Place assimilation has been widely attested in many languages. In a typological study on the targeting of place assimilation (Jun, 1995, 2004), some languages demonstrate three places of articulation (coronal, labial, and velar) as a target (e.g., Diola Fogny, Malay, Thai, Nchufie, and Yoruba), and others only allow coronal (e.g., Catalan, English, German, Toba Batak, and Yakut). With respect to its occurrences, coronal is known to be the most frequently selected target. However, targeting exclusively limited to coronal and labial is typologically very infrequent (Jun, 1995, 2004; see also Browman & Goldstein, 1995; Byrd, 1992; Chen, 2003). Place assimilation has been relatively well examined using various theoretical frameworks. At the outset of generative phonology, feature change was responsible for the phonological process (e.g., [+labial]→[+velar] / __ [+velar] in Kim-Renaud (1974)). However, this traditional approach does not provide an explicit mechanism to explain how a speaker probes into and singles out (a set of) specific feature(s) under consideration in his/her phonological system. Later, in feature geometry theory (Clements, 1985), feature specifications were defined in a hierarchical way; for that reason, the target [+labial] and the trigger [+velar] were both governed by the Place node; consequently, an adjacent set of the target and trigger is close enough so that the former is directly influenced by the latter.
V-to-C formant transitions are important perceptual cues to identify place of articulation (Borden & Harris, 1984). With respect to anticipatory place assimilation (e.g., C1C2→C2C2), it can be motivated by weak perceptibility inherent in coda (Byrd, 1992; Fujimura et al., 1978; Jun, 2004; Krakow, 1989; Steriade 2000, 2001). Acoustic formant transition is the result of coarticulation called gestural overlap (Browman & Goldstein, 1986, 1989, 1990a, 1990b, 1990c, 1992, 1995, among others). With increased gestural overlap between C1 and C2, the formant transitional information of C1 can be gradually replaced by that of C2, all things being equal, which is in turn hypothesized to affect speech perception (Browman & Goldstein, 1990c; Byrd, 1992). Another articulatory factor contributing to place assimilation is gestural reduction in C1, either categorical, gradient, or both (Jun, 2004; Kochetov & Pouplier, 2008; Son, 2008; Son et al., 2007). In Chen’s (2003) perceptual recovery algorithm study using the outputs of gestural simulation, she found that complete recovery of voiced coronal stop /d/ in the coda of /d#b/ sequences was less likely to take place with increased (or more) overlap, while recovery of underlying voiced labial stop /b/ in the reversed order (/b#d/) was consistently stronger in the performance of the perceptual recovery algorithm. This means that increased gestural overlap in the assimilating /d(#)b/ sequences is responsible for a weakened perceptual bias to [b(#)b]. Based on this, increased (or more) gestural overlap is considered to be responsible for a reduced perceptual bias, which could have subsequently induced speakers’ miscopying of /d/ as articulatorily diminished.
Revisiting gestural reduction of C1, Browman & Goldstein (ms.) are somewhat reserved in deciding on what could have caused gestural reduction of C1; rather than giving full credit to perceived reduction of C1 with increased overlap, they are also open to the possibility of ‘some independent reason’ (p. 16). Meanwhile, the reversed order of labial-coronal (e.g., /p#t/→[pt], not *[tt], ‘pumptires’ (Byrd, 1992: 19)) avoids place assimilation. Using the output of synthesized speech as the input of a perceptual experiment with human listeners, coronal /d/ in C1, not labial coda, was perceived as assimilated when C1 and C2 were quite overlapped. Byrd (1992) argued that this might be because the lips gesture in C1 moves slower than the tongue tip gesture; consequently, labial in coda is unlikely to be obscured by the subsequent agile articulation of the tongue tip, and thus less susceptible to an ‘acoustically hidden’ segment (Browman & Goldstein, 1990a: 304). To quote Browman & Goldstein (1990b: 422), “Gestures, however, have quantitative (gradient) articulatory properties...” According to them, gesture-based input specifications in C1C2 sequences are considered to provide explanatory accounts for place assimilation as attested in American English (e.g., the string of ‘perfectmemory’ available in the University of Wisconsin X-ray microbeam data). They observed that a small tongue tip gesture still remained after most of the tongue tip constriction degree had lessened, with increased gestural overlap even in fluent and casual speech.
Korean assimilating sequences in the low-vowel context (/a/-to-/a/) demonstrate inter-speaker variability in an EMMA (Perkell et al., 1992) study (Kochetov & Pouplier, 2008). The tongue tip gesture was either categorically deleted, partially (or gradiently) reduced, or unreduced (e.g., /t(#)p/ and /t(#)k/). In particular, partially reduced tokens occurred with inter-speaker variability (e.g., one out of the three speakers), while categorical reduction was observed across speakers. Only one subject from Kochetov & Pouplier (2008) showed categorical as well as partial (or gradient) reduction in the production of coronal /t/: partial (or gradient) reduction was quite limited in occurrence if there was any.
With respect to the nature of gestural reduction in a labial target within /p(#)k/ sequences using EMMA (Perkell et al., 1992), Korean place assimilation has been characterized by categorical reduction (Son et al., 2007). However, subsequent articulatory studies have shown evidence for rare occurrences of gradient (or partial) reduction in terms of spatiotemporal measurements of the lip aperture gesture such as constriction minima combined with constriction duration (Son, 2008). Gradiently reduced coda /p/ in the assimilating /ap(#)ka/ sequences was more frequent in occurrence (3.8% in the coda of production, being limited to the across-word boundary condition), although there was inter-speaker variability (one out of five speakers), compared to gradiently reduced onset /k/ in C2 (1.3% in the onset of production, including both the within-/across-word boundary conditions). Son (2013) showed gestural reduction in terms of kinematic measurements such as closing acceleration duration, overall closing duration, and constriction duration in the lip aperture gesture (/apka/ (assimilation)</apta/ (non-assimilation)). Jun (1996) attributed perceptual place assimilation to the incomplete lips closing gesture (i.e., partial reduction in Jun (1996)) for /p/ in the environment of __/k/) based on his aerodynamic study, which can, in principle, fall in anywhere along a continuum with categorical reduction at one extreme and complete closure at the other. In that case, the emergence of categorically reduced /p/ tokens, if not all, could be interpreted as an extreme manifestation of gradient reduction along a continuum. His oral pressure data from fourteen speakers demonstrated reduction of C1 for /VpkV/ sequences in 47% of production in various vocalic contexts (e.g., /ipki/, /upku/, /ipku/, /upki/).
With regard to perceived assimilation, previous studies agree that labial is less susceptible to place assimilation as compared to coronal. This is because more sluggish movement is less likely to be obscured by a following segment with an agile articulatory movement (Browman & Goldstein, 1990a, 1990c). In the tongue musculature of primates, type I fibers are hypothesized to be related to slower movements of the posterior of the tongue body and type IIA fibers to rapid movements of the apex of the tongue (DePaul & Abbs, 1996). In this sense, the movements of the lips can be gradually covered by the sluggish movements of the tongue back in C2. However, in a perceptual study using various types of /pk/ sequences acquired through simultaneous air pressure-acoustic methodology (e.g., non-overlapped, highly overlapped, reduced C1, and control), highly overlapped /pk/ sequences with fully produced C1 were still perceived as unassimilated (Jun, 1996). Likewise, perceptual consequences have also not been manifested through manipulating degrees of gestural overlap: either: human listeners’ perception was not finely sensitive to C1 reduction (Son et al., 2007).
In the current study, we focus on gestural reduction of C1. Partly due to different experimental or analytical methodologies employed for kinematic studies on assimilating contexts and their controls, there has not been a single kinematic study which has provided, to the best of our knowledge, reduction frequency (either categorical, gradient, or both) along with kinematic characteristics during lip aperture closing movement. In the current study, we are concerned with systematically describing some articulatory reflexes in the assimilating context /pk/ sequences, and we mainly classify lip aperture data from a larger group of subjects.
For the purpose of our current study, we take into account two vocalic contexts (high front vowel /i/-to-/i/ vs. low central vowel /a/-to-/a/), two speech rates (e.g., normal vs. fast), and assimilating /pk/ sequences along with two homorganic controls (/pp/ and /kk/). Note that American English is relatively well studied in terms of numerous vocalic contexts (cf., an aerodynamic study on Korean place-assimilating sequences using various vocalic contexts in Jun (1995)). Various vowels (e.g., non-high back vowels as in ‘pop’, ‘tot’, and ‘caulk’) were used in Browman & Goldstein (1995) and a high front lax vowel (or a schwa) (e.g., ‘perfect’) in Browman & Goldstein (1990c). Using gestural simulations, a low front vowel (e.g., ‘bad’) was used in Byrd (1992) and Chen (2003). In Öhman’s (1967) X-ray data, different articulators involved distinct vocalic and consonantal tiers: the tongue tip articulation for voiced stop /d/ fairly consistently achieved its constriction degree regardless of vocalic context (/i/-to-/i/, /a/-to-/a/, /u/-to-/u/) are, and the articulations of the tongue body and the lips for flanking vowels were vowel-dependent. By hypothesis, gestural tiers are divided into the consonantal tier and the vocalic tier (Browman & Goldstein, 1992; Öhman, 1967), where a consonant is superimposed over the vocalic tier. However, invariant target achievement has not been kinematically examined in an assimilating context, where articulatory reduction can potentially be applied. It is of interest to learn whether this invariance which is independent of vocalic context (high vowel /i/-to-/i/ and low vowel /a/-to-/a/) holds true for target achievement in the assimilating /pk/ sequences.
We are also interested in describing vertical jaw movement. The jaw articulator is shared by both vocalic and consonantal tiers (e.g., jaw height for front vowels gradually decreasing in the order /i/>/ɪ/>/ɛ/>/æ/ and for back vowels /u/>/ʊ/>/ɑ/ in Ladefoged (2001); jaw height for consonants in the order coronal> (labial, velar) in Keating et al. (1994)). As we examine vocalic context effects (/i/-to-/i/ vs. /a/-to-/a/) on jaw height, we also examine whether jaw height of C1 varies as a function of C2 (e.g., assimilating heterorganic /pk/ sequences vs. control homorganic /pp/ sequences) and speech rates (normal vs. fast). Previous kinematic studies on flapping in a high vowel context ([iɾi]) (Son, 2015b) and a low-vowel context ([aɾa]) (Son, 2015a) exhibited no speech rate effects on jaw height. Although previous studies have reported inter-/within-speaker variability with respect to articulatory reduction of C1 in assimilating /pk/ sequences in terms of primary articulator (e.g., either categorical, gradient, or both) (Jun, 1995; Son, 2008; Son et al., 2007), they are lacking in jaw movement data. In the current study, we examine whether, and if so how, jaw movement varies as a function of different vowel (/i/-to-/i/ vs. /a/-to-/a/), C1C2 type (/pk/ vs. /pp/ vs. /kk/), and speech rate (normal vs. fast) conditions. In doing this, we attempt to provide a more detailed description of C1C2 sequences.
We used an electromagnetic midsagittal articulometer (EMMA in Perkell et al., 1992; for a detailed description of this system and subsequent post-processing procedures, see Son (2008)). We made use of kinematic data from four transducers (the upper and lower lips for lip aperture, the tongue dorsum, and the lower central incisor for the jaw) for further analysis.
Eight native speakers (five females and three males) of Seoul Korean participated in the EMMA experiment. Kinematic data from Son’s (2013) /pk/ sequences in the low-vowel (/a/-to-/a/) context at two speech rates (normal vs. fast) were reused. In addition, we examined three consonantal sequences (/pk/ and controls (/pp/ and /kk/)) in the high-vowel (/i/-to-/i/) context as well as control /pp/ and /kk/ sequences in the low-vowel (/a/-to-/a/) context, which were elicited simultaneously at the time of data collection in Son (2013). The subjects, ranging from their mid-twenties to early thirties, were living in Connecticut, U.S.A. when we collected their articulatory data, where they had been pursuing graduate or post-doctoral research. In the pre-experimental, paper-and-pencil questionnaire, all of them identified themselves as native speakers of the Seoul-Korean dialect, reporting that they had lived in the Seoul metropolitan area for at least twenty-three years.1 They had all spent for approximately four years abroad on average and belonged to Korean communities of various kinds while abroad (e.g., Korean churches, Korean student associations, etc.). None of them had speech/hearing deficits. They were not informed of the purpose of the experiment before or after the experiment and were all financially rewarded after completing a production experiment.
The stimuli list is provided in (1). A total of 256 tokens (2 (Vowel contexts)×2 (Speech rates)×8 (Repetitions)×8 (Subjects)) were available both for /pk/ and /pp/ sequences. A total of 251 tokens were available for /kk/ since we failed to acquire four /kk/ tokens at normal rate and we excluded one /kk/ token at normal rate due to poor trajectory of the tongue dorsum.
Test sequences: /pk/
/ʧənapkanɨn kjesanhaki himtɨlə/ (borrowed from Son (2013: 670))
전압가는 계산하기 힘들어.
‘Voltage values are difficult to estimate.’
/ʧənipkie ta nawaissə/
전입기에 다 나와 있어.
‘Everything (you need) is on the resident document.’
Control sequences: /kk/
Control sequences: /pp/
We semi-automatically demarcated three gestural landmarks (e.g., the movement onset, peak velocity, and constriction onset) for the lip aperture gesture as we use the function lp_Findgest and constriction minima as we use the function lp_Snapex in MVIEW (Tiede, 2005) (Figure 1.i) (see Son (2011) for more detailed descriptions of data analysis using Tiede’s (2005) algorithm based on a velocity threshold). Using the function lp_Snapex, constriction maxima were also demarcated in the tongue dorsum and the jaw gestures. Constriction minima and constriction maxima shown with a red dot are pertinent to maximum constriction in the kinematic trajectories (e.g., the vertical tongue dorsum, lip aperture, and vertical jaw position) (Figure 1.ii). When the function lp_Snapex failed to capture minimum lip aperture points in time, we used values corresponding to the time point of the maximum constriction of the vertical tongue dorsum trajectory (Figure 1.iii). Based on the measures as shown in figures (1.i), (1.ii), and (1.iii), we borrowed criteria for token classification established in Son (2008)2.
C1 of heterorganic CiCk cluster sequences is unreduced if:
C1 of heterorganic CiCk cluster sequences are partially (or gradiently) reduced if:
lp_Findgest detects gestural landmarks (i.e., gesture onset, target attainment, and release) within a properly selected window, and
Spatial values for maximum constriction are less constricted than 3 times the standard deviation of the interquartile mean of heterorganic CiCk cluster sequences, but more constricted than 3 times the standard deviation of the interquartile mean in homorganic control utterances without the relevant constriction (in CkCk control utterances, a time point of the relevant constriction of C1 in CiCk was measured at the constriction maxima of homorganic CkCk clusters).
C1 of heterorganic CiCk cluster sequences are categorically reduced if:
lp_Findgest fails to detect gestural landmarks (i.e., gesture onset, target attainment, and release) within a properly selected window, or
Spatial values for maximum constriction are less constricted than 3 times the standard deviation of the interquartile mean of heterorganic CiCk cluster sequences, and not more constricted than 3 times the standard deviation of the interquartile mean in homorganic control utterances without the relevant constriction (in CkCk control utterances, a time point of the relevant constriction of C1 in CiCk was measured at the constriction maxima of homorganic CkCk clusters).
We also applied the criteria for token classification in (2) to categorize control /pp/ sequences into two gestural types (e.g., unreduced and partially reduced). Doing this for control /kk/ sequences, we categorize tokens as partially reduced if spatial values for the tongue dorsum constriction are less constricted than 3 times the standard deviation of the interquartile mean.
Raw data were converted to z-scores and used as input for further analysis. Linear mixed-effects models were constructed in R (R Development Core Team, 2014). The results of articulatory analysis were fitted with the lmer function from the lme4 packages (Bates et al., 2015). Specifically, we fitted a linear regression model with four kinematic measurements of the lip aperture closing movement (peak velocity, spatial displacement, acceleration duration, and closing movement duration), minimum lip aperture, and maximum vertical jaw position while taking into account Vowel context (low vowel /a/-to-/a/ vs. high vowel /i/-to-/i/), Speech rate (normal vs. fast), and Consonant sequence type (control /kk/ vs. assimilating /pk/ vs. control /pp/). Tukey HSD tests were used for post-hoc analysis.
With regard to a main effect of Vowel context (/i/-to-/i/ vs. /a/-to-/a/), the peak velocity and spatial displacement were smaller in the high-vowel context, with a reduction of –0.797 (SE±0.154) and of –0.613 (SE±0.155), respectively [t(498)=–5.179; t(491.1)=–3.954, all at p<0.0001] (/a/-to-/a/>/i/-to-/i/) (Figures 2.a and 2.b). Meanwhile, the duration of the lip aperture closing acceleration and the lip aperture closing movement were longer in the high-vowel context, lengthened by 0.397 (SE±0.167) and by 0.453 (SE±0.156), respectively [t(498)=2.371, p<0.05; t(498)=2.914, p<0.01] (/a/-to-/a/> /i/-to-/i/) (Figure 2.c). As for a main effect of Speech rate, there was no significant effect on any dependent variables (all at p>0.05) (Figures 2.d, 2.e., and 2.f).
The results of lip aperture closing movement also showed a main effect of Consonant sequence type on spatial displacement, acceleration duration, and movement duration [t(491.1)=3.153, p<0.01; t(498)=2.521, p<0.05; t(498)=6.101, p<0.0001], where /pp/ sequences exhibited greater vertical displacement by 0.489 (SE±0.155) and greater duration by 0.422 (SE±0.167) and 0.949 (SE±0.156), respectively (/pk/</pp) (Figures 2.h and 2.i). There was an interaction between Consonant sequence type and Speech rate on the duration of the lip aperture closing movement, with a reduction of –0.642 (SE±0.223) in /pp/ sequences when combined with fast rate [t(498)=–2.881, p<0.01], where shorter duration in /pk/ sequences was observed only in the normal rate (/pk/</pp/), and not in the fast rate (/pk/=/pp/) (Figure 2.j).
Taking everything into account, the peak velocity only varied with vocalic environments. Spatiotemporal properties showed the opposite patterns of one other with respect to vocalic contexts: greater spatial displacement was observed in the low-vowel (/a/-to-/a/) context and temporally longer closing duration in the high-vowel (/i/-to-/i/) context. Lastly, spatiotemporal reduction during the lip aperture closing movement was generally observed in the assimilating /pk/ sequences in comparison with control /pp/ sequences (/pk/</pp/) (Figure 2.i), with one exception (e.g., the duration of lip aperture closing movement in fast rate (/pk/=/pp/) (Figure 2.j).
Regarding lip aperture minima, there was interaction between Vowel context and Consonant sequence type, with a reduction of 0.236 (SE±0.075) in /ipki/ and of 0.212 (SE±0.075) in /ippi/ sequences [t(751)=3.125; t(751)=2.811, all at p<0.01], and between Speech rate and Consonant sequence type, with a reduction of 0.248 (SE±0.075) in /pk/ sequences when combined with fast rate [t(751)= 3.311, p<0.001]. The maximum constriction degree between the upper and lower lips did not vary with different vocalic contexts within the assimilating context (/apka/=/ipki/) (Figure 3.d). This was also true for homorganic control /pp/ sequences (/appa/=ippi/). In addition, there were Speech rate effects on the lip aperture minima as long as /p/ was included (normal (more constriction)<fast (less constriction) in both /VpkV/ and /VppV/), indicating that the fast rate is associated with less constriction (Figure 3.f). With regard to interaction between the two factors, from another perspective, the lip aperture was consistently less constricted in the order /kk/ (less constriction)>/pk/>/pp/ (more constriction) in each combination (Figures. 3.e (interacting with Vowel context) & 3.g (interacting with Speech rate)).
Referring to the criteria established to classify partially (or gradiently) reduced tokens (see (2) in section 2.2), we learned that 49 tokens were partially reduced and 6 tokens categorically reduced out of a total of 763 tokens. As shown in Table 1, partial (or gradient) reduction occurs dominantly in fast rate in 82% of tokens (40 out of 49) and categorical reduction in 100% of tokens (6 out of 6). With respect to vocalic environment, the high-vowel context is a condition for more frequent partial reduction in 59% of tokens (29 out of 49), compared to the low vowel in 41% (20 out of 49). As for Consonant sequence type, not a single sequence type demonstrated a dominant reduction frequency over the others in terms of partial reduction (37% (18 out of 49 tokens) for /kk/ sequences; 33% (16 out of 49 tokens) for /pk/ sequences; 31% (15 out of 49 tokens) for /pp/ sequences). Categorical reduction was typical of the assimilating /pk/ sequences. If homorganic control sequences showed any instance of gestural reduction, so did the assimilating /pk/ sequences.
With regard to vertical jaw position, there was a main effect of Vowel context, with an increase of 0.967 (SE±0.162) in the high-vowel context [t(647)=5.984, p<0.0001] (/a/-to-/a/</i/-to-/i/) (Figure 4.a), Speech rate, with a decrease of –0.424 (SE±0.163) in the fast fate [t(647)=–2.609, p<0.01] (normal>fast) (Figure 4.b), and Consonant sequence type, with an increase of 0.611 (SE±0.148) in the /pk/ sequence and of 0.358 (SE±0.147) in the /pp/ sequence [t(647)=4.124, p<0.0001; t(647)=2.441, p<0.05)] (/kk/<(/pk/=/pp/)) (Figure 4.c). Specifically, vertical jaw position was higher when a C1C2 sequence was flanked by high vowels (/i/-to-/i/). As for different consonantal sequence types, they demonstrated binary distribution (/kk/<(/pk/=/pp/) (cf., ternary distribution for maximum lip constriction (/kk/</pk/</pp/). Lastly, vertical jaw position varied with speech rate, indicating that the mouth is more open when speakers talk faster.
Examining lip aperture maxima, results from the current study indicated that the number of tokens classified as partially reduced were very similar across different consonantal sequence types (18 tokens for control /kk/ sequences vs. 16 tokens for test /pk/ sequences vs. 15 tokens for control /pp/ sequences). Although we were not able to separate the first consonant from the second in control CiCi (/pp/) and CkCk (/kk/) sequences in terms of maximum constriction, the maximum constriction of these homorganic sequences is at least not considered to be reduced but fortified, compared to lenis (/pk/). Evaluated within the same cluster sequence type, the probability across different sequences was quite comparable (6.3% for the test /pk/ sequences vs. 5.9% and 7.2% for the control /pp/ and /kk/ sequences, respectively). Based on this, we tentatively conclude that i) partial reduction occurs to some extent across the board, if there is any, and ii) gradient reduction in magnitude may be attributed to the general quantitative properties of gestures (Browman & Goldstein, 1986, 1989), rather than being specific to place-assimilating sequences such as /pk/ (e.g., Jun, 1996).
However, caution should be taken before we come to a conclusion since this is a preliminary study which covers only a subset of data: the current analysis only includes the within-word boundary condition, and not the across-word boundary condition. Another disadvantage may be stem from following Son’s (2008) arbitrary requirements for categorizing partially (or gradiently) reduced tokens, which referred to an interquartile mean and interquartile standard deviation in search of their resistance to outliers. Use of an interquartile mean and three times its standard deviation to define partial (or gradient) reduction still remains an arbitrary method. An alternative can be found in Son et al. (2012), where frequency distributions are interpreted using a histogram as they provide a mathematical analysis based on point-attractors in the task-dynamics model of speech production (Nam et al., 2012).
In the current study, several kinematic properties of the lip aperture gesture reflected weaker articulation in the spatialtemporal dimension (e.g., closing spatial displacement, closing acceleration duration, and closing movement duration) in the assimilating context (/pk/</pp/). Traditionally, Korean place assimilation is driven by applying a feature-changing rule as shown in (3). Using narrow phonetic transcription, two phonological processes are involved in Korean assimilating /pk/ sequences. One is a fortification rule in which a lenis obstruent in the onset is fortified after an obstruent in the coda (Silverman, 2017). The other is a place-assimilation rule in which the place feature of the coda becomes identical with that of the onset (Kim-Renaud, 1974). In addition, fortification occurs regardless of the application of the place-assimilating rule and the lenis stop is assumed to be realized in the coda (e.g., /pk/→[kk*]; /pp/→[pp*]).
Recall that spatiotemporal reduction is not peculiar to the heterorganic assimilating context but is even applied to homorganic controls. We further fitted a linear regression model for the three dependent variables after excluding reduced tokens of any kind. The results showed that a weaker gestural event of C1 from heterorganic assimilating sequences was also consistently observed [t(460.4)=3.096, p<0.01 for the closing displacement; t(467)=2.511, p<0.05 for the closing acceleration duration; t(467)=6.124, p<0.0001 for the closing movement duration] (/pk/</pp/). Given this, we speculate on two possibilities to account for this. One is that spatiotemporal reduction observed even in unreduced tokens generally occurs at the phonetic execution level. Speakers’ preemptive action with more reduction in C1 may, in part if not completely, reflect speakers’ strategy that they simply do not exert much articulatory efforts towards a perceptually unrewarding gestural event. At the phonological level of representation, this reduction could be a consequence associated with gestural overlap or independent of gestural overlap in the gestural score (Chen, 2003; Browman & Goldstein, 1990a, 1990b, 1990c; among others). Alternatively, this could be an output of the application of probability-based constraints in optimality theory (Jun, 1995). Since it is beyond the scope of the current study to discuss which theoretical frameworks are a better fit to account for data we have acquired, we leave this issue for further study. The other possibility to account for a weaker gestural event of C1 being consistently observed from heterorganic assimilating sequences is that stronger articulation of fortified onset [p*] may have extended onto coda [p] in the homorganic C1C2 sequences. Note that Barry (1991) observed in his electropalatography study that in an assimilating sequence from English there was a 43% increase in fast rate, compared to slow rate, in terms of the duration of the rear tongue body gesture in C2 and a 78% increase in terms of the duration of velar closure in C2. (e.g., ‘handgrenade’). Given that, it is plausible to assume that C2 in /pp/ sequences may enhance articulatory strength not only its temporal domain but also spatial domain over C1 by extending its articulatory strengthening to its homorganic C1 (e.g., gestural blending in Romero (1992)). As such, strengthening may be spatiotemporally overriding the closing movement of C1.
Caution should be taken, however, since similar spatiotemporal properties in the lip aperture closing movement are observed between lenis and fortis when a single segment is tested in /a/-to-/a/ and /i/-to-/i/ contexts (/VpV/=/Vp*V/ in Son et al. (2012)). We should include a more balanced set of articulatory data so that we may test the nature of coda in intervocalic homorganic lenis-lenis (e.g., assimilating and non-assimilating) and lenis-fortis (e.g., homorganic and heterorganic) sequences, as well as in an intervocalic singleton fortis stop control. Doing this will enable a better understanding of articulatory characteristics of lenis in coda, irrespective of the target of place assimilation.
Öhman (1967) observed that consonantal articulation (e.g., coronal) is achieved, regardless of different vocalic contexts. Being compatible with this, Korean /pk/ sequences also demonstrated invariable maximum constriction degrees between two different vocalic contexts (/apka/=/ipki/). This was also true for homorganic control /pp/ sequences (/appa/=/ippi/). The results of the current study support that articulatory tiers are bifurcated so that consonantal articulation occupies a separate tier, independent of vocalic articulation (Browman & Goldstein, 1992; Öhman, 1967). In particular, consonantal articulation is less open than vocalic articulation in terms of constriction degree in the oral tract. Focusing within a single tier of articulation (e.g., the vocalic tier), low vowels, for example, are less constricted than high vowels (Ladefoged, 2001).
Although the scope of the current study is limited morphologically to the within-word condition, categorical reduction of the lip aperture gesture did not at all occur in the high-vowel context (/ipki/), though we sometimes observed such cases in the low-vowel context (/apka/) in conjunction with fast rate. This elision indicates that speakers are sensitive not only to paralinguistic factors such as speech rates (see speech rate/style formally incorporated within a probability-based, optimality-theoretic approach in Jun (2004); frequency effects in Pouplier et al. (2017)), but also linguistic factors such as vocalic contexts. With respect to occasional categorical reduction, one can conjecture that speakers are aware that it may require too much effort to complete a lip aperture gesture from lower jaw position (e.g., /a/-to-/a/), especially in conjunction with fast rate. At the speech-planning level, therefore, segmental deletion happens as a consequence of this (cf., residual tongue tip gesture detected in C1 for American English in Browman & Goldstein (1990c)). Under this assumption, articulatory efforts can be maximally reserved when the energy cost of reaching a target is unrewarding from the standpoint of speakers, although inter-/intra-spearker variability still remains. Referring to the results of the current study, a significant increase in energy may signal a greater peak velocity and greater spatial displacement in the low-vowel context (e.g., /a/-to-/a/>/i/-to-/i/).
The occurrence of partial (or gradient) as well as categorical reduction in the lip aperture gesture was obviously rate-dependent, demonstrating higher frequency in fast rate (82% for partial reduction and 100% for categorical reduction as shown in Table 1). This is compatible with a general observation where lenition is more likely to occur easily in fast rate and the vocal tract is more open in this condition (e.g., close (unfricated) approximant in slow natural speech vs. more open approximant in faster speech rate in Kirchner (1998: 257)). Not having acoustic descriptions available which correspond to all articulatory data used in the current study, we are not ready to provide a fully balanced acoustic analysis of C1 /p/ along with articulatory counterparts. Nevertheless, in conformity with Kirchner’s (1998) observation of more open articulation in faster speech rate, we found that the jaw is also lower in fast rate (normal>fast), in addition to overall rate effects on the lip aperture minima in the assimilating /pk/ and control /pp/ sequences. This indicates that speech-rate effects are also immediately reflected in jaw articulation as well as the primary articulator. Through this, the spatially reduced mandibular cycle (see basic mandibular cycle for consonants, vowels, and syllables, etc. (MacNeilege & Davis (1993: 341)) can facilitate the production of a syllable (e.g., V-to-C) in fast rate, which contributes to a reserve of articulatory efforts as a consequence (Jun, 1996).