laterals of Gaoth Dobhair
Irish and Hiberno English', in Occasional Papers
in Linguistics and Language Learning 654-78. Ní
Chasaide, A. (1999) 'Irish', in The Handbook of
the International Phonetic Association.
Cambridge Cambridge University Press.
32
References (cont.)
Ní Chiosáin, M. (1991) Topics in the Phonology of
Irish. PhD Dissertation. Amherst, MA University
of Massachusetts. Ní Chiosáin, M. J. Padgett
(2001) 'Markedness, segment realization and
locality spreading', in Segmental Phonology in
Optimality Theory (ed. Lombardi L.). Cambridge
Cambridge University Press. Peterson, G. I.
Lehiste (1960) 'Duration of syllable nuclei in
English'. Journal of the Acoustical Society of
America 32693-703. Quiggin, E. (1906) A Dialect
of Donegal. Cambridge Cambridge University
Press. Repp, B. (1975) 'Dichotic masking of
consonants by vowels'. Journal of the Acoustical
Society of America 57724-735. Sommerfelt, A.
(1922) The Dialect of Torr, Co. Donegal.
Christiania J. Dybwad. Sproat, R. O. Fujimura
(1993) Allophonic variation in English /l/ and
its implications for phonetic implementation.
Jounal of Phonetics 21291-312. Stevens, K.
(1998) Acoustic Phonetics. Cambridge, MA MIT
Press. Tarnóczy, T. (1948) 'Resonance data
concerning nasals, laterals and trills'. Word
471-77. Walley, A. J. Flege (1999) 'Effects of
lexical status on children's and adults'
perception of native and non-native vowels'.
Journal of Phonetics 27307-332. Watson, S.
(1994) 'Scottish Gaelic', in Stáir na Gaeilge
(eds. McKone, K., D. McManus, C. Ó Háinle, N.
Williams L. Bratnach), 661-702. Maynooth
Coláiste Phádraig. (in Irish). Wright, J. (1986)
'The behaviour of nasalized vowels in the
perceptual vowel space', in Experimental
Phonology (eds. Ohala, J. J. Jaeger), 45-68.
London Academic Press Inc.