Acoustics and perception of velar softening for unaspirated stops

Abstract

This paper provides articulatory, acoustic and perceptual data in support of the hypothesis that the velar softening process through which /k/ becomes /tʃ/ is based on articulation rather than on acoustic equivalence if operating on unaspirated stops. Production data for unaspirated /k/ are analyzed for five speakers of Majorcan Catalan, where the velar stop phoneme exhibits (alveolo)palatal or velar allophones depending on vowel context and position. Data on several parameters, i.e., contact anteriority and dorsopalatal contact degree, burst spectral peak frequency, energy and duration, and F2 and F3 vowel transition endpoints and ranges, suggest that /tʃ/ may have originated from (alveolo)palatal stop realizations not only before front vocalic segments but also before low and central vowels and word finally. Perception results are consistent with the production data in indicating that the most significant /tʃ/ perception cues are burst energy before /a, u/ and burst duration word/utterance finally. They also suggest that velar softening for unaspirated /k/ before front vocalic segments is triggered by an increase in burst frication energy and duration resulting from the narrowing of an (alveolo)palatal central channel occurring at stop closure release. These findings are in agreement with the existence of contextual and positional (alveolo)palatal stop allophones of /k/, and with evidence from sound change, in the Romance languages.

Introduction

Velar softening involves the replacement of /k/ by the alveolopalatal affricate /tʃ/ and even by the alveolar affricate /ts/ which may then give rise to the fricatives /ʃ/ and /s/, respectively. The affricate /tʃ/ is generally produced with the tongue blade at the central or back alveolar zone and possibly the tongue predorsum at the prepalate, and will be referred to as alveolopalatal rather than as palatoalveolar throughout this paper in view of data for the Catalan language reported in the literature (see Recasens & Espinosa, 2006). Illustrative examples of this velar softening process may be found in the Romance languages, e.g., Latin /ˈkεnto/ CENTU ‘one hundred’ has evolved into [ˈtʃento] cento in Italian and into [s

] cent in French.

The change /k/> /tʃ/ is widely attested before front vowels and /j/ as in the case of the first Slavic palatalization and in Bantu, just before front vowels as in the case of the second Slavic palatalization, in Indo-Iranian and in Old English word initially, or just before /i/ as in Cowlitz Salish or before /j/ as in Chinese (Guion, 1996). As for the Romance language family, a first palatalization wave caused /k/ to become /tʃ/ (also /ts/) before front vowels and /j/ in all languages except for Sardinian which suggests that the process should be traced back to late Latin and must have been completed by the 5th century A.D.

The reason why velar softening is implemented preferably before front vocalic segments is consistent with the well-known fact that velar stops are articulated at a more anterior location in this contextual condition than before low and back rounded vowels and /w/. Indeed, closure location occurs generally at the rear hard palate and therefore, at the postpalatal or postpalato-velar regions for front velars, and at the soft palate or velar region for back velars. Palatographic and X-ray data indicate that this contextual variation in place of articulation for velars is widespread among the world's languages, i.e., it occurs in languages as diverse as English (Keating & Lahiri, 1993; Liker & Gibbon, 2007), Czech, Hungarian (Keating & Lahiri, 1993), Telugu (Reddy, 1981) and French (Corneau, Soquet, & Demolin, 2000), as well as in the Eastern Catalan dialect (Recasens & Pallarès, 2001).

Two different explanatory hypotheses based on articulation (Section 1.1.1) and on acoustic equivalence (Section 1.1.2) have been put forth in order to account for velar softening.

The articulation-based hypothesis has been the prevailing explanation of velar softening during the 20th century and even earlier. Among other scholars, Romance phoneticians and dialectologists in the past have argued that velar softening takes place after an increase in closure fronting for front velar stops causing a change in place of articulation from postpalatal or postpalato-velar to mediopalatal, prepalatal, alveolopalatal or even alveolar. After this change in place, the stop burst, which is composed of a burst spike, a frication phase and an aspiration period if available, is integrated as the frication component of an affricate (Anttila, 1972, p. 73; Grammont, 1971, p. 214; Hock, 1986, p. 76; Rousselot, 1924–1925, pp. 607–631). In summary, the articulation-based hypothesis claims that velar softening (a change in manner, from stop to affricate) is necessarily preceded by velar palatalization (a change in place of articulation, from velar to a more anterior articulation approaching the closure and constriction location for /tʃ/). In this paper, the terms ‘velar softening’ and ‘velar palatalization’ will be kept separate. The stop realizations generated through velar palatalization will be referred to as (alveolo)palatal, and transcribed as [c] if voiceless and as [ɟ] if voiced, in line with the fact that they show most typically two articulatory patterns: alveolopalatal involving a simultaneous alveolar and palatal closure; and palatal proper with closure occurring exclusively at the palatal zone (Recasens, 1990). The main articulator is the tongue dorsum if the articulatory outcome is palatal, and the blade and predorsum if it is alveolopalatal.

In support of the articulation-based hypothesis, there is a good deal of palatographic data showing that (alveolo)palatal stop realizations derived from /k/ differing in closure fronting degree may occur before front vocalic segments in languages and dialects as diverse as Greek (Nicolaidis, 2001, pp. 71–73), Romanian (Dukelski, 1960, p. 44), Majorcan Catalan (Recasens & Espinosa, 2006) and Occitan (Maurand, 1974, pp. 116, 131–132). These realizations range auditorily from more /k/-like to more /t/-like, which accounts for why phoneticians have often used slightly different phonetic symbols in order to transcribe those sound qualities (Rousselot, 1924–1925). Accordingly, replacements such as [keˈki] celui-ci ‘this one’ by [keˈti] in Occitan could very well arise from the existing form [keˈci] in the same dialect (Bouvier, 1976, p. 83).

Moreover, whenever enough frication is generated through the lingual constriction at stop release, the affricate outcomes of [c] through velar softening may exhibit different degrees of closure fronting which parallel those for the (alveolo)palatal stop (Scripture, 1902). Thus, the resulting affricate may vary from palatal ([tç]) to alveolopalatal ([tʃ]) to alveolar ([ts]) depending on whether closure for [c] occurs at the palatal, alveolopalatal or alveolar zone, respectively. Supporting evidence derives from the presence of these affricate types in the same or close dialectal areas, often cooccurring with realizations of [c]: [c], [tç], [tʃ] and [ts] in Rhaetoromance (see Section 1.2.2); [c] and [tʃ] in Francoprovençal ([ky], [c(j)y], [tʃy] cul CULU ‘bottom’; Gardette, 1967–1984, map 1112); [c] in Standard Modern Greek but [tç] in Cretan Greek and [tʃ] in Cypriot Greek (Trudgill, 2003, p. 54).

A more recent proposal, the acoustic equivalence hypothesis, claims that the categorization of front velars as affricates has not taken place through additional closure fronting. If velar softening is achieved through assimilation via an (alveolo)palatal stop, it remains unclear why such a stop should be articulated so much further forward than the vocalic segment causing the assimilatory process to take place (Ohala, 1992). Ohala and colleagues argue instead that velar softening is caused by acoustic equivalence between the front velar stop burst and the /tʃ/ frication noise which, as found for several languages, share a spectral peak frequency at about 2500–3500 Hz (English: Guion, 1998; Stevens, 1998, p. 416; Zue, 1980; German, Polish, Catalan: Zygis, Recasens, & Espinosa, 2008). Acoustic similarity is enhanced by the high-frequency F2 vowel transition endpoints for both front /k/ and /tʃ/.

Evidence for this hypothesis is said to derive from the confusion of /ki/ with /ti/. This confusion has been reported to occur for English excerpts with either aspirated /k/ excised from utterance initial CV sequences (Winitz, Scheib, & Reeds, 1972) or else unaspirated /k/ excised from /ski/ after degradation of its 3000 Hz burst peak (Chang, Plauché, & Ohala, 2001). It should be noticed, however, that these perceptual experiments do not test velar softening proper since they do not show that /k/ may be replaced by /tʃ/. Indeed, there are at least two relevant differences between /t/ and /tʃ/ which suggests that an alternative explanation may be needed: the absence of sufficient frication noise at /t/ release, mostly if the stop is unaspirated; a substantial frequency difference between the spectral peak for the /t/ burst (at about 4000 Hz; Zue, 1980) and for the /tʃ/ frication component (at 2500–3500 Hz). Moreover, filtering important spectral information may render speech stimuli unnatural.

In a pioneering production and perception study, Guion (1998) has shown that front /k/ may be confused with /tʃ/ under specific methodological conditions, thus providing some evidence in support of the acoustic equivalence hypothesis. Presentation of natural speech perception stimuli containing a velar stop and the following vowels /i/, /ɑ/ or /u/ yielded /tʃ/ responses only if white noise was superimposed on them. Natural speech stimuli without noise overimposed did not yield any /tʃ/ responses. In these circumstances, /k/ before /i/ was confused mostly with /tʃ/ (35%), but also with /ɡ/ (10%) and with /dʒ/ (12%) perhaps since the low signal-to-noise ratio used in the study (+2B) rendered these stimuli hard to categorize. As expected, /k/ before /ɑ/ yielded relatively few /tʃ/ responses (13%) and practically no confusions with /ɡ/ or /dʒ/; unexpectedly high confusion percentages for /ku/ were attributed by the author to the front realization of /u/ in the Texan American English dialect of the speakers who provided the production data for the perceptual tests. A remarkable finding of Guion's study is that not only the high-frequency burst spectral peak but also VOT duration and, therefore, the duration of the burst frication and aspiration period, plays a relevant role in the implementation of velar softening (i.e., the longer the VOT period, the higher the percentage of affricate reponses).

Descriptive, experimental and sound change data reported above appear to be in support of an articulatory motivation of velar softening at least for Romance and other language families. The main argument is the presence of (alveolo)palatal realizations of front /k/ and the fact that these realizations may cooccur synchronically with affricates in ways which are consistent with differences in contact size and anteriority for [c]. Other arguments in favor of the articulation-based hypothesis will be adduced next.

A problem with an interpretation of velar softening based on acoustic equivalence is that this process appears to have affected not only aspirated voiceless stops with long and salient releases composed of frication and aspiration (e.g., in the Germanic languages), but also shorter and less perceptible unaspirated voiceless stop releases (e.g., in Romance). Analysis data show that word initial VOT values are clearly shorter for unaspirated /k/ (e.g., 29 ms in Puerto Rican Spanish; Lisker & Abramson, 1964) than for aspirated /k/ (e.g., 74 ms in English; Zue, 1980). This large VOT duration difference appears to be associated primarily with aspiration though in practice it is hard to separate aspiration from local frication since the two events overlap with each other (Hanson & Stevens, 2003). The frication noise is typically longer for affricates than for stops (Ladefoged & Maddieson, 1996, p. 90), much more so if stops are unaspirated. Data for /tʃ/ in Spanish and Italian indicate indeed that the fricative portion of the affricate is about 80 ms in slow speech and 65 ms in fast speech, which is much longer than the VOT values for unaspirated stops (Maddieson, 1980).

Differences in burst acoustic prominence between unaspirated and aspirated stops should render the perceptual confusion between front /k/ and /tʃ/ more unlikely for the former class of stop consonants than for the latter. Informal listening of CV excerpts taken from Catalan meaningful sentences reveal, indeed, that unaspirated front velar stop bursts are not confusable with /tʃ/. This observation is consistent with Guion's finding that aspirated front velar stops are harder to be confused with /tʃ/ when burst frication is shortened (Guion, 1998), and suggests that a noticeable increase in burst frication duration and/or intensity is needed for an unaspirated front velar stop to yield an affricate percept.

The little acoustic salience of the burst for voiced unaspirated stops renders the acoustic equivalence explanation even less feasible in case that velar softening is associated with front /ɡ/, e.g., /ɡ/ > /dʒ/ as in Catalan [dʒεl] gel from Latin /ˈɡεlo/ GELU ‘ice’.

Velar softening may also take place in contexts and positions where /k, ɡ/ are usually back rather than front. In Gallo-Romance, velar softening was implemented through a second palatalization wave dated about the 5–7th centuries A.D. and even later, and operating not only before front vocalic segments but also before /a/ and also /ə/ as well as in word final position (French [ʃεvʀ] chèvre CAPRA ‘goat’, [ʃaʀb

] charbon CARBONE ‘coal’, Rhaetoromance [satʃ], [satç], [saç] SACCU ‘sack’; Goebl, 1998, map 678; Pope, 1934, p. 128). This change appears to be at odds with the acoustic equivalence hypothesis since the spectral burst for back /k/ exhibits a much lower frequency than the /tʃ/ frication noise, i.e., it is located about 1600–1800 Hz before /a/ and about 750–1100 Hz before back rounded vowels (Zygis et al., 2008; Zue, 1980).

In support of the articulation-based hypothesis, the (alveolo)palatal stop realization [c] of unaspirated /k/ may occur in those same conditions in languages and dialects of the Gallo-Romance family. This is the case for the Rhaetoromance dialects Romansh, Ladin and Friulian ([can] CANE ‘dog’, [caw] CAPUT ‘head’; Brunner, 1963; Haiman & Benincà, 1992, pp. 66–70; Lutta, 1923, p. 40) and for Northern Italian dialects ([car] CARU ‘dear’, [sac] SACCU ‘sack’; Rohlfs, 1966; Salvioni, 1901). Moreover, (alveolo)palatal stops and affricates may cooccur before /a/ and word finally in the dialectal areas referred to, e.g., Rhaetoromance [ˈcawɾa], [ˈtʃawɾa] CAPRA ‘goat’, [rεc], [rεtʃ], [rəts] REGE ‘king’ (Jaberg & Jud, 1928–1935, map 1079; Luzi, 1904, p. 769). Velar palatalization has taken place in the same circumstances in the Majorcan Catalan dialect under analysis in the present paper (Recasens, 1996, pp. 243–244), as well as in other language families, e.g., in Alsean in the Northwest coast of America where [c] occurs before /a/ and /ə/ and word finally (Buckley, unpublished (a)), and in Persian where word final /k/ is realized [c] ([xac] ‘earth’; Bhat, 1974, p. 46; Buckley, unpublished (b)).

The acoustic equivalence hypothesis would be compatible with an especially anterior, [ε]-like realization of /a/. It has been argued that this realization must have existed in Gallo-Romance at the time velar softening applied, based mostly on the fact that /a/ has yielded a mid front vowel in open syllables in languages such as French (e.g., mer MARE, the example chèvre above; Buckley, unpublished (b)). It is by no means clear, however, whether /a/ raising preceded or followed the generation of the outcomes [c] or [tʃ] of /k/. Progressive palatalization of /a/ by a preceding (alveolo)palatal consonant is a common phenomenon (Valencian Catalan [ʎεrk] llarg LARGU ‘long’; Recasens, 1996, p. 101), and may operate in parallel to progressive /j/ insertion as exemplified by Old French chier derived from Latin CARU ‘dear’ (Lausberg, 1970, pp. 261–262; Pope, 1934, pp. 127–128, 163). Moreover, the vowel /a/ has stayed unchanged in several Romance domains where /ka/ has evolved into [ca], [tʃa] or [tsa] (i.e., in those mentioned above as well as in Northern Occitan and Francoprovençal), and velar palatalization before an unfronted variant of /a/ occurs in present-day Northern French (see Coustenoble, 1945, p. 77 and palatographic material in Rousselot, 1924–1925, p. 607). It is also the case that velar softening in French operated before the diphthong /aw/ where an [ε]-like realization of /a/ is problematic (e.g., chose CAUSA ‘thing’, where o was issued from /aw/ through monophthongization). Another point to be made is that the spectral peak for /k/ before /ε/ falls between 2000 and 2500 Hz and thus, below that for /tʃ/ (Zygis et al., 2008; Zue, 1980), and that velar softening applies less frequently before /ε/ than before the higher cognates /i, e/ in the world's languages (Bhat, 1974, p. 30).

As shown in the preceding sections, evidence in support of the articulation-based hypothesis derives from the observation that the (alveolo)palatal stop and the outcoming affricate may share an analogous closure placement in velar softening processes. Voicing, prosodic and lexical characteristics may also play a role. Thus, the affricate or fricative outcomes may be more anterior for original /k/ than for original /ɡ/ (French [s

] cent CENTU ‘one hundred’, [ʒ

] gens GENTE ‘people’), as well as word initially than intervocalically (Ladin [ˈtsento] CENTU ‘one hundred’, [veˈtʃin] VICINU ‘neighbour’; Gartner, 1910, p. 189) and in stressed than unstressed syllables (Friulian [ˈtʃazε] CASA ‘house’, [kaˈmiŋ] CAMINU ‘path’; Jaberg & Jud, 1928–1935, map 269). All these cases may be accounted for assuming that tongue contact size and closure fronting, as well as duration, are greater for voiceless stops than for voiced stops and increase with stress and at the edges of the lexical and prosodic domains (Cho, 2001; Farnetani, 1990; Fougeron, 2001; Fougeron & Keating, 1997; Keating, Wright, & Zhang, 1999).

Another finding more in support of the articulation-based hypothesis than of the acoustic equivalence hypothesis appears to be the alveolar affricate outcome /ts/ of velar softening. Assuming that the alveolar affricate has been issued directly from a stop, it is hard to envisage how a front velar stop burst may be confused perceptually with the sharp /s/ frication noise exhibiting a spectral peak well above the typical 2500–3500 Hz frequency for front /k/. In principle, the articulation-based hypothesis could handle better this sound replacement, as revealed by articulatory data showing that very front (alveolo)palatal stops may be articulated at the alveolar zone and, therefore, at the same place of articulation as /ts/ (Section 1.1.1).

In the light of the data described above, velar palatalization should not be treated simply as an assimilation through which /k/ acquires the palatal place of articulation of a following front vocalic segment. We would like to suggest instead that the change /k/>[c] may be achieved through two production mechanisms, i.e., gestural blending and gestural strengthening.

Front velar palatalization in the adjacency of a dorsopalatal vowel or glide is achieved through gestural blending since both the consonant and the vocalic segment are produced with the same tongue dorsum articulator in this case. Blending may be implemented through two different production strategies. It may yield an intermediate articulation between the stop and the front vowel or glide, as for postpalatal realizations of /k/ before a front vocalic segment (Browman & Goldstein (1989), Browman & Goldstein (1992)). Another option is for blending to give rise to an articulation encompassing or exceeding the combined closure or constriction areas of the two consecutive phonetic segments, as for several Catalan clusters made up of /t, n/ and the alveolopalatals /ʎ, ɲ/ (Recasens, 2006). The latter mechanism could account for why front velar stops may exhibit so much closure fronting when implemented as (alveolo)palatal stops.

Specific conditions upon which velar palatalization occurs suggest that gestural strengthening may also act as a velar palatalization trigger (Straka, 1965). This suggestion is valid for the word initial and word final positions and for stressed syllables which favor lengthening and an increase in tongue contact in the consonant, as well as before open vowels where (alveolo)palatal closure formation requires much tongue dorsum raising. In view of the descriptive materials adduced in Section 1.2.2, it looks feasible to claim that velar palatalization through strengthening may apply across the board except word internally before /ɔ, o, u/ perhaps since keeping a very posterior closure location and the lips relatively rounded and protuded in anticipation to a back rounded vowel renders tongue dorsum raising towards the palate especially hard.

Diachronic evidence in support of these two velar palatalization mechanisms comes from Romance where velar softening occurred twice in two different sets of conditions: only before front vocalic segments in Late Latin and in practically all early Romance languages presumably through velar palatalization induced by gestural blending (see Section 1.1); before other vowels and word finally in a subset of Romance languages and dialects a few centuries later, presumably through velar palatalization induced by gestural strengthening (see Section 1.2.2).

The goal of the present study is to test the articulation-based hypothesis by looking for the acoustic cues which could contribute to the generation of affricates from (alveolo)palatal stops, and by speculating on the perceptual effectiveness of those cues. Articulatory data will be adduced in order to substantiate the claim that velar palatalization and velar softening may apply not only before front vocalic segments but also in other conditions allowing an increase in linguopalatal contact fronting to occur.

Acoustic analyses and perceptual tests will be carried out on the same electropalatographic (EPG) data for Majorcan Catalan used in Recasens and Espinosa (2006), i.e., on productions of unaspirated /k/ next to /i, a, u/ in word/utterance initial, intervocalic and word/utterance final position. The main reason for selecting Majorcan Catalan is because this dialect has the (alveolo)palatal stop [c] in its sound inventory, which was probably issued as an internal development and has been reported to sound [tʃ]-like sporadically (Bibiloni, 1983). Descriptive and experimental studies indicate indeed that, in Majorcan, [c] is an allophone of /k/ appearing most typically before a front vowel but also before /a/ and /ə/ and word finally after any vowel but not word internally before a back rounded vowel ([ci] qui ‘who’, [ˈcazə] casa ‘house’, [sac] sac ‘sack’, but [kus] cus ‘he/she sows’; Recasens & Espinosa, 2006; Veny, 1983). This scenario differs from that for the other Catalan dialects such as Eastern Catalan where /k/ may be slightly more anterior or more posterior depending on the following vowel (see Section 1.1).

The chances that (alveolo)palatal [c] may be categorized as /tʃ/ should depend on the following predictions about the spectral peak frequency, intensity and duration of the stop burst, and the endpoint and range of the vowel formant transitions.

As for the burst spectral frequency, [c] should show a 3000–4000 Hz front-cavity-dependent spectral peak appropriate for or similar to that for /tʃ/ (Keating & Lahiri, 1993). Vowel environment is expected not to affect these spectral characteristics much since (alveolo)palatals should coarticulate little with the contextual vowels (Recasens, 1984). Given that stop bursts and affricate noises differ in intensity and duration, these two parameters ought to be less favorable to velar softening than spectral frequency: the intensity of the frication noise increases with airflow volume velocity and the cross-sectional area of the constriction (Stevens, 1998) and, therefore, is higher for affricates than for unaspirated stops; moreover, affricate frication noises are longer than unaspirated stop bursts (see Section 1.2.1). In specific circumstances, however, the intensity level and duration of the [c] burst may approach those for /tʃ/. Indeed, the stop burst intensity is also determined by front-cavity size at stop release and, thus, should be especially high before low and back vowels which are articulated with a larger front cavity than front vowels (Dorman, Studdert-Kennedy, & Raphael, 1977). This observation is consistent with data for the Australian languages showing that (alveolo)palatal stops combine better with back vowels than with front vowels, i.e., /#ca/ and /#cu/ are preferred to /#ci/ (Butcher, 2004). As for duration, VOT data reported in the literature (Docherty, 1992, pp. 26–27) indicate that bursts may be longer before high vs. low vowels. Burst duration may also be an effective velar softening cue word finally where /k/ may give rise to affricate and fricative realizations (see evidence from Rhaetoromance in Section 1.2.2). Since Catalan and other Romance languages do not have a spirantization rule affecting voiceless stops, the replacement of [c] by [ç] in word final position is prone to be caused by prepausal lengthening of the stop burst frication period rather than by shortening or elimination of the stop closing phase.

Regarding the vowel transitions, the acoustic theory of speech production predicts that fronting closure location from the postpalate and mediopalate towards the alveolopalatal zone should cause the endpoint frequency of the F2 vowel transitions to lower slightly or stay the same, and the endpoint frequency of the F3 vowel transitions to rise (Fant, 1960; Ladefoged & Bladon, 1982). The rationale for these formant frequency variations is to be sought in the fact that F2 for (alveolo)palatals is a half-wavelength resonance of the back cavity and is positively associated with the tongue pass rather than with the lip section, while F3 is inversely related to front-cavity length (Fant, 1960; Stevens, 1998). Unpublished data on Eastern Catalan /tʃ/ (referred to partly in Recasens, 1986) reveal that, for the vowel formant transitions for [c] to approach those for the affricate consonant, the F2 transition endpoints should occur at about 2000–2100 Hz (in the context of /i, e, ε/), 1800 Hz (/a, ə/), 1700 Hz (/ɔ, o/) and 1550 Hz (/u/), while the F3 transition endpoints should take place at about 2700–2900 Hz (/i, e, ε, a, ə/) and 2400–2600 Hz (/ɔ, o, u/). These formant frequency values are highly similar to those for other (alveolo)palatal consonants though slightly lower in the case of F2 and slightly higher in the case of F3 (Recasens, 1984).

The frequency extent of the formant transitions (i.e., the formant frequency range) varies with vowel context in a similar way for all consonants produced at the palatal and alveolopalatal zones. In view of the formant transition endpoints just mentioned, the F2 frequency range should be greater for low and back vowels and ought to increase as front vowels become lower. Vowel intensity, which increases with oral opening and is thus greater for /a/ than for /i, u/, ought to contribute to the prominence of the vowel transitions as well (English: Lehiste and Peterson, 1959; Eastern Catalan: Recasens, 1986). A possible road to velar softening may be the integration of long-range and intense formant transitions in CV sequences with [c] as the glide [j], followed by the perception of [j] as /ʃ/ whenever an increase in the degree of lingual constriction at stop release causes much frication to occur. Present-day phonetic variants and written forms in documents from the past suggest that the change /k/ > /tʃ/ before /a/ may have developed indeed through a transitional glide, which was inserted in Old French (see Section 1.2.2) and may still be heard in Francoprovençal ([tʃjε(r)] cher CARU ‘dear’; Gauchat, Jeanjaquet, & Tappolet, 1925, p. 60) and Wallon and Picard ([k(j)εr], [cjεr], [tʃ(j)εr] cher CARU ‘dear’, [tje], [cjε], [tʃ(j)e˜] chien CANE ‘dog’, [dʒjãb] jambe CAMBA ‘leg’; Bruneau, 1913, p. 410; Carton, 1972, p. 454; Haust, 1953–1997, pp. 106, 109).

In summary, the identification of (alveolo)palatal [c] with /tʃ/ and thus, the articulation-based hypothesis of velar softening, is expected to work out mostly for the frequency component of the burst and the formant transition endpoints largely irrespective of vowel context and position. Burst intensity and formant transition ranges are expected to favor velar softening mostly before low and back vowels, and burst duration could give rise to a /tʃ/ percept most easily before high vowels and to an affricate or a fricative word finally. The effectiveness of all these parameters is subjected to the presence of a prominent burst frication noise which is associated with a narrow constriction and a high airflow volume at stop release.

Section 2 of the present article tests these predictions on the best candidates for velar softening through an analysis of the articulatory and acoustic characteristics of the (alveolo)palatal and velar allophones of /k/ in Majorcan. Section 3 reports results from a series of perception experiments where (alveolo)palatal and velar realizations of /k/ differing articulatorily and acoustically were presented to Catalan listeners for identification.