Scottish Gaelic (Gàidhlig, gla) is an endangered Celtic language spoken in parts of Scotland. Descriptions of the phonology include either a distinctively nasalized fricative /ṽ/ or distinctively nasalized vowels that may affect neighboring fricatives, but it is questionable whether nasalized fricatives are aerodynamically possible. This leads to the question of how such sounds in Gaelic are pronounced. Historically, Gaelic distinguished oral /v/ (orthographic bh) from nasalized /ṽ/ (orthographic mh) (Gillies 2009: 238), as in the modern words sèimh /ʃɛːv/ 1 ‘calm’ vs. gheibh /jɛv/ ‘will get.’ There are several analyses of this distinction in the modern language, but it is widely agreed that there is some difference between mh and bh involving nasalization at some timepoint, and that Gaelic is likely to have at least this particular nasalized fricative (MacAulay 1992; Ó Maolalaigh 2003) and possibly others (Ternes 2006 ).
However, Ohala (1975) and Ohala and Ohala (1993) point out that if the velum is open by any substantial amount to allow air to flow through the nasal cavity, oral air pressure would be vented through the nose, and therefore air probably would not flow past the point of the oral constriction fast enough to create frication noise. Thus, posited nasalized fricatives might be phonetically something else. For example, nasalization in Gaelic could occur exclusively on vowels, thus maintaining the historical distinction of bh vs. mh without affecting the fricative. This is in fact what Borgstrøm (1941), Gillies (2009), and Bauer (2011) describe. However, some phoneticians have claimed that true nasalized fricatives are possible (Gerfen 2001; and others reviewed in Shosted 2006), and Gaelic has often been named as having them, and as a counterexample to Ohala’s claim (Ó Maolalaigh 2003). The purpose of this paper is to determine how these sounds are produced in this language, how variable their realizations are, and how the potential aerodynamic conflict between nasalization and frication is resolved.
There are several obvious possibilities for how Gaelic speakers might realize the claimed /ṽ/ and other nasalized fricatives. They might produce nasalization on a neighboring segment, such as the preceding vowel, and produce no nasalization on the fricative. They might produce nasalization only on those fricatives that pose no aerodynamic conflict with nasalization, namely those with the place of articulation below the split between the oral and nasal cavities. In these, opening the velum does not vent pressure behind the oral constriction. They might produce nasalization only on consonants that are pronounced as approximants rather than fricatives. In Scottish Gaelic, a single orthographic symbol and historical segment such as bh is realized either as a fricative or an approximant depending on environment (Gillies 2009). That is, historical changes have created a situation in which many consonants that are often grouped with the fricatives are actually approximants. Furthermore, in many languages, voiced non-sibilant fricative phonemes such as /v, ð/ are often variably realized as approximants without clear frication noise, e.g., /ð/, in most tokens of English the. There is no aerodynamic problem with nasalizing during approximants, since air pressure in the oral cavity does not have to be high enough to produce frication noise. Therefore, Gaelic speakers might nasalize during the target consonants, but only for those that present no aerodynamic conflict, such as /h/ and approximants. Alternatively, Gaelic speakers might simply neutralize the fricative nasalization distinction entirely. Or, finally, they might be producing true nasalized fricatives despite the apparent aerodynamic problem.
Before turning to the current experiments, we will review in more detail what has been claimed about these consonants in Scottish Gaelic. Gillies (2009) describes historical Early Irish, the ancestor of Scottish Gaelic, as having had a three-way distinction between /f, v, ṽ/, which neutralized to /f, v/ and to approximant realizations such as /w, j/ in the modern language. He also states that stressed vowels have a nasalization distinction, which developed in the environment of a historical nasal consonant (e.g., /m, n/, etc.) or historical /ṽ/. That is, he describes modern Scottish Gaelic as having distinctively nasalized vowels, and he does not mention any nasalization occurring on consonants other than nasal stops such as /m, n/, etc. For sèimh this account would give /ʃɛ̃ːv/. This is similar to the descriptions given by Borgstrøm (1941) and Bauer (2011), although they leave some ambiguity about environments. For the same environments, Bauer describes orthographic mh as an approximant “with nearby nasality” (2011: 504), and he states that nasalization of historical mh is neutralized in some words and maintained in others. Descriptions of Scottish Gaelic describe various dialects, usually with each source documenting a particular dialect. Thus some variability among sources is to be expected.
MacAulay (1992), however, lists /ṽ/ as a phoneme of Scottish Gaelic, and does not posit any distinctively nasalized vowels. Presumably vowels neighboring a nasal consonant, whether it be a nasal stop like /m, n/ or whether it be /ṽ/, are nasalized by assimilation, as in other languages. MacAulay’s description makes nasalization a distinctive property of consonants, not vowels, so that /ṽ/ would be expected to be nasalized in some way. Stewart (2004) describes mh as distinct from bh in that mh causes nasalization on the surrounding vowels, but the consonant itself is realized as [v] for both (2004: 15). Ó Maolalaigh (2003) addresses sporadic historical developments, rather than making a general claim about the realization of mh. However, he states specifically that Ohala (1975) “is of course incorrect, as the Gaelic languages provide ample testimony for the existence of this phone” ([ṽ]; 2003: 120).
Ternes (2006), working on Applecross Gaelic, 2 gives a somewhat surprising analysis. Like Gillies (2009), he states that Scottish Gaelic has distinctively nasal vowel phonemes, and that there is no nasalized /ṽ/ phoneme. However, he makes the novel claim that nasalization spreads both leftward and rightward from nasal vowels until it hits either an oral stop or the word boundary. This leads to examples like friamh [f̃r̃ĩãːṽ] ‘root’ and amhach [ãh̃ũç̃] ‘neck’ (both from Ternes), with nasalization throughout the word, including on two fricatives in each word. Ternes states that there are no distinctively nasal upper-mid vowels corresponding to oral /e, o/. Bauer (2011) and Gillies (2009) state no such limitation, and give examples that may conflict with it, although dialectal variability in the vowels makes this unclear. Presumably in Ternes’ account, /e, o/ before mh would not be distinctively nasalized, so nasalization would not spread throughout the word if the stressed vowel is /e, o/.
The online dictionary Faclair Beag, developed by Bauer and MacDhonnchaidh (2014), and not representing any single particular dialect, does not mark nasalization on vowels except variably in the environment of an orthographic mh. For example, it gives cumha ‘lament’ as /kũ.ə/ and snàmh ‘swim!’ as /sNã:v/, 3 but deamhas ‘shears’ as /dʲɛu.əs/. For the two examples from Ternes above, it transcribes both words with no nasalization in them whatsoever, despite the orthographic mh. Thus, the Faclair transcriptions variably neutralize the historic mh nasalization distinction. One other source, Ladefoged et al. (1998), does not mention a nasalization distinction either on vowels or on /ṽ/, but their work does not address the mh cases, so this may be an omission from the paper rather than a claim of neutralization.
We turn now to the phonetic debate about aerodynamics and nasalized fricatives, in other languages or in the vocal tract in general. Ohala (1975) and Ohala and Ohala (1993) claim that aerodynamic constraints make nasalized fricatives unlikely. Substantial frication noise generated at a constriction in the oral cavity can only occur if there is high air pressure behind the oral constriction and a very narrow opening at the constriction. This combination results in a considerable volume of air flowing quickly past the surfaces of the vocal tract at the constriction, creating aperiodic noise. If the velum is lowered by any substantial amount, air will flow out the nasal cavity, resulting in lower oral pressure behind the constriction, which means that less air will flow through the oral constriction. That air flow would likely not be enough to generate frication noise. Solé (1999) uses tubes of a variety of diameters, inserted between the cheek and teeth and then around the back molars, to vent oral pressure artificially while speakers attempt to produce fricatives. She finds that a very small air vent of 7.9 mm2 does not alter the percept of frication noise, but that an opening of 17.8 mm2 does. However, she finds that it might still allow some frication for voiceless non-sibilant fricatives, although sibilants attempted with this size vent sounded non-sibilant, and attempted voiced fricatives sounded like approximants.
Shosted (2006) points out that the lowest amount of velic opening that has been reported to produce somewhat perceptible nasalization is 10 mm2 (Warren et al. 1993). This threshold applies to consistently hypernasal speech in a clinical setting, not to a single segment in unimpaired speech, so it is not clear that this is comparable. Thus, at most there could be a range of velic opening from 10 to 17.8 mm2 in which it is possible to have both perceptible nasalization and oral frication, but probably with altered oral frication quality, and probably only for voiceless non-sibilants. Shosted (2006: 99) provides results on artificial fricatives that further suggest oral frication quality might be affected in this range. Producing such segments consistently would require speakers to open the velum to a specific range.
It is important to clarify what types of sounds Ohala (1975) and Ohala and Ohala (1993) do not exclude by claiming that nasalized fricatives are impossible. A phonologically oriented grammar may describe a fricative as nasalized to mean that underlying fricatives become nasal stops in a given environment, e.g., /s/ becomes [n], and this does not refer to a phonetic nasalized fricative at all. A historical description, similarly, could state that historical fricatives became nasalized, meaning they became /m, n, ŋ/. A segment which is phonologically best described as a nasalized fricative phoneme could be produced as nasalization on the preceding vowel followed by an oral fricative. None of these situations poses any aerodynamic problem. Nasalization can be produced on fricatives with the constriction behind the velic opening, as in [h̃], discussed above, and approximants that are the realization of fricative phonemes can be nasalized. Ohala’s claim also leaves open the possibility of fricatives with very low airflow through the nasal cavity, from a very small velic opening: the airflow would go predominantly through the oral constriction, creating frication noise there, with very low velic leakage, possibly not enough for perceptible nasalization (Shosted 2006). The opposite case would be substantial nasalization through a larger velic opening but little airflow through the oral constriction, not enough to create frication noise. In a voiced segment, this would appear as an approximant. In a voiceless one, the amplitude of noise during the consonant would simply be very low, approaching silence.
Several phoneticians have investigated whether nasalized fricatives are aerodynamically possible in various languages. Schadeberg (1982) claims there is a true nasalized fricative [ṽ] in Umbundu. However, Shosted (2006) shows that the evidence for Schadeberg’s claim is rather weak. Gerfen (2001) provides aerodynamic evidence for voiceless fricatives in Coatzospan Mixtec, a language of Mexico. Gerfen’s data include some examples that do have nasal airflow during the consonant, but Gerfen shows only one figure to demonstrate that there can be oral frication along with the nasalization. Furthermore, Gerfen’s (2001)Figure 3 shows that none of the tokens in his work had nasalization throughout the duration of the fricative, and one of three speakers produced nasalization only for a small proportion of the duration at the end of the fricative. These consonants might begin with oral frication and then conclude with nasalization, rather than having perceptible nasalization and frication at the same time. Shosted (2006) details some potential methodological doubts about the interpretation of these results, and Gerfen’s (2001) work is primarily qualitative, but Gerfen does show convincingly that there can be substantial nasal airflow during at least part of some fricative tokens.
Shosted’s (2006) own study uses languages that are not claimed to have distinctive nasalized fricatives, but which have distinctively nasal vowels adjacent to fricatives. Studying one speaker each of Brazilian Portuguese, French, and Hindi, he finds that nasalization can overlap from a distinctively nasalized vowel into a neighboring fricative, but that this has a substantial influence on the frication quality. He suggests that distinctively nasalized fricatives, at least for fricatives that normally have a clear spectral peak, especially [s, ʃ], would only be likely if the language lacks oral fricatives that typically have a broad spectrum, e.g., [h, f]. This is because nasalization tends to reduce the spectral peak of fricatives that normally have one, and Shosted extrapolates that this would make them acoustically similar to the oral broad-spectrum fricatives.
There have been very few aerodynamic studies of languages that are claimed to have distinctively nasalized fricatives, since many of these languages are understudied and spoken in remote locations. Gerfen’s (2001) work is perhaps the only one. We therefore address this issue for Scottish Gaelic. We present aerodynamic data from 14 native speakers, collected on the Isle of Skye, to determine how the potential nasalized fricatives in this language are realized, at least among the dialects of our speakers, discussed below. We also present perceptual data to determine whether listeners are able to hear the distinction, and when in the signal perceptual cues might be located.
Beyond determining whether there is nasalization during the target consonants in Gaelic, we also need to determine whether these consonants are realized as fricatives or approximants. We can then determine what variant Gaelic speakers produce for these sounds (e.g., nasalization on the preceding vowel, neutralization of nasalization, true nasalized fricatives, nasalization only for approximants). The phonology of Gaelic raises two other questions about these sounds. First, in addition to orthographic mh, the other potential source of phonetic nasalized fricatives is through assimilation to a later non-contiguous nasal consonant. One might then wonder whether such consonants have more nasalization if the nasalization is underlying than if it is derived, or perhaps vice versa.
Second, for any fricatives that are nasalized by assimilation, spreading of nasality might be a phonetic process of opening the velum early and leaving it that way, or it might be a more abstract phonological process of spreading nasality to targeted sounds. Ternes (2006), unlike other accounts, claims that the nasalization spreads both leftward and rightward until it reaches either an oral stop or a word boundary. This implies that any fricatives that become nasalized by assimilation should also have nasalization throughout the neighboring stressed vowel, which Ternes describes as distinctively nasal. For example, in a bhean /ə vɛn/ ‘his wife,’ Ternes’ account would predict [a ṽɛ̃n]. Investigating the pattern of nasalization during segments surrounding the target consonant can address this question.
These questions allow us to go beyond investigating the physically possible range of sounds to address phonological aspects of the phenomenon as well. Furthermore, combining aerodynamic production data with perceptual data can provide convergent evidence about whether the distinction is maintained at all, and about what portion of the signal contains any perceptual cues to the nasalization.
2 Production experiment: Aerodynamics
Our first question is the descriptive one of how speakers of Scottish Gaelic produce sounds that have been described as nasalized fricatives, and how they produce the comparable oral fricatives. We turn first to aerodynamic data on oral and nasal airflow.
We chose a total of 120 Scottish Gaelic words containing orthographic fricatives (Appendix A), either ones that are expected to involve nasalization such as deimhinnte /dʲɛviNʲtʲə/ ‘certain’ (with nasalization on either /ɛ̃/ or /ṽ/, depending on the analysis), or corresponding ones that are expected not to, such as cobhair /kovərʲ/ ‘help.’ Some of the phonemes are expected to be realized normally as approximants or as merged with the preceding vowel rather than as fricatives, e.g., cumha /kuə/ ‘mourning’ (could be considered /kũə/) or ghearain /jɛrɛNj/ ‘complained,’ but they were included because they fall into a historical class with the others, and to provide a test of whether the nasalization is maintained on vowels or approximants, where there is no aerodynamic conflict. The target consonants could be in word-initial, -medial, or -final position. Some of the orthographic target consonants may even have been deleted historically, but these may have left nasalization behind on the vowels they merged with. Recognizing the challenges that come with relying on a spelling system, we defined the conditions based on the orthographic consonants because this typically reflects the historical nasalization categories which are the question of interest and because of dialectal variation in the modern pronunciations. The medial and final positions have the nasalization contrast only for /v/ vs. /ṽ/ (orthographic bh vs. mh, respectively, e.g., gabh /gav/ ‘take’ vs. thàmh /hã:v/ or /ha:ṽ/ ‘rest’), and the nasality for consonants in these positions is underlying rather than derived. We refer to the conditions where nasalization is predicted as ‘nasal conditions’ and to the conditions where no nasalization is expected as ‘oral conditions.’
For word-initial position, we chose to use potential nasalized fricatives that are derived through a combination of two phonological processes: mutation triggered by a preceding particle or a grammatical class and nasal assimilation triggered by a nasal consonant after the intervening vowel. For example, the target bh in a bhean /a vɛn/ ‘his wife’ comes from the word bean ‘wife’ (with /b/). The particle a ‘his’ triggers mutation, which changes the /b/ to a fricative /v/. If Ternes (2006) is correct about spreading of nasalization, the final /n/ after the intervening vowel would indirectly trigger nasalization on the word-initial bh consonant. That is, beginning from an underlying /b/, the mutation changes it to a fricative or approximant, and the nasal assimilation gives it nasalization, so that the combination of phonological processes from the left and the right would result in a nasalized fricative. 4 We refer to these consonants also as being in nasal conditions. See Borgstrøm (1941), MacAulay (1992), Ternes (2006), or other sources on the language for a general discussion of Gaelic initial consonant mutation. An oral condition comparison case is do bheul /dɔ vjiaL/ ‘your mouth,’ where the particle do triggers mutation (from beul), but there is no following nasal, so the bh is expected not to be nasalized. For these word-initial cases, the orthography does not mark nasalization on the consonant itself; any nasalization is only marked through the presence of the later nasal consonant.
One further type of word-initial condition is exemplified by a mhamaidh /ə vami/ ‘Mummy (voc.),’ where the nasalization is underlyingly present (cf. nominative mamaidh /mami/), but frication is derived through mutation. Thus, the word-initial mh condition is the only word-initial condition where nasalization is underlyingly present as in the medial and final conditions. In all other word-initial conditions, it is derived. The conditions are summarized in Table 1.
The items included both the palatalized and non-palatalized consonants. (See MacAulay 1992; Gillies 2009; or other sources on the language for discussion of this distinction, which is not always phonetically realized as palatalization.) We group the palatalized and non-palatalized consonants together, e.g., analyzing /v/ and /vj/, non-palatalized and palatalized orthographic bh, together. This is partly because dividing the conditions by palatalization would leave very few items per condition, and also because the predictions regarding nasalization do not differ depending on palatalization. Although consonants with a strong palatal constriction or a neighboring high vowel could potentially have higher nasalization because the close constriction of the tongue to the palate reduces oral airflow, the phonologically palatalized consonants in Gaelic often do not have a clear, strong palatal offglide that would suggest a much closer oral constriction than in consonants without phonological palatalization.
A longer list of potential items was initially chosen from dictionaries, and possible items were then all checked with the native speaker author (Fisher). Many words listed in dictionaries of Scottish Gaelic are archaic and not in modern usage, are used in another dialect, or are too low frequency to be easily recognizable. Fisher rejected doubtful items and suggested alternatives. While the speakers vary in their lexical usage, most of the items that were selected were familiar to most speakers.
Fourteen native speakers of Scottish Gaelic, ages 24–66, participated in this experiment. All but two report that only Gaelic was spoken in their homes at least until they began school, the other two reporting half Gaelic or primarily Gaelic. All continue to use Gaelic in their daily lives. The speakers were recruited through contacts of the native speaker author (Fisher), who is from Skye and who also served as one of the speakers. The speakers include people who use Gaelic in their work, for example in education, media, or the arts. Thus many of the speakers are highly educated about Gaelic, but professions of the speakers also include mechanic and receptionist. Only the youngest two speakers have had substantial Gaelic-medium primary education, with most having exclusively English education, or Gaelic only as a subject in high school. All of the speakers are literate in Gaelic, however.
Most of the speakers are from various parts of the Isle of Skye, where the experiment was conducted. A few are from Lewis, the Uists, or Vatarsay. Dialectal variation is strong in Scottish Gaelic (MacAulay 1992; Ternes 2006), so the speakers’ pronunciations of the items occasionally differ apart from the issue of nasalized fricatives. A few productions could not be measured for nasalization because of dialectal variability affecting the target consonant, as described below.
The 120 items were written in Scottish Gaelic orthography in a randomized list. Speakers participated in this and several other experiments (e.g., Warner et al. 2011; Hammond et al. 2014) during a visit to either the Gaelic language college Sabhal Mòr Ostaig or the Columba 1400 community center in Staffin, both on the Isle of Skye. All of the experiments together took approximately two hours per participant; this one took approximately 15 minutes. Participants received monetary compensation.
The experiment was conducted in a quiet room of the college or community center. Speakers held an oral airflow mask over the mouth and had a separate nasal airflow mask strapped to their head. The airflow and acoustic data were collected by a MacQuirer X16 system (Scicon R&D), set to record for 50 seconds, after which recording was restarted for the next file. Calibration was performed on the airflow system at the authors’ university before departure for fieldwork because of the weight of the calibration equipment, but absolute volume of airflow is not at issue in the measurements below. Speakers read the randomized word list with both masks on, so the resulting acoustic recording is muffled, but sufficient for determining time points at which segments begin and end, and for distinguishing an approximant from frication/silence. However, as is typical with aerodynamic data (Shosted 2006), the acoustic quality is not good enough to reliably distinguish between frication and silence. We did not collect high-quality audio recordings without masks, because one cannot assume that nasalization is consistent token-to-token, and because we could not increase the lengthy experiment session.
Most speakers mispronounced or did not read a few of the items on the list, often because they found a few words unfamiliar. Some speakers occasionally pronounced a word twice, especially at the end of one 50-second recording and then again at the beginning of the next, as it is difficult to be sure in real time whether the recording included the last word said. In such cases, both repetitions were measured. The total number of tokens measured for each speaker varied from 111 to 123.
Oral and nasal airflow traces were low-pass filtered at 30 Hz using a Hanning filter. This is between Zajac and Yates’ (1997) 15 Hz and Shosted’s (2006) 75 Hz; this cut-off smoothed the data while not losing too much information. DC-offset was removed manually by measuring the average raw flow during a time period when the speaker was neither breathing in/out nor speaking, and subtracting that value from all values in the file.
The time periods of the target consonant, its preceding vowel, its following vowel, and the consonant immediately after that were identified (Figure 1). For example in glè mhinig /gle: vɪnək/ ‘very often,’ the mh is the target consonant, and the preceding /eː/, the target /v/, the following /ɪ/, and the /n/ after it were labeled. These four segments were identified in order to assess whether there is nasalization during the target consonant and/or during a neighboring vowel, and whether there is nasalization on either that vowel or the consonant after it, /n/ in this case, that could be the trigger of nasalization.
For labeling these segments, criteria were as follows. The boundary between an obstruent and a vowel was defined as onset/offset of the second formant frequency. The boundary between a nasal stop and a vowel was taken as the sudden discontinuity in energy in the spectrogram. Since the audio microphone in the setup is inside the oral mask, it shows low amplitude for nasal consonants, so the boundary between a nasal consonant and a vowel usually shows a large increase/decrease in amplitude in the waveform as well. The boundary between a glide and a vowel was determined as accurately as possible based on change in amplitude of the voicing, sudden changes in amplitude of the formants as visible in the spectrogram, or failing that, as one-third of the way through the duration of the change in the second formant frequency from a vowel into a glide, and two-thirds of the way through the F2 change from a glide to a vowel. Because many Gaelic consonants have become approximants historically, and because one of the questions is whether nasalization is maintained specifically in approximants, the need to place boundaries between vowels and glides in this data cannot be avoided. For word-final segments, the boundary between speech and following out-breath or silence was identified as a sudden drop in amplitude visible in the spectrogram for final vowels, and as the cessation of the low-amplitude frication noise or nasal vibration in the waveform for final fricatives and nasals. One cannot use the oral or nasal airflow to determine the end of a word-final segment, because it usually continues during the following out-breath. The same criteria in reverse were applied for word-initial segments.
For some of the items, there is no following vowel, i.e., for all word-final target consonants; no consonant after the following vowel (e.g., cumha /kuə/ ‘mourning’); or no preceding vowel, as in word-initial cases where mutation is triggered by a morphological feature such as tense or mood instead of a particle, for example, bhiodh /viu/ ‘would be’. For such items, whichever of the usual four segments existed were located. Because Gaelic orthography is not straightforward, and there is dialectal variation, it was sometimes hard to determine whether a given segment was present but difficult to perceive, or whether it was categorically deleted, perhaps historically. For example, some speakers’ productions of amharc /awərk/ ‘view’ have very little perceptible trace of an /r/, making it difficult to determine whether the consonant after the following vowel is the /r/ or the final /k/. In such cases, Fisher listened to the recording and judged what segment should be labeled, but the data on the consonant after the target consonant is not presented in this paper. Furthermore, the target consonant itself was occasionally completely deleted in a given speaker’s dialect. Fisher also evaluated all such tokens, and 23 tokens out of the total of 1,666 were measured as having no target consonant present, even as an approximant, as for example in abhainn pronounced as /aɪNj/. These tokens were included for analyses of nasalization in the vowels, but were not used for analyses of the target consonant. Such cases are useful despite the absence of the target consonant, because if the target consonant has been deleted historically, it may have left nasalization behind, perhaps leading to distinctively nasalized vowels as suggested by Gillies (2009) and others.
We also measured which portions of the signal during the four segments had elevated nasal airflow. This was done manually because even after filtering, airflow signals contain noise, and automatic measurement would be prone to error. Any portion of the speech with nasal airflow substantially above the zero line that lasted longer and rose higher than in the background noise was considered a period of elevated nasal airflow (Figure 1). The exact time point for onset of elevated nasal airflow was identified as the point when the nasal airflow trace exceeded the range of values appearing in the non-nasalized speech adjacent to (before) the period of nasalization. For offset, the reverse criterion was used, that is the value entering the range of values represented in the speech adjacent to the period of nasalization. This means that the time points of onset and offset of elevated nasal airflow do not necessarily fall at sudden changes in the slope of the nasal airflow signal, but rather are defined based on the range of the adjacent, local, background noise. The degree of uncertainty in visually determining onset and offset of elevated nasal airflow is similar to the uncertainty of other decisions one makes in placing boundaries for hand-measured phonetic dependent variables, e.g., in determining the point of onset of voicing for VOT or the point of offset of F2 for vowel duration.
From this information, several measures were calculated. A categorical measure of presence/absence of elevated nasal airflow at any time during each of the four segments was coded. The duration of elevated nasal airflow within each segment and duration of the segment was also calculated, from which the proportion of each segment containing nasal airflow was calculated. The peak nasal airflow during each of the four segments was also extracted. 5
2.2.1 Examples of outcomes
Quantifying aerodynamic results on nasalization is challenging, because a consonant with nasalization may have slight nasal airflow throughout the segment, high nasal airflow for only 10 milliseconds at the onset of the consonant, probably reflecting segmentation criteria for the vowel-consonant boundary rather than nasalization of the consonant itself, or anything in between. A token may have slight nasalization continuing from a previous sound, or word-final targets may have nasalization overlapping with exhalation through the nose after the speech ends. Because a finding of nasalization for a given proportion of a consonant can mean various things, and because some past papers show few examples, we first present several examples of airflow traces to demonstrate possible pronunciations of the target consonants (Figures 1–3). Figure 1 above shows a case with no elevation of nasal airflow at all during the target consonant, although there is nasalization during the /n/ after the following vowel. This token of the target consonant was produced as an oral fricative with weak voicing, a non-nasalized /v/.
Figure 2 shows several additional examples. Figure 2(a) shows nasalization during a token of /h/ in the nasal condition. This token has strong oral and nasal flow throughout most of the target consonant, as well as strong nasal flow during the /n/, the segment that could trigger nasalization of the /h/. The /h/ is produced as a nasalized voiceless fricative. Nasalization drops back to the zero line during the vowel between the /h/ and the /n/. Figure 2(b) shows an example where the target consonant (nasal condition) is an approximant-like realization of /ɣ/ in glè ghann /gle: ɣauN/ ‘very scarce.’ This token shows no elevated nasal airflow during the target consonant, but has nasalization during most of the following vowel as well as the /N/ after it. Figure 2(c), do pheanas /dɔ fjanəs/ ‘your punishment’ shows a token with strong nasal airflow overlapping just slightly from the following vowel into the final milliseconds of the target consonant. Because onset of F2 is used as the criterion for onset of the vowel, this token would be counted as having nasalization during the target consonant, but clearly this is not a nasalized fricative. Nasalization continues in this token throughout the following vowel, the consonant /n/ after that, and the vowel after that, ending as the word-final fricative /s/ starts. Figure 2(d) shows a token with relatively strong nasal airflow during the preceding vowel that continues through nearly all of the target consonant, with voicing evident in the waveform during the consonant, and no formants visible in the spectrogram. No frication noise is visible in the waveform or spectrogram. The target consonant has no elevation in the oral airflow at all, which could indicate either an absence of frication noise or a failure of the oral mask to seal well to the face 6 during this token. Thus, this token has some nasalization, but at the cost of frication, even though it is not an approximant.
Figure 3 displays three tokens that may have nasalization with frication. Figure 3(a) shows a word-final /v/ that begins with strong oral airflow indicative of frication, then has nasal airflow beginning partway through the segment. The onset of nasal airflow coincides with a drop in oral airflow and a loss of amplitude of frication in the waveform. This is a biphasic segment, with oral frication followed by nasalization, not both at the same time. Both oral and nasal airflow continue as the speaker breathes out. Figure 3(b) and (c) show two cases that could perhaps be true nasalized fricatives. Both have nasal airflow throughout the consonant (weak in 3b, strong in 3c), and both have at least a slight elevation of oral airflow, suggesting frication, during at least part of the consonant. Neither has approximant-like formant structure in the consonant, and 3c is voiceless. However, in 3c, the speaker begins the nasal airflow well before the word begins, suggesting this word-initial fricative is the continuation of breathing out through the nose preparatory to starting talking rather than a velic opening gesture timed to a segment.
2.2.2 Quantitative measures of nasalization during the target consonant
We turn now to quantitative analysis of how often nasalization occurs during target consonants and how strong it is. We will examine the categorical variable of whether nasalization occurred during a given consonant (proportion of tokens with nasalization) rather than gradient measures such as peak nasal airflow during the target consonant. We choose to focus on the categorical measure because relatively few tokens were produced with nasalization during the consonant at all, making statistical analysis of peak nasal airflow problematic, and because of potential problems with scaling of the data from separate recordings and speakers. However, preliminary analysis of peak nasal airflow largely supported the patterns discussed here for proportion of tokens with nasalization.
Each token was evaluated for whether any time period identified as having elevated nasal airflow occurred during the target consonant. Overall, only 22.9% of the target consonants in nasal conditions had nasal airflow at any time during the consonant. Of target consonants in oral conditions, 8.3% had nasal airflow during some part of the consonant. While it might seem surprising that so many oral consonants have nasalization during them, this serves as a comparison condition to show how often the measurement methods will detect some elevated nasal airflow during a consonant that is not considered nasalized at all, whether through overlap with nasal airflow during the outbreath at the onset or offset of speech (Figure 4), frequent excessive nasalization by a speaker, coarticulation with a neighboring segment, or other reasons.
Figure 5 shows the proportion of target consonants with nasal airflow, for each orthographic segment and environment. In word-initial position for segments that have both oral and nasal conditions, nasalization is numerically more common in the nasal condition than in the oral condition in every pair. For the statistical analysis, variance is unequal across segments because some conditions have very few tokens with nasal airflow, a floor effect. Therefore each oral vs. nasal consonant pair was analyzed separately using a one-factor within-subjects ANOVA. 7 Data was first averaged over items, to give the proportion of all items a given speaker produced with nasal airflow during the consonant. 8 In these word-initial pairs, the nasal condition had nasal airflow significantly more often than the oral condition for all but two pairs (significant: bh: F(1,13)=14.34, p<0.005; gh: F(1,13)=5.20, p<0.05; sh: F(1,13)=15.83, p<.005; th: F(1,13)=35.00, p<0.001; non-significant: ph and dh both F<1). One of the two with no significant difference, dh, has a floor effect, so that there is very little variation to be tested.
The mh initial consonant cannot have an oral counterpart because nasal status is underlying for this consonant, but it shows nasal airflow as often as nasal condition bh, its closest match (F<1 for comparison of these two conditions). Nasal ch (/x/) does not have an oral counterpart because of accidental omission from the word list, and this segment will be discussed below.
For word-medial and word-final position, the consonants to compare are mh (nasal condition) and bh (oral condition). A two-factor within-subjects ANOVA, with Word Position (medial, final) and Nasal Status (mh, bh) showed a significant main effect of Nasal Status (F(1,13)=21.66, p<.001) and a significant interaction (F(1,13)=7.42, p<0.02), but no main effect of Word Position (F<1). The simple effect of Nasal Status was significant in both word positions (medial: F(1,13)=22.97, p<.001; final: F(1,13)=4.82, p<0.05). Thus, in all word positions, consonants in the nasal conditions are more likely to have nasal airflow than those in the oral conditions (non-significant only for ph, dh).
A second question one can ask with this data is whether nasal airflow during consonants is more likely when there is no aerodynamic conflict than when there is. Two types of target consonants have no aerodynamic conflict with nasalization: /h/ because its source of frication is behind the velic opening, and approximants. The dh, gh orthographic consonants are realized as approximants or as [ɣ], which is often approximant-like with little frication, and sh, th are categorically /h/. For ch (/x/), it is not clear exactly how far back the oral constriction is, 9 and thus whether it would pose a problem for nasalization or not. We will group ch with sh and th as back fricatives.
To address these questions of place and manner, we will use only the nasal conditions, not the oral comparison conditions. Among the three back fricative nasal conditions (ch /x/, sh /h/, th /h/), th shows nasalization significantly more often than either ch (F(1,13)=43.56, p<0.001) or sh (F(1,13)=10.43, p<0.01). ch and sh do not differ significantly from each other (F(1,13)=3.95, p=0.07). In order to control the number of pairwise comparisons, we will compare other consonants to ch, the lowest value of the three, to determine whether they are significantly different from the back fricatives. Each of the non-back consonants that is often strongly fricated shows significantly less frequent nasalization than ch (ph: F(1,13)=32.74, p<0.001; mh: F(1,13)=33.49, p<0.001; bh: F(1,13)=18.12, p<0.005). Thus, we can conclude that the categories that are often fricatives, not primarily approximants, and that have constrictions clearly in front of the velum are less likely to have nasalization than back fricatives are.
The consonants that are very often realized as glides, dh, gh, have the least frequent nasal airflow of all conditions. Even when these are /ɣ/ rather than /j/, the /ɣ/ is often produced without loud frication noise, somewhat similar to an approximant. To control the number of pairwise comparisons, we compare them just to ph. The ph category is a representative of the non-back fricatives with relatively low frequency of nasalization, and it is consistently realized as a fricative, never as an approximant. Neither dh nor gh differs significantly from ph (dh: F(1,13)=1.53, p>0.1; gh: F<1). Both do have significantly less frequent nasalization than ch (dh: F(1,13)=64.33, p<0.001; gh: F(1,13)=35.41, p<0.001).
Thus, the back fricatives, realized as /x/ or /h/, have nasalization during the consonant more often than either the front fricatives that are variably realized as approximants or the predominantly approximant consonants. The predominantly approximant consonants (dh, gh) are numerically even less likely to have nasalization than the front fricative/approximants, but this difference is not significant. Statistically, all of the nasal conditions have comparably rare nasalization except the back fricatives.
A third question one can ask based on this data is whether the mh consonant, MacAulay’s (1992) /ṽ/, which has nasalization underlyingly or historically, is more likely to be nasalized than consonants that become nasalized allophonically. To answer this question, we compare the nasal category of word-initial bh (allophonic nasalization), the word-medial mh, and the word-final mh (the latter two underlying). These categories differ significantly (F(2,26)=4.47, p<0.03). However, it is not clear what the grouping is within these three categories: word-initial bh has nasalization significantly less often than word-medial mh (F(1,13)=18.41, p<0.005), but the word-initial category does not differ significantly from the word-final category (F(1,13)=2.96, p>0.1), and the medial and final categories do not differ from each other (F<1). Thus, nasalization is somewhat more common where the target segment mh is underlyingly (MacAulay 1992) or historically (Gillies 2009) nasal, but there is no clear statistical grouping of those two conditions as separate from the derived word-initial condition. See Section 2.2.4 below for discussion of a particular type of realization common word-finally for one speaker.
2.2.3 Nasal airflow in other segments
If the target consonant itself is not nasalized, has nasalization simply been shifted somewhere else? Some analyses of Scottish Gaelic claim that the nasalization historically associated with the mh segment is realized on the neighboring vowel, not on the consonant itself, which is sometimes even deleted (Borgstrøm 1941; Ternes 2006; Gillies 2009). Even if the segment is underlyingly /ṽ/, with the nasalization phonologically a feature of the consonant, one way to avoid the aerodynamic problem would be to produce the nasalization on a neighboring segment, most likely the preceding vowel. Furthermore, for the word-initial consonants other than mh, e.g., a dhàn /ə ɣa:n/ ‘his poem’, nasalization is only expected on the target consonant if it has spread from a following segment (the vowel if it is analyzed as distinctively nasalized, here /ãː/, or the /n/ after it). Thus for word-initial nasal conditions other than mh, nasalization is expected on the following vowel (Ternes 2006).
To determine whether nasalization appears on a neighboring segment in addition to or instead of the target consonant, we classified each token by whether it had elevated nasal airflow during the target consonant, the preceding vowel, both of those, neither of those, or during the following vowel. As mentioned above, segments sometimes have elevated nasal airflow only during a few milliseconds at the onset or offset of the segment, usually if the neighboring segment has nasal airflow, where this reflects segmentation criteria rather than nasalization of any substantial portion of the segment. For this analysis, we wanted to avoid counting nasalization during segments that it only slightly overlapped into. Therefore, we examined whether there was elevated nasal airflow for at least 15% of the duration of each segment. This is an arbitrary cut-off point chosen to exclude tokens where nasal airflow overlapped slightly from a neighboring segment without excluding tokens where the target consonant itself could be described as a nasalized consonant. Of all tokens with elevated nasal airflow during any of the target consonant, 15% fell below this cut-off and were excluded.
Figure 6 shows the counts of how many tokens in the nasal conditions have elevated nasal airflow during 15% or more of both the target consonant and preceding vowel, neither of those segments, or one but not the other. The following vowel will be considered separately below. For every condition except th, the most common outcome is to have no nasalization greater than 15% of duration in either the preceding vowel or the consonant. Furthermore, the option of having nasalization only during the preceding vowel, shifting the nasality to a preceding segment that does not conflict aerodynamically with it, is only common in the word-medial and word-final conditions, and not very common even there. Nasalization does not shift onto the preceding vowel across the word boundary in word-initial conditions, in line with Ternes’ (2006) prediction. Even word-initial mh, where the nasalization is a property of this consonant rather than of following segments, does not have this outcome. Finally, Figure 6 shows that the realization with nasalization only during the target consonant is only common for /h/ (sh, th).
Another possibility is that nasalization might be on the target consonant’s following vowel. For the word-initial nasal conditions other than mh, as in a dhàn /a ɣa:n/ ‘his poem,’ this is in fact what the analyses that posit distinctively nasalized vowels (Borgstrøm 1941; Gillies 2009) predict. Ternes (2006) predicts nasalization on both the target consonant and the following vowel. However, almost all items in these conditions have a nasal stop such as /m, n, N, Nj/ as the next consonant to the right, for example the /n/ in a dhàn /ə ɣa:n/ ‘his poem.’ In these words, the vowel after the target consonant is likely to be partially nasalized just by phonetic assimilation to the nasal stop after it, as it would be in most if not all languages, and this has nothing to do with the target consonant. Furthermore, word-final target consonants of course do not have a following vowel. Therefore, we examined the word-medial nasal condition in order to determine whether nasalization was shifted to the following vowel instead of the target consonant. Of the 131 medial mh tokens among all speakers, 20 had nasal airflow during at least 15% of the following vowel, but 15 of those also had a nasal following consonant, e.g., the /Nj/ in deimhinnte /dʲɛviNʲtʲə/ ‘certain,’ that accounts for allophonic nasalization of the vowel. Thus, in only 5 of these tokens did speakers produce the nasalization for the mh segment on the following vowel instead of the target consonant.
The word-initial mh condition provides another environment where we can test whether the expected nasalization is shifted from the target consonant to its following vowel, because the nasalization is also underlying in this case. Some of these words also have a nasal as the next consonant after the target consonant, e.g., a mhuinntir /ə vɯiNʲtʲirʲ/ ‘his people,’ meaning that nasalization would be expected on the intervening vowel regardless of whether the target consonant contributes any nasalization or not. In this condition, 20 tokens out of the total of 154 have nasalization during at least 15% of the following vowel but not during either the preceding vowel, which is across a word boundary, or during the target consonant, and do not have a nasal stop as the next consonant. Thus, in the entire dataset containing 951 tokens in nasal conditions, only 25 tokens indicate that the speaker might have shifted the nasalization of the target consonant onto the following vowel instead. Eight of those are tokens of glè mhoiteil /gle: vɔtjəl/ [vɔhtʃəl] ‘very proud’ (Figure 7), which has strong preaspiration on the medial consonant. In this case, the nasal airflow occurs during the latter part of the vowel /ɔ/ and the preaspiration, thus not adjacent to the target consonant. This may be a case of spontaneous nasalization on segments with high oral airflow such as preaspiration, because of acoustic similarity to nasalization (Ó Maolalaigh 2003, based on earlier work by Ohala), and may or may not have any relationship to the mh segment earlier in the word.
To sum up the investigation of which segments other than the target consonants might carry nasalization, by far the most common outcome is to neutralize the nasalization distinction, with these consonants neither being nasalized themselves, nor introducing nasalization onto a neighboring segment. Thus, the low percentages showing nasalization during the target consonant (except for /x, h/ targets) in Figure 5 above do not reflect a shift of nasalization onto the preceding or following vowel instead of the consonant, but instead reflect lack of nasalization with fricatives.
2.2.4 When there is nasalization, is there also frication?
The results reported above all address the question of whether there is nasalization in the words that are expected to be nasalized, and if so, when. However, another part of the question is whether any target consonants with nasalization also have oral frication, particularly for those with place of articulation in front of the velum, the buccal fricatives. One possible alternative is that all the tokens with substantial nasalization are approximants, either because speakers intentionally pronounce nasalized approximants or because the venting of air pressure through the nasal cavity prevents air from making frication noise at the oral constriction. A given consonant might have nasalization or frication, but not both. This question comes down to whether there are tokens that are true phonetic nasalized fricatives, despite the expected aerodynamic problem.
To determine whether there are such tokens, we turned to a qualitative analysis of subsets of tokens that offer the opportunity for this combination to occur. Since dh and gh are very often realized as glides rather than fricatives, and even fricative tokens of them may have only weak frication, those conditions are unlikely to contain any tokens with a true phonetic nasalized fricative. Similarly, ch, th, and sh are out of consideration, since the place of frication is near or behind the split to the nasal cavity. Tokens with extremely little nasalization at all cannot be phonetic nasalized fricatives, and tokens with nasalization only at the boundary with a neighboring segment, as discussed above, probably reflect segmentation criteria rather than a true nasalized fricative segment. Furthermore, tokens like Figure 3(a) show a biphasic realization, with oral frication first, followed by nasalization later. This realization was common word-finally for one speaker. Therefore, we examined all word-initial and word-medial tokens in the nasal conditions for ph, mh, and bh that have elevated nasal airflow during more than 15% of the duration of the target fricative and that have a peak nasal airflow value during the target consonant of more than 0.08. A histogram showed that this cut-off is a reasonable value in between tokens with minimal vs. strong nasal airflow. There are 25 tokens in the dataset that meet these criteria out of the 538 initial or medial mh, bh, and ph tokens, and these 25 should be the tokens with the greatest chance of showing a true phonetic nasalized fricative with place of articulation in front of the velum.
For each of these tokens, we examined the airflow data, waveform, and spectrogram visually to determine whether the target consonant might be a true nasalized fricative. Some tokens were approximants, as shown by voicing and clear formant structure similar to Figure 2(b). Some other tokens had oral airflow not even elevated above that of the neighboring vowels as in Figure 2(d) and no clear frication in the waveform or spectrogram, indicating that the consonant was at most very weakly fricated. 10 Only seven tokens could perhaps be classified as true nasalized fricatives, meaning that they were not approximants and they had at least some elevation of oral airflow during at least part of the consonant. Figure 3(b) shows one of the two most likely examples of a true nasalized fricative, and even in that case, nasal airflow decreased as oral flow increased, demonstrating the trade-off between air venting through the nose and passing through the oral constriction. Figure 3(c) shows the second good example, but as mentioned above, this token had the nasalization as a characteristic of starting up speech rather than being timed to the target consonant, so nasalization may not be a feature the speaker intentionally realized on the consonant. This qualitative analysis demonstrates that if speakers produce nasalization together with oral frication at all, they do so rarely and with a reduction of both airstreams.
2.2.5 Is allophonic nasalization in consonants phonetic or phonological spreading?
In words such as do phuing /dɔ fɯiŋʲgʲ/ ‘your point’ or a’ ghainmheach /a ɣɛnɛəx/ ‘the sand,’ any nasalization present on the target consonant comes about through spreading of nasalization from the following vowel or the consonant after it, as discussed above. When these initial consonants are in fact nasalized, one might wonder whether this spreading of nasalization is a matter of phonetic coarticulation, or an abstract phonological process of spreading the feature of nasalization to the earlier consonant.
To address this question, we examined all tokens in word-initial nasal conditions, except mh, where nasalization is not allophonic, that had elevated nasal airflow during at least 15% of the duration of the target consonant. For these tokens, we analyzed the proportion of the intervening vowel’s duration that had nasalization (for example the /ɯi/ in do phuing). If nasalization spreads phonetically leftward from the following /n, m/ or other nasal consonant, and spreads far enough to affect the target consonant, then there should be nasalization throughout the intervening vowel. This would be the case if the nasalization is simply a matter of lowering the velum early and leaving it open. Figure 8 shows how often what proportion of the intervening vowel was nasalized. Only sh and th, both realized as /h/, have many tokens with nasalization throughout the intervening vowel, and all conditions have tokens where nasalization stops for some portion of the vowel, as in Figure 2(a) above. Thus, the nasalization is not simply spreading phonetically leftward from the following consonant in most cases, since it usually misses a portion of the vowel even when nasalization is present in both consonants. Instead, this could be a phonological spreading process, where the nasalization targets specifically the preceding consonant without making the intervening vowel nasalized. Alternatively the tokens with a gap in nasalization during the intervening vowel could have nasalization on the target consonant for some other, unrelated reason (perhaps like the sporadic nasalization with high airflow or with /h/ [rhinoglottophilia] discussed by Ò Maolalaigh 2003).
The results of this production experiment demonstrate that nasalization during the target consonant is not a common option for any of the Gaelic consonants that have sometimes been described as nasalized fricatives, either orthographic mh or fricatives preceding a distinctively nasalized vowel. Rather, there are several possible ways to realize these sounds, with complete loss of nasalization the most common. However, a minority of tokens demonstrate nasalization in a neighboring segment or for a portion of the target consonant. A very few tokens may contain true nasalized fricatives, but with a trade-off of volume in the two airstreams. However, these results do not address whether the nasalization is sufficient for listeners to perceive it and maintain a perceptual distinction between, for example, /v/ and /ṽ/ (orthographic bh and mh).
3 Perception experiment
To address the question of whether the variable differences in production are sufficient for a perceptual distinction, we performed a phonetic identification experiment in which we presented various portions of words similar to those in the production experiment, and asked listeners to choose which of two spellings was a better match for the partial word they heard. For example, we presented portions of the word càmhal /kaːvəL/ 11 ‘camel’ and asked listeners to judge whether what they heard was a better match to …àmha… or …àbha…. We used pairs of words, with the vowels preceding and following the target consonant matched as closely as possible. For example, tàbhachd /taːvəxk/ ‘solidity’ formed a pair with càmhal and had the same response options. Only portions of the words were presented in order to determine the location of perceptual cues, and to prevent listeners from recognizing the word and responding based on lexical information, since the language has very few minimal pairs for this.
Since most tokens in nasal conditions in the production experiment were produced without any elevation in nasal airflow, and the tokens with nasalization might not all have perceptible nasalization, it is possible that listeners would be unable to perceive the distinction, even though speakers do produce nasalization more often in the nasal than oral conditions. That is, the current generation of speakers could be still marginally producing the distinction, even if no one can hear the variable distinction they produce. The current experiment allows us to address this question.
Nine pairs of Scottish Gaelic words were selected (Appendix B), with one containing a fricative (or approximant) where nasalization might be expected, e.g., càmhal ‘camel,’ and one where it is not, e.g., tàbhachd ‘solidity.’ The definition of these conditions is the same as for the production experiment above. The surrounding environment was matched as closely as possible, within the constraint of using readily familiar words. Since the words are not presented in pairs, it is not necessary for the environment to match exactly. As with the production experiment, words were initially selected from a dictionary, and then checked with the native speaker author (Fisher). Fisher was recorded at a sampling rate of 44.1 kHz reading the words in a sound-protected booth at the authors’ university, using a high-quality head-mounted microphone. Later examination of the production data above for individual speakers verified that the difference in how often this speaker produces nasalization during nasal vs. oral target consonants is approximately average relative to the other speakers in the production experiment: she neither makes an exceptionally consistent nor an exceptionally neutralized distinction.
Three portions were excised from each word with an intervocalic target consonant, and two from each word with a word-final target consonant. For this purpose, consonants that are word-initial but after a particle, such as glè shàmhach /gleː haːwəx/ ‘very quiet,’ are also intervocalic. The three portions were the preceding vowel only, e.g., /eː/ in glè shàmhach, the preceding vowel plus the target consonant, e.g., /eːh/, and the stretch from halfway through the preceding vowel to halfway through the following vowel, e.g., /eːhaː/ but with only half the duration of each vowel. We will refer to these as the V, VC, and VCV conditions. The reason to cut at vowel midpoints for the VCV condition is that using the entire VCV might provide enough coarticulatory cues for listeners to recognize the word and use lexical information in choosing their response. Using these three portions allows us to determine when in the signal information about nasalization is perceptible, for example, whether nasalization is perceived through the preceding or following vowel, or whether nasalization during the consonant itself is a perceptual cue. There might be other portions of the word one might wish to test, but we could not extend the total experimental time.
For word-final consonants as in gabh /gav/ ‘take!’, the two portions used were the preceding vowel only, e.g., /a/ in gabh, and the preceding vowel and target consonant, but starting at halfway through the duration of the preceding vowel. 12 Thus, word-final target consonants had only the V and VC conditions. Again, halfway through the vowel in the VC condition was used as the stimulus onset to prevent listeners from recognizing the lexical item. The boundary between a vowel and an obstruent was defined by onset and offset of F2 in the spectrogram. Vowels in many of the recordings had somewhat breathy voice quality at the onset or offset of the vowel, making the F2 less clear during a portion of the vowel. Such breathy portions were considered to be part of the vowel. For the boundary between vowels and glides, a sudden change in the amplitude of F2 was visible in all stimulus recordings, and this point was identified as the onset/offset of the vowel.
All portions were extracted from the recording of the word, and the amplitude of the waveform was ramped up linearly from 0 over the first 5 ms and ramped down to 0 over the final 5 ms. This avoids introducing perceptual artifacts by cutting suddenly to or from 0 (Smits et al. 2003).
Seventeen native speakers of Scottish Gaelic began participation in this experiment, but one was excluded because of equipment failure and one because of lack of literacy in Gaelic, so data for 15 listeners can be analyzed. These participants were a superset of those for the production experiment, except that Fisher did not participate in this experiment. The additional participants, tested on days when the airflow equipment was not available, had similar characteristics to those in the production experiment.
The stimuli for this experiment were randomized together with similar stimuli for several other phonological distinctions of Gaelic, e.g., preaspiration, palatalization, and epenthetic vs. underlying vowels (Warner et al. 2011). For each of these distinctions, similar portions of Gaelic words were used as stimuli, and similar response options were presented. Thus, randomizing the stimuli for these related experiments together with the nasalization stimuli for the current experiment served to provide filler items so that not all stimuli involved a nasalization contrast. The total number of stimuli, including those for other distinctions, was 166. Five similar practice items were also constructed. The stimuli were presented in the same random order for all participants.
This experiment was conducted at the same locations as the production experiment. Listeners participated during the same visit, but the two experiments were separated by a break. Each listener was seated in front of a laptop in a quiet room and heard the stimuli over headphones. The EPrime software was used to present stimuli and record responses. With each auditory stimulus, listeners saw two partial word spellings on the laptop screen, one representing a target consonant in the nasal condition and one in the oral condition, as explained above. For example, the response alternatives for càmhal ‘camel’ were …àmha… and …àbha….. Ellipses were used around the responses to indicate that the stimuli were part of a longer word that listeners would not hear in its entirety. The response alternatives were matched to each stimulus in spelling and word position, so for example the response alternatives for gabh /gav/ ‘take!’ were ...abh and ...amh. For items where the target consonant would only be allophonically nasalized by assimilation to a later nasal consonant, enough of the orthography of the word was presented to indicate the nasalization, e.g., for glè thana /gleː hanə/ ‘very thin,’ the response alternatives were ...èthad... and …èthan... The n in this case is the trigger for nasalization. The stimuli for other distinctions presented in the same session had similar response alternatives, making the task consistent throughout.
Listeners heard the stimuli over headphones, and then chose which response alternative they thought better matched the stimulus. Most listeners used a button box to respond, but because of an equipment failure, the final five listeners responded by pressing keys on the laptop instead. If a listener failed to respond to an item, the next stimulus followed after a 6-second time-out. This occurred for approximately 3.5% of trials. The entire experiment took approximately 10 minutes.
Results of the perception experiment, presented as the proportion of stimuli for which listeners chose the nasal response, appear in Figure 9. We examine the results in terms of the proportion of nasal responses because this allows us to determine whether listeners are more likely to perceive the stimulus as matching a nasalized consonant when it actually comes from a word in the nasal condition than when it comes from a word in the oral condition. Figure 9 shows that listeners are slightly more likely to choose the orthographic nasal response for stimuli in the nasal condition than the oral condition.
Because only two time portions were presented for word-final consonants, but three were presented for word-medial consonants, we analyze the data using two separate statistical designs. One uses the V and VC portions of both word positions, with the three factors Nasal Status (nasal, oral), Portion Presented (V, VC), and Word Position (medial, final). The other uses only the word-medial conditions, with the two factors Nasal Status (nasal, oral) and Portion Presented (V, VC, VCV). Both analyses use by-subjects and by-items ANOVAs. All factors are repeated measures in the by-subjects analysis. For the by-items analysis, Nasal Status and Portion Presented are repeated measures.
For the three-factor design (using only V and VC portions), listeners chose the nasal response significantly more often for nasal conditions (significant main effect of Nasal Status, F1(1,14)=7.69, p<0.02, F2(1,7)=6.57, p<0.04). The main effect of Portion Presented was significant only in the by-subjects analysis (F1(1,14)=6.04, p<0.03, F2(1,7)=4.20, p=0.08), reflecting slightly more frequent choice of the nasal response if the consonant was heard in addition to the preceding vowel. However, there was no interaction between Nasal Status and Portion Presented (F1 & F2<1), and neither the main effect of Word Position nor any of the other interactions were significant either (F1 & F2<1). Thus, listeners did not become better able to distinguish nasalized from non-nasalized target consonants on hearing the consonant itself; they were equally able to hear the distinction based on just the preceding vowel. This could reflect the fact that nasalization can overlap slightly across segment boundaries, as shown in the perception experiment above. Since vowel-glide-vowel sequences have to be included because of the nature of the target consonants, overlap across segmental boundaries is particularly likely.
For the two-factor design (using only medial target consonants), the main effect of Nasal Status was significant by subjects and nearly significant by items (F1(1,14)=14.33, p<0.005, F2(1,4)=7.26, p=0.054). Since there were only 5 item pairs in the medial condition, this by-items analysis has very low power. The main effect of Portion presented among the medial conditions was significant only by subjects (F1(2,28)=5.23, p<0.02, F2(2,8)=1.11, p>0.05), and the interaction was not significant (F1 & F2<1).
The production data above showed sporadic realization of nasalization in the nasal conditions, with only a minority of tokens having nasalization at any location near the target consonant. Furthermore, it showed that nasalization on nasal condition consonants is more common for [h]. Thus listeners’ ability to perceive nasalization in the target consonants could be highly item specific, as well. This could lead to the significant but rather small effect of Nasal Status we see in the perception results. However, it is also possible that the speaker for the stimuli produced nasalization relatively consistently when recording items for the stimuli. Overall, listeners identify the Nasal Status of the stimuli correctly in 57% of the trials they responded to (55% of all trials). Is this actually an average over responses to a few items with highly perceptible nasalization and many stimuli with the nasalization distinction neutralized?
To determine this, we calculated the difference between average proportion nasal responses to the nasal condition stimulus and to the oral condition stimulus for each pair, averaged over the portion presented (V, CV, VCV). The three pairs with the largest difference, indicating the most ability to perceive the distinction, were the pairs with còmhla, glè shamhach, and glè sheimh. This represents two pairs with /h/ and one with the target consonant merged into the vowel. The difference in proportion nasal responses to the nasal vs. the oral condition for these three pairs is 0.37, 0.30, and 0.26, respectively, indicating moderate perceptibility of the difference (dʹ (detectability)=0.96 for the largest difference). However, except for two pairs, those containing riamh and càmhal, each pair showed a difference in proportion of at least .15 (dʹ=0.38 for the weakest pair among these). Thus, the ability to perceive nasalization in the target consonants, although weak overall, is not limited to a few word pairs or to one particular phonological environment. The effect is not driven by a few exceptional words or tokens.
One might then wonder whether the overall accuracy rate reflects differences among listeners, even though the ability to perceive the distinction is significant, especially considering the dialectal variability within Scottish Gaelic. The results of the perception experiment do not appear to reflect dialectal differences: of the five listeners who were least accurate to perceive the distinction, all were raised at least partly on Skye, as was the speaker of the stimuli. Only one of these listeners was also raised partly on North Uist. Two listeners who perceived the distinction very well come from the islands of Lewis and Vatarsay, where the dialects differ more from the stimuli (Borgstrøm 1941). The pattern for the perception results also does not reflect age in any obvious way: both the most and least accurate listeners include listeners in their 60s–70s as well as a listener below age 30.
The results of the perception experiment show that Scottish Gaelic listeners are able to perceive the distinction between fricatives that are sometimes claimed to be nasalized and oral fricatives significantly better than chance, but their perception of the distinction is rather weak. One would expect accuracy for a stable distinction of the language to be much higher than an average of 57% correct. This could perhaps simply indicate that the participants, who are for the most part not used to participating in button-press response experiments about their language, had difficulty with the task. However, results for the other distinctions in the language that were tested in the same session such as palatalization and preaspiration rule that explanation out: average accuracy across all portions presented for the palatalization distinction was 77% correct, for example (Warner et al. 2011). Thus, it appears that listeners are indeed able to perceive the distinction at a level significantly better than chance, but not consistently and reliably.
The lack of an interaction between nasal status and portion presented shows that listeners do not perceive the distinction more accurately as they hear a larger portion of the context. They perceive the distinction just as well from the preceding vowel as they do from the vowel plus consonant. This suggests that information about nasalization of the fricative is not located exclusively within the consonant.
4 General discussion
Comparison across the production and perception results shows a clear picture: speakers of Scottish Gaelic, at least the dialects represented by our participants, have largely neutralized the distinction between ‘nasalized fricatives’ and regular, non-nasalized ones, but they still maintain the distinction in a minority of tokens they produce. This is true regardless of whether the phonological locus or phonetic realization of nasalization is assumed to be in the consonant (MacAulay 1992) or the vowel (e.g., Borgstrøm 1941; Bauer 2011). Native listeners hearing these consonants reflect that pattern: they are able to perceive the distinction with better-than-chance accuracy, but only just barely so. Listeners’ moderate but significant ability to perceive the distinction is a good reflection of the variability in the production data: the distinction is neither made nor perceived consistently, but it is variably present. This matches with anecdotal observations: on Skye, there are two road signs located very near each other pointing the way to the town of Ardvasar. One spells the Gaelic name Ard a Mhasair while the other spells it Àird a’ Bhàsair, suggesting that the orthographic mh/bh distinction is variable.
The overall accuracy rate of 57%, above chance but far from ceiling, is reminiscent of several other types of distinctions in the phonetic or sociolinguistic literature, such as incomplete neutralization (Warner et al. 2004; Kharlamov 2014; and others cited therein) and near-merger (e.g., Labov et al. 1991). The literature is inconclusive about what causes such marginal distinctions even in well-studied languages such as English, German, and Russian. Some possible influences include orthography, presence of minimal pairs in the experiment, reading vs. speaking tasks, past exposure to dialectal variability, and sound change in progress. We are not able to collect enough data for Gaelic to answer these questions. However, since Early Irish, the ancestor of the language, is described as having /ṽ/ 13 (Gillies 2009), and the current production data show the distinction between mh (historical /ṽ/) and bh (/v/) as being marginal, it seems likely that there is a sound change in progress moving toward neutralization. Still, since we do not find a clear age effect in the perception data among our participants, the neutralization does not seem to be reaching completion within current adult speakers.
Apart from the causes of such a marginal distinction, the implications of the findings for aerodynamics are clear: Gaelic speakers do not produce strong frication at a source in the oral cavity above the velic split while also producing strong nasalization. Such a sound should be almost, but perhaps not quite, aerodynamically impossible (Shosted 2006). However, they also do not consistently take any of the approaches to realizing such a segment that might seem the most likely: (1) they do not simply produce strong nasalization on an adjacent vowel instead of on the consonant; (2) they do not maintain nasalization only in conditions where there is no aerodynamic conflict, namely /h/ and approximants; and (3) they do not simply neutralize the nasalization distinction entirely. Instead, they produce the distinction between historical mh and bh variably, neutralizing nasalization in most tokens, but maintaining nasalization on or near the target consonant for a minority of tokens in any of several ways. The tokens that maintain nasalization can involve any of (1) producing nasalization on a neighboring vowel, (2) producing first frication and then nasalization in turn (a biphasic consonant), (3) producing nasalization but no frication noise, or (4) producing both a small amount of nasalization and low amplitude frication noise. This last case is the closest to the original claim of nasalization during a fricative, but such tokens are extremely rare (at most 7 in our entire dataset), and show an inevitable trade-off between nasal and oral airflow, demonstrating that nasalization and oral frication occur at the expense of each other.
We tested two types of consonants that have been described as being nasalized for different reasons in the literature. The more obvious case is the orthographic mh vs. bh distinction, our word-medial and word-final conditions. This is the case where some sources describe an underlyingly nasal fricative, /ṽ/ (MacAulay 1992), and other sources state that this consonant itself is nasalized, although without specifying whether the phonological distinction is marked on the consonant (Ó Maolalaigh 2003). Other sources describe the historical consonant as having caused distinctive vowel nasalization on the adjacent stressed vowel (Borgstrøm 1941; Gillies 2009; Bauer 2011). The type of consonant we test in word-initial position, as in do phuing /dɔ fɯiŋʲgʲ/ ‘your point’ is perhaps only claimed by Ternes (2006) to have nasalization, because Ternes’ account involves distinctive nasalization on stressed vowels spreading both leftward and rightward until it reaches either a stop or the word boundary.
Despite the difference in what aspect of the phonology could cause nasalization in these two types of cases (mh vs. allophonic nasalization), and the difference in how widely nasalization has been claimed for these consonants in the literature, we found considerable similarity in these two situations. In the production data, both underlying mh conditions, the word-medial and the word-final, are somewhat more likely to have nasalization in the consonant than the word-initial bh condition with allophonic nasalization is. However, only the difference between initial and medial, not initial and final, is significant. Thus, it seems that the segment with underlying or historic nasalization as an inherent part of the consonant, the mh case, may be nasalized more often than the corresponding allophonically nasalized fricative, but the difference is not large enough to form a clear statistical pattern. Turning to perception, listeners’ ability to distinguish nasalization was not limited to one type or the other.
Furthermore, in the production data, the segments with the strongest nasalization were those realized as back fricatives (th, sh /h/; ch /x/), which are only nasalized allophonically. These consonants never have nasalization as a historical or underlying property of the consonant itself. In the perception data, two of the pairs with the strongest detectability had /h/ consonants. Furthermore, /h/ targets in the production experiment were the only ones to commonly have nasalization throughout the vowel between the target consonant (/h/ in this case) and the nasal stop after the vowel (such as /m, n, N/) that triggers nasalization. That is, /h/ was the only target consonant to frequently show phonetic spreading of nasalization all the way from the next consonant leftward through the vowel and the target consonant, although not all tokens of /h/ had this, either.
It is perhaps surprising that nasal condition /h/ shows such strong nasalization, considering that it does not have nasalization underlyingly, and only Ternes (2006) has claimed that it would be nasalized. Ternes does not, however, claim greater nasalization for /h/ than for other fricatives. This may fit in with Ó Maolalaigh’s (2003) conclusion that many words of Scottish Gaelic show sporadic nasalization due to rhinoglottophilia, triggered by high oral airflow consonants and/or /h/ rather than by nasals. /h/ is an extremely high airflow consonant. The combination of an /h/ and a nearby nasal consonant such as /m, n, N/, having only an intervening vowel that is nasalized through assimilation to the following consonant anyway, may be sufficient to increase the chances of nasalization throughout a large portion of the word. However, a contributing factor to the greater nasalization of /h/ target consonants is also the fact that there is no aerodynamic problem in nasalizing a fricative with the place of articulation farther back than the velum (Ohala 1975): this is what allows the nasalization on the /h/ to occur.
Word-initial consonants other than mh in this study, such as /f/ in do phuing /dɔ fɯiŋʲgʲ/ ‘your point’, do not have underlying nasalization, and speakers did produce some tokens with nasalization. They did so significantly more often than for matched consonants in oral conditions, so this is not entirely a matter of spontaneous nasalization (as in Ó Maolalaigh 2003); it has to be caused by the nasal stop such as /m, n, N/ later in the word. This might seem like validation of Ternes’ (2006) claim for Applecross Gaelic that distinctive nasalization on vowels spreads leftward and rightward until it hits either a stop consonant or a word boundary, resulting in forms like friamh [f̃r̃ĩãːṽ] ‘root,’ with nasalization throughout the word, including on three consonants. However, our results do not support this analysis. First, the fact that most of the tokens with nasalization on the word-initial consonant do not have nasalization throughout the following vowel rules this interpretation out, at least for our speakers’ dialects. Ternes’ analysis does not predict that nasalization could skip the intervening vowel. This suggests that the leftward spread of nasalization may be an abstract phonological process rather than just an early lowering of the velum, as Ternes seems to suggest. Furthermore, many of the figures given as examples here fail to show the pattern Ternes predicts in categorical transcriptions that must either put a nasalization mark or not on every segment in a predictable pattern. Every example in Figures 1–3 and Figure 7 shows failure of nasalization to spread as far as Ternes’ analysis would predict. Vowel height is also relevant for this issue. Ternes (2006) and Bauer (2011) state that Scottish Gaelic has distinctively nasalized stressed vowels but not for the mid vowels (upper-mid in Ternes’ analysis). Borgstrøm’s (1941) analysis is similar, but leaves some ambiguities. Therefore, even Ternes’ analysis with long-distance nasal spreading should not apply if the stressed vowel is upper-mid. We have very few upper-mid stressed vowels near target consonants in our materials, though.
A thorough analysis of nasalization during the vowels of Gaelic, including investigation of vowel height and of whether nasalization is phonologically marked on the vowel vs. the consonant, is beyond the scope of the current work. Our primary purpose is to address the aerodynamic issue, regardless of whether the source of any nasalization is a distinctively nasal vowel or consonant. However, initial qualitative inspection of the nasal airflow traces given as examples here does not seem to support Ternes’ (2006) or Bauer’s (2011) analyses with the vowel as the locus of distinctive nasalization, and neither do the results fully support MacAulay’s analysis with orthographic mh as a distinctively nasalized fricative /ṽ/. Instead, the nasalization distinction that existed historically for the mh consonant seems to be most often neutralized, and realized in variable ways when it is maintained.
To summarize, the production and perception results together show that the aerodynamically problematic nasalized vs. oral fricative distinction that has been claimed for Gaelic is marginal whether its phonological locus is understood as the consonant or the vowel, but variably present. Speakers neutralize the distinction, producing no nasalization near the target consonant at all, for most tokens. However, they do produce nasalization on a minority of tokens, with one of several options for timing the nasalization relative to the target consonant. In turn, the perception results show variable perception of the distinction, significantly better than chance, but much less accurate than for typical linguistic distinctions such as the palatalization and preaspiration distinctions (Warner et al. 2011). Together, these results show a picture of variable maintenance of a distinction that poses an aerodynamic conflict, with merger of the distinction, rather than an alternative such as producing nasalization on a neighboring segment, surfacing as the majority outcome.
We would like to thank all of the participants in the experiments for sharing their knowledge of their language with us. We also thank Boyd Robertson, John Norman MacDonald, Maoilios Caimbeul, Sabhal Mòr Ostaig, and the Columba 1400 center in Staffin for making this work possible. We are grateful to Micaya Clymer, Colin Gorrie, and Lio Mathieu for assistance in preparing and running the experiments, and to Ian Clayton and Colleen Patton for assistance in preparing parts of this paper. We also thank Anna Bosch, Diana Archangeli, and two anonymous reviewers for their thoughtful suggestions and commentary on this work. This work was funded by grant BCS-0921685 from the U.S. National Science Foundation.
Appendix A. Items for the production experiment
Appendix B. Items for the perception experiment
Bauer, Michael. 2011. Blas Na Gàidhlig: the practical guide to Scottish Gaelic pronunciation. Glasgow: Akerbeltz. Google Scholar
Bauer, Michael & UilleamMacDhonnchaidh. 2014. Am Faclair Beag. http://www.faclair.com/index.aspx?Language=en (accessed September 2014).
Borgstrøm, Carl Hjalmar. 1941. A linguistic survey of the Gaelic dialects of Scotland, vol. II: the dialects of Skye and Ross-shire. Norsk Tidsskrift for Sprogvidenskap. Oslo: Aschehoug. Google Scholar
Gillies, William. 2009. Scottish Gaelic. In The Celtic languages, 2nd edn. (Routledge Language Family Series), Martin J.Ball & NicoleMüller (eds.), 230–304. London: Routledge. Google Scholar
Kharlamov, Viktor. 2014. Incomplete neutralization of the voicing contrast in word-final obstruents in Russian: phonological, lexical, and methodological influences. Journal of Phonetics43. 47–56. Web of ScienceCrossrefGoogle Scholar
Ladefoged, Peter, JennyLadefoged, AliceTurk, KevinHind & St. JohnSkilton. 1998. Phonetic structures of Scottish Gaelic. Journal of the International Phonetic Association28. 1–41. CrossrefWeb of ScienceGoogle Scholar
MacAulay, Donald. 1992. The Scottish Gaelic language. In The Celtic languages, DonaldMacAulay (ed.), 137–248. Cambridge: Cambridge University Press. Google Scholar
Ohala, John J. 1975. Phonetic explanations for nasal sound patterns. In Nasálfest: papers from a symposium on nasals and nasalization, Charles A.Ferguson, Larry M.Hyman & John J.Ohala (eds.), 289–316. Stanford: Language Universals Project. Google Scholar
Ohala, John J. & ManjariOhala. 1993. The phonetics of nasal phonology: theorems and data. In Nasals, nasalization, and the velum (Phonetics and Phonology Series, Vol. 5), Marie K.Huffman & Rena A.Krakow (eds.), 225–49. San Diego, CA: Academic Press. Google Scholar
Ó Maolalaigh, Roibeard. 2003. Processes in nasalization and related issues. Ériu53. 109–32. Google Scholar
Shosted, Ryan K. 2006. The aeroacoustics of nasalized fricatives. Berkeley: University of California dissertation. Google Scholar
Smits, Roel, NatashaWarner, JamesMcQueen & AnneCutler. 2003. Unfolding of phonetic information over time: a database of Dutch diphone perception. Journal of the Acoustical Society of America113. 563–74. CrossrefGoogle Scholar
Solé, Maria-Josep. 1999. The phonetic basis of phonological structure: The role of aerodynamic factors. In Proceedings of the 1st Congress of Experimental Phonetics, 77–94. Tarragona, Spain: Universitat Autònoma de Barcelona. Google Scholar
Ternes, Elmar. 2006 . The phonemic analysis of Scottish Gaelic, based on the dialect of Applecross, Ross-shire, 3rd edn. Dublin: School of Celtic Studies, Dublin Institute for Advanced Studies. Original edition, Hamburg: Helmut Buske Verlag. Google Scholar
Stewart, Thomas W. 2004. Mutation as morphology: bases, stems, and shapes in Scottish Gaelic. Columbus: The Ohio State University dissertation. Google Scholar
Warner, Natasha, AndrewCarnie, MurielFisher, JessamynSchertz, LionelMathieu, ColinGorrie, MichaelHammond & DianaArchangeli. 2011. Timing of perceptual cues in Scots Gaelic. Poster presentation at the Acoustical Society of America meeting, San Diego, CA, November 2011. Google Scholar
Warner, Natasha, AllardJongman, JoanSereno & RachèlKemps. 2004. Incomplete neutralization and other sub-phonemic durational differences in production and perception: evidence from Dutch. Journal of Phonetics32. 251–76. CrossrefGoogle Scholar
Warren, Donald W., Rodger M.Dalston & RobertMayo. 1993. Aerodynamics of nasalization. Phonetics and Phonology5. 119–46. Google Scholar
Zajac, David J. & Campbell C.Yates. 1997. Speech aerodynamics. In Instrumental clinical phonetics, Martin J.Ball & ChristopherCode (eds.), 87–118. London: Whurr Publishers. Google Scholar
We underline the target consonant in orthographic representations of example words throughout. We will not mark nasalization in transcriptions, since the issue we are investigating is exactly that: whether, and where, there is nasalization. We will mark it only when necessary to show what distinction is being discussed, or when reviewing how other authors have described the nasalization of a given word or segment. For transcriptions of whole words, we use a phonemic level transcription given in phonemic slash marks based on the pronunciation of the native speaker co-author (Fisher). We use phonetic level transcription in square brackets only when necessary to discuss the realization of a particular production.
We use /N, L/ for the ‘fortis’ consonants, following the convention for Gaelic, e.g., Gillies (2009).
In the account of Gaelic with distinctive nasalized vowels, the vowel /ɛ̃/ in this word historically became a nasalized vowel because of the following /n/, and synchronically this vowel would be the source of any allophonic nasalization that might affect the preceding fricative.
The process of exporting the data files from MacQuirer to Praat may have obscured the units of the raw airflow measures. However, gain was corrected before export of the files, so the scale is the same for all speakers. For this and other reasons explained below, we focus primarily on the data regarding when in the word elevated airflow occurs, which is not dependent on the raw airflow units. Several additional measures were taken but will not be reported here. For example, we measured peak oral airflow and average nasal and oral airflow for each segment, but we found the average airflow measures too noisy to be useful because the portion of the segment containing elevated airflow varies greatly.
Figure 2(b) and (d) may show such failure of the oral mask to seal fully to the face, allowing leakage around the oral mask. This would make the oral airflow trace unreliable. Given the fieldwork setting of using an oral and nasal mask with older speakers of an endangered language, it is difficult to avoid this entirely. We make minimal use of the oral flow measurements in our analyses, and use the waveform and spectrogram as additional information about frication during target consonants, the crucial point for the oral flow. The nasal mask is not subject to the same problem, since it is held on by velcro straps and it does not have to seal around the chin.
A two-factor ANOVA including orthographic consonant and nasalization condition did show a significant interaction (F(5,65)=9.87, p<0.001), so investigation of the simple effects of nasalization status is justified apart from the concern about unequal variance. Because of the large proportion of tokens without nasalization, we were unable to use a logistic regression approach with this dataset.
If palatalized, ch is [ç], but all of our words in the ch condition are non-palatalized, so [x] is expected. Some tokens sound like they may be realized as glottal [h], with no strong velar constriction. Without tongue position data using articulograph or ultrasound methods, and with the muffled sound quality of recording with an oral mask, it is difficult to be sure whether there is a substantial velar constriction in all cases.
As mentioned above, for some such cases the oral mask may not have been sealed well to the face. Figure 2(d) is probably such a case. However, if there were strong frication in any of these tokens, it should still be visible in the waveform or spectrogram, since the audio signal does not depend on the seal of the oral mask. We did not find tokens with clear frication visible in the waveform or spectrogram but lacking elevated oral airflow through leakage.
This word can also be pronounced as [khã:əL], with nasalization on the vowel as the reflex of the historical mh, and this is the pronunciation given by Faclair Beag. Fisher pronounced it with a [v] in the stimulus recording, without being instructed to do so. For the purpose of the perception experiment, the question is whether any cues to the mh vs. bh distinction are perceived (presumably nasalization), regardless of whether a consonant is synchronically present or has been deleted. This is not a common word of Gaelic, which also contributes to its variability.
We cannot know how /ṽ/ was realized phonetically at older stages of the language: speakers may have produced nasalization on neighboring vowels and not on the fricative even at the time of Early Irish, or they may have produced a nasalized but approximant rather than fricative version of the consonant.