Intrinsic fundamental frequency of Amharic vowels

Intrinsic fundamental frequency (IF0) refers to the cross-linguistic tendency for high vowels to have a higher F0 than low vowels (Whalen and Levitt, 1995). IF0 is often thought to be a result of mechanical coupling, where tongue movement causes changes in the laryngeal structures, perturbing the F0. Ohala and Eukel (1987) propose that high vowels lead to increased vertical tension on the true vocal folds, which raises F0. In contrast, the tongue-compression hypothesis proposes that low vowels induce vocal fold slackening, which lowers F0 (Ewan, 1979). The purpose of the present study is to investigate the cause of IF0 by comparing high, mid, and low vowels in Amharic, using audio and EGG signals to measure F0 and contact quotient for different vowel heights. Amharic was chosen because it contrasts three heights of central vowels. The goals are to confirm that IF0 effects are found in Amharic and determine if voice quality changes further support either hypothesis. Discussion will focus on how F0 and contact quotient differ by vowel height, and whether the results support the tongue-pull hypothesis (where IF0 is driven by high tongue position), the tongue-compression hypothesis (where IF0 is driven by low-back tongue position), or both.Intrinsic fundamental frequency (IF0) refers to the cross-linguistic tendency for high vowels to have a higher F0 than low vowels (Whalen and Levitt, 1995). IF0 is often thought to be a result of mechanical coupling, where tongue movement causes changes in the laryngeal structures, perturbing the F0. Ohala and Eukel (1987) propose that high vowels lead to increased vertical tension on the true vocal folds, which raises F0. In contrast, the tongue-compression hypothesis proposes that low vowels induce vocal fold slackening, which lowers F0 (Ewan, 1979). The purpose of the present study is to investigate the cause of IF0 by comparing high, mid, and low vowels in Amharic, using audio and EGG signals to measure F0 and contact quotient for different vowel heights. Amharic was chosen because it contrasts three heights of central vowels. The goals are to confirm that IF0 effects are found in Amharic and determine if voice quality changes further support either hypothesis. Discussion will focus on how F0 and cont...


INTRODUCTION 1
Cross-linguistically, the fundamental frequency (F0) of vowels has been found to correlate 2 with vowel height, with high vowels having a higher F0 than low vowels . 3 This phenomenon is referred to as the intrinsic fundamental frequency of vowels (IF0) or intrinsic 4 pitch of vowels (IPV). In a study of 35 languages representing 11 different language families, Whalen 5 and  found evidence of intrinsic F0 of vowels in every reported language, with effect 6 size varying from 5 Hz to over 20 Hz . The authors suggest that the 7 phenomenon is universal, but the cause of the phenomenon remains unclear. Commonly cited 8 hypotheses attribute IF0 to mechanical coupling of the oral vocal tract and laryngeal structures, 9 where the F0 perturbation is due to tongue or jaw movement inducing changes on laryngeal 10 structures (Ladefoged, 1964;Ohala, 1978;Ewan, 1979;Chen, Whalen & Tiede, 2019). Yet, precisely 11 how tongue or jaw movement might lead to changes in vocal fold vibration remains unclear. Other 12 hypotheses posit that IF0 is due to volitional control for vowel production, either as auditory 13 enhancement to cue vowel category (Diehl, 1991) or as part of the articulatory gesture for the vowel 14 (Sapir, 1989). The goal of the present study is to test predictions of articulatory accounts of IF0 and 15 examine the relationship between oral vocal tract movement (i.e., tongue and jaw) and laryngeal 16 movement proposed by each hypothesis. 17

a. Background 18
Perhaps the most frequently cited account of IF0 is the TONGUE-PULL HYPOTHESIS, 19 originally proposed by Ladefoged (1964). The tongue-pull hypothesis states that IF0 is driven by the 20 tongue body's high position pulling on the hyoid bone. The upward movement of the hyoid bone in 21 turn pulls the thyroid cartilage forward, leading to increased laryngeal height and vocal fold tension. 22 Together, the increased laryngeal height and vocal fold tension are believed to raise F0 (Ladefoged, 23 1964). This is schematized in Figure 1. however, data from  and Sapir (1989) suggest the opposite pattern holds: 48 /u/ has a higher F0 than /a/ despite having a lower laryngeal position. It is also important to note 49 that an increase in F0 is not caused exclusively by larynx raising. Volitional F0 raising is primarily 50 caused by activation of the cricothyroid, the muscle located between thyroid and cricoid cartilage 51 (refer to Figure 2 for identification of thyroid and cricoid cartilage). Contraction of this muscle 52 causes the thyroid cartilage to tilt forward, lengthening the vocal folds and increasing F0. F0 can also 53 raise due to an increase in vocal fold stiffness and increased airflow (Zhang, 2016). Therefore, it is 54 not surprising that laryngeal height alone does not explain IF0 despite finding correlation between 55 laryngeal height and fundamental frequency in other tasks (e.g., pitch glides, singing). 56 Ohala (1978) provides another version of the tongue-pull hypothesis. Rather than relating 57 tongue position to laryngeal height, he proposes that the high tongue position of high vowels leads 58 to a pulling of the aryepiglottic folds, which then pull the ventricular ("false") vocal folds away from 59 the true vocal folds; this in turn increases vertical tension of the true vocal folds. The vertical 60 movement of the false vocal folds reduces the damping effect of the false vocal folds on the true 61 vocal folds; together with the increase of vertical tension of the true vocal folds, this leads to the 62 increased F0 in high vowels (Ohala, 1978;Ohala & Eukel, 1987). This is schematized in Figure 2. 63 6 increased tongue-pull for bite blocks of greater width. They found that found high vowels had a 67 higher pitch than low and mid vowels in all conditions, with low and mid vowels pairing together. 68 The magnitude of the effect increased only with the largest bite block (10 mm). These findings were 69 interpreted as support for the revised tongue-pull hypothesis. 70 This hypothesis has several underlying assumptions, some of which remain untested. The 71 first is that IF0 is due to passive laryngeal movement that is an automatic consequence of tongue 72 movement for vowel articulation. Evidence from Ohala and Eukel's (1987) bite block study as well 73 as findings in infant speech (Whalen, Levitt, Hsiao & Smorodinsky, 1995) and speech of Deaf adults 74 (Bush, 1981) support this assumption. Additionally, Whalen et al. (1998) completed an EMG study 75 of cricothyroid activity during vowel production. If IF0 were volitional (i.e., not passive), they would 76 expect to see increased cricothyroid activity for high vowels (high F0) compared to low vowels (low 77 F0). Instead, the authors found that cricothyroid activation did not follow the F0 pattern found in 78 vowels. They interpreted their results as support of the hypothesis that IF0 is a result of passive 79 laryngeal movement and an automatic consequence of vowel articulation since there was no 80 evidence of volitional muscle contraction leading to changes in F0 (Whalen, Gick, Kumada & 81 Honda,1998). Second, Ohala's tongue-pull hypothesis assumes that the false vocal folds dampen 82 true vocal fold vibration in a neutral (i.e., mid) tongue position and that vocal fold dampening lowers 83 F0; therefore, the tongue pulling the false vocal folds vertically allows for increased F0 by reducing 84 the dampening effect. There is some research to bear on these assumptions: Bailly   Source: book 'Anatomy and Physiology, https://openstax.org/details/books/anatomy-and-physiology. Adapted by B.
Ramos for this paper.

9
Crevier-Buchman & Halimi, 2019). Similarly, Lancia and Grawunder (2014) found that /a/ was 120 more prone to laryngealization compared to a more fronted vowel, such as /i/, due to tongue 121 retraction required for /a/ (Lancia & Grawunder, 2014). This is consistent with the model of the 122 vowel space provided by Esling (2005), who claims that low-back vowels are produced by the 123 laryngeal articulator; that is, vowels such as /a/ can be considered pharyngeal vowels given the 124 relationship between tongue retraction and the larynx. In contrast, other vowels do not involve the 125 laryngeal articulator as a primary articulator. Esling adds that vowels involving tongue retraction are 126 more susceptible to increased laryngeal constriction. With this view, the findings presented in Esling  The support for the tongue-compression hypothesis is only partial, however. Recall that Ewan's 131 tongue-compression hypothesis also stated that tongue retraction leads to vocal fold slackening. Yet 132 findings from the studies cited here suggest that tongue retraction causes decreased F0 by increasing 133 laryngeal constriction, not by vocal fold slackening, as Ewan predicts. 134 In addition to movement of the tongue, there is some evidence that jaw height also 135 influences IF0, though the results can vary by speaker (Zawadzki & Gilbert, 1989 completed a study of IF0 in 8 English vowels using x-ray microbeam data of 40 speakers from two 142 databases. While tongue height and F1 show high correlations with F0, their findings suggest that 143 jaw height contributes more to IF0 than tongue height. The authors state that the muscular chain 144 between the mandible, hyoid, and larynx may be behind the connection between jaw height and 145 changes in F0, though they do not offer a predicted sequence of actions responsible for IF0. 146 Erickson et al. (2017) propose that jaw opening leads to backward translation of the hyoid bone. 147 The hyoid bone is connected to the thyroid cartilage through thyrohyoid muscles; when the hyoid 148 bone is moved back, they propose that the thyroid cartilage is also forced to rotate posteriorly 149  The hypothesized link between jaw movement and IF0 remains untested in the literature. 152 Interestingly, in both Zawadzki & Gilbert (1989) and Chen et al. (2019), /a/ and /ɔ/ seem to have 153 lower F0 values than expected given their jaw height alone when compared to vowels of similar jaw 154 height. Though neither study compared F0 of front and back low vowels, the results would be 155 compatible with the tongue compression hypothesis and suggest that IF0 has multiple sources rather 156 than a link between the vocal folds and a single articulator, such as the jaw. 157 Unlike the previous hypotheses, Sapir (1989) proposes instead that the laryngeal 158 configuration during vowel production is volitional and forms part of the articulatory goal for the 159 vowel to reach its acoustic target. His HORIZONTAL-VERTICAL PULL HYPOTHESIS proposes that IF0 160 is due to extrinsic laryngeal muscle activation leading to changes in laryngeal configuration during 161 vowel production (Sapir, 1989). According to his account, extrinsic laryngeal muscle activation, not 162 tongue movement, causes changes in laryngeal and pharyngeal configurations in order to reach an 163 acoustic target for vowel quality (i.e., F1 and F2 targets). The laryngeal configurations also have 164 consequences for fundamental frequency. As summarized in Sapir (1989), during production of /i/, 165 contraction of the suprahyoid muscles cause the hyoid to move forward and superiorly, which leads 166 to increased pharyngeal space in the anterior-posterior dimension and lowers F1 for /i/. This same 167 movement causes increased true vocal fold tension, which increases F0. Since low vowels do not 168 target a low F1, there is less activation in the suprahyoid muscles, and the vocal fold tension is not 169 altered, leading to a lower F0 for low vowels compared to high vowels. This hypothesis remains 170 largely untested outside of Sapir's own studies (Sapir 1989 Table I Horizontal-vertical pull (Sapir) Phoneme-specific extrinsic laryngeal muscle activation leads to changes in F0

179
Non-articulatory explanations of IF0 have also been proposed, including the acoustic 180 coupling hypothesis (Flanagan & Landgraf, 1968) and the auditory enhancement hypothesis (Diehl, 181 1991). The auditory enhancement hypothesis proposes that phonetic features of consonants and 182 vowels covary to enhance the auditory feature of that segment. With respect to IF0, the hypothesis 183 suggests that changes in IF0 are due to purposeful enhancement of vowels to make contrasts 184 perceptually distinct (Diehl, 1991).   Finally, the acoustic coupling hypothesis proposes that the increased F0 of high vowels is an 206 effect of the first formant on the fundamental frequency (Flanagan & Landgraf, 1968). The 207 hypothesis states that when the vocal folds vibrate at a rate that is near the resonant frequency of F1, 208 acoustic coupling between the vocal tract and vocal folds occurs, such that F0 increases due to 209 coupling with F1. Since high vowels have a low F1 that can be near to F0, acoustic coupling occurs 210 for high vowels, causing an increase in F0 for high vowels. Ewan (1979) tested the acoustic coupling 211 hypothesis, comparing nasal /m/ in two phonetic environments, /ama/ and /umu/. Despite the 212 first nasal formant being low in both environments, the F0 of /m/ differed depending on the vowel 213 14 context. The nasal in /ama/ had lower F0 than the nasal in /umu/. In both cases, the F0 of the 214 intervocalic /m/ was consistent with the vowel context and did not appear to be influenced by the 215 first nasal formant (Ewan, 1979). Guérin and Boë (1980) also found evidence that acoustic coupling 216 cannot account for IF0. Using a two-mass voice source model to simulate the effect of vocal tract 217 changes on F0, the authors found that F0 was positively correlated with F1. This is the opposite 218 pattern found in natural speech as it predicts increased F0 with low vowels, not high vowels (Guérin 219 & Boë, 1980). Due to the evidence suggesting acoustic coupling does not account for IF0, this 220 hypothesis has generally been abandoned in the literature as a possible explanation. 221

b. Current Study 222
The cause of IF0 remains unknown, though the summary of previous research presented above 223 suggests IF0 may have multiple sources. The purpose of the present study is to investigate the 224 predictions of articulatory hypotheses by comparing high, mid, and low vowels in Amharic using 225 acoustic data and electroglottography (EGG) to examine the relationship between tongue position, 226 F0, and Contact Quotient (CQ), a measure of vocal fold contact area during vibration that is 227 thought to relate to vocal fold tension (Herbst, 2020). Amharic is a Semitic language spoken in 228 Ethiopia and by a large diaspora community. Its seven-vowel inventory allows for a detailed 229 comparison of IF0 along the dimensions of both height and frontness as it contrasts front, central, 230 Figure 5. Amharic vowel inventory and back vowels among high and mid vowels in addition to a single low vowel, /a/ ( Figure 5) 231 (Hayward & Hayward, 1992). Although previous research by Ado (2011) on Amharic vowels did not 232 find a statistically significant difference between F0 of high vowel /i/ and low vowel /a/, the 233 difference in F0 values for /i/ and /a/ was consistent with effect sizes found in Whalen and Levitt 234 (1995). Previous studies examining the cause of IF0 have relied heavily on data from European 235 languages (e.g., English or German). This study contributes acoustic and articulatory data from an 236 under-represented language to address a long-standing question in phonetics. 237 The predictions of each hypothesis are as follows (Table II): under the tongue-pull hypothesis, 238 high vowels in Amharic should have the highest F0, followed by mid and low vowels, which should 239 pair together. Conversely, under the tongue-compression hypothesis, high and mid vowels would be 240 expected to pair together given the lack of tongue retraction required for those vowels. However, 241 since we know from the recent body of literature on IF0 in vowels that IF0 is gradient, it is likely 242 that both tongue pull and tongue compression would be responsible for IF0. In this case, high, mid, 243 and low vowels should all differ in F0. Additionally, since both hypotheses ultimately relate the 244 cause of IF0 to differences in vocal fold tension, differences in contact quotient would be expected 245 to follow the predicted F0 patterns. Since testing the horizontal-vertical pull hypothesis requires 246 measuring muscle activation (e.g., with electromyography), this hypothesis is not tested directly in 247 this study. Similarly, the current study does not directly assess the contribution of jaw height to F0, 248 but predictions based on these hypotheses are provided in Table II. 249 Three of the articulatory hypotheses (tongue-pull, tongue-compress, and jaw height) assume that 250 IF0 is an automatic consequence of articulator movement. That is, as the tongue or jaw moves, it 251 induces changes on the larynx leading to F0 perturbations. It follows, then, that these changes 252 should occur regardless of vowel category. For example, under the tongue-pull hypothesis, an /i/ 253 that is produced with a higher tongue position should have a higher F0 and more tense vocal folds 254 than an /i/ produced with a lower tongue position. Therefore, predictions are also provided for F0 255 and CQ based on F1 and F2, independent of vowel category in Table III Participants included 8 native speakers of Amharic (5 men and 3 women) living in the San 266 Diego area but who were originally from Addis Ababa, Ethiopia (Table IV). All participants were L2 267 English speakers and speak Amharic as the primary language in their home. Recordings from 268 Speaker 1 and Speaker 7 were excluded due to noisy EGG signals that resulted in inability to 269 calculate contact quotient. Individuals with a history of speech, language, hearing, or neurological 270 disorders were excluded from participation, as were those who were not literate in Amharic, because 271 the study protocol required reading the Amharic orthography.  sixth order vowel, /ɨ/ was not included as a second vowel in the /tVtV/ frame as it cannot occur 279 word-finally. All permutations resulted in nonce words, which were reviewed by a native Amharic 280 speaker prior to use in the study to ensure that none of the words were in fact real words. Nonce Stimulus sentences were presented one at a time on a screen using Amharic orthography, which 291 is an alphasyllabary, where one character represents a CV syllable. Participants were instructed to 292 read each sentence aloud at a comfortable rate and loudness and to self-advance to the next sentence 293 at their own pace. Target words within the carrier sentences were repeated five times for each 294 participant, with sentences presented in the same order each time. A sixth recording was collected 295 for Speaker 6 due to frequent reading disfluencies during the first recording. There were 30 296 repetitions of each target vowel, yielding 210 stimulus items per participant. The author was present 297 in the recording booth at the time of the recording to ensure participants read all stimulus items. If a 298 word was read disfluently or in error, the speaker was asked to repeat it, and the second production 299 was used for analysis. Reading was judged to be disfluent if the speaker produced repetition, 300 prolongation, or blocking of segments. In the case of Speaker 6, who had a higher rate of 301 disfluencies in reading compared to other participants, 88% of his recordings were segmented for 302 analysis (compared to 100-98% for other participants), though the overall number of tokens 303 analyzed was equal to other participants given the additional recording. 304

e. Instrumentation and segmentation 305
Recordings were done in a sound-attenuated booth. Audio data were recorded at a 44.1 kHz 306 sampling rate and 16-bit quantization rate using a Shure SM10A head-mounted microphone. The 307 electroglottography (EGG) signal was recorded using an EG2-PCX electroglottograph from Glottal 308 Enterprises. EGG and audio signals were collected simultaneously using a Focusrite Scarlett 6i6 2nd 309 generation pre-amplifier and digitizer. EGG measures vocal fold contact area during vocal fold 310 vibration (Herbst, 2020). Contact quotient (CQ) is a measure of the duration vocal fold contact in 311 relation to the total duration of the vibratory cycle. An increase in contact quotient should be 312 interpreted as increased vocal fold adduction (e.g., through general laryngeal and pharyngeal 313 constriction) or increased contraction of the thyroarytenoid--the muscle responsible for increased 314 true vocal fold tension (Herbst, 2020). Therefore, CQ can be used to examine differences in tension 315 proposed by the tongue-pull and tongue-compression hypotheses. 316 Audio and EGG data were segmented in Praat (Boersma & Weenink, 2020). Target vowels were 317 segmented from the onset of clear formants to the end of clear formants (Figure 6(a)). In cases of 318 short vowels consisting of three or fewer glottal pulses in the EGG waveform, the boundary for 319 vowel segments were extended just beyond the beginning and end of glottal pulses regardless of 320 formant onset and offset in the spectrogram. This was done in order to capture all EGG pulses for 321 analysis (Figure 6(b)). automatically in VoiceSauce using the Praat settings. If those values were also judged to be 330 mistracked by the same criteria, values were corrected manually in Praat. EGG waveforms were 331 analyzed using EGGworks. Mean CQ was collected over the duration of the segmented vowel using 332 the hybrid method (Howard, 1995). The hybrid method uses two different methods for defining 333 vocal fold contacting and decontacting. Onset of vocal fold closure is taken to be the peak in the 334 derivative of the EGG signal. The end of contact was determined using a 25% threshold. Contact 335 quotient values below 0.2 and above 0.8 were excluded, as these values are believed to be outside of 336 the range of normal voicing. Mean values were calculated for all measures from the middle one third 337 of each vowel for analysis. This was done to further control for co-articulatory effects of the 338 adjacent consonants on F0, F1, or F2 of the vowel. F1, F2, F0, and CQ values were z-scored for 339 each speaker, and values greater than 2.5 standard deviations away from each speaker's mean were 340 excluded as these represented outliers. 341

f. Analysis 342
The data were analyzed in R (version 3.6.2, R Core Team, 2019) using raw values from the 343 middle one-third of each vowel. Raw data for each token were z-scored for each speaker and values 344 falling greater than 2.5 standard deviations away from the mean were excluded as these were 345 believed to represent outliers. Linear mixed effects analyses were performed using lme4 (Bates,

363
The means and standard deviations of fundamental frequency (F0), contact quotient (CQ), 364 F1, F2, and duration for each vowel category are provided in Table V. F0 is highest for /i/ and 365 lowest for /a/, consistent with findings of previous studies in IF0 Ado, 366 2011). Overall, high vowels have the highest mean F0, followed by mid vowels, and finally /a/, 367 though there are within-height differences between high and mid vowels. CQ results will be 368 discussed in section 3.

369
A linear mixed effects analysis was performed to assess the effect of each factor (vowel, F1, 370 F2) on F0. Significance of a targeted factor was assessed with nested model comparison between a 371 model that included the targeted factor to a baseline one where that factor was removed from the 372 model, but whose random error structure was identical. This was done using the anova() function in 373 R (R Core Team, 2019) There was a significant main effect of vowel quality on F0 (X 2 (1) = 271.33, p 374 < .001). See Table VI for coefficients and standard errors for the model of F0 as a function of vowel 375 category. Post-hoc pairwise comparisons between vowels in the model with Tukey alpha correction 376 found no difference between any two pairs of high vowels /i, ɨ, u/, but F0 values for all high vowels 377 were significantly higher than mid and low vowels. Mid vowels /e, ə, o/ were not significantly 378 different from each other with respect to F0. All non-low vowels differed significantly from F0 of 379 /a/, with /a/ having the lowest F0, as expected. These findings are illustrated in Figure 7, where 380 high vowels have the highest F0s, followed by mid vowels, and finally low vowel /a/. Notably, there 381 are no significant differences in F0 between front and back vowels at the same phonological height 382 category. 383  To ensure that differences across vowel categories were not due to inherent differences in 388 was not found to be significant (X 2 (1) = 0.7568, p = 0.3843), which suggests that duration does not 391 affect F0 once vowel quality is accounted for. 392 Next, we turn to the effect of F1 and F2 on F0. Recall, that the tongue-pull, tongue-393 compression, and jaw height hypotheses state that intrinsic F0 of vowels is caused by mechanical 394 coupling of the tongue and larynx. Based on this, we would predict that F0 perturbations would 395 differ even within vowel categories if tongue or jaw height differs between two vowels of the same 396 category. Therefore, the effects of F1 and F2 on F0 were also examined independent of vowel 397 category to investigate changes in F0 as a result of tongue position. The relationship between F1, F2, 398 and F0 was examined with a linear mixed effects model including participant as a random intercept. 399 Table VII shows the summary of the full model with coefficients and standard errors of fixed 400 effects. The interaction between F1 and F2 was not found to be significant (X 2 (1) = 0.7121, p = 401 0.3988). That is, changes in F1 do not contribute to the effect of F2 on F0. This is expected as we 402 do not see an effect of vowel backness on F0 that differs depending on vowel height in Figure 8. 403 Results revealed a significant main effect of F1 (X 2 (1) = 192.31, p < 0.001), indicating that 404 F0 decreases as F1 increases. This is seen in Figure 8 and Figure 9(a), which show that F0 decreases 405 as speakers produce vowels lower in the vowel space. The main effect of F2 was not significant (X 2 406 (1) = 0.2034, p = 0.652), indicating that vowel frontness does not have an effect on F0. This is seen 407 in Figure 9(b); as F2 increases (increased frontness), F0 remains largely unchanged. 408 409 CQ was used as measure of vocal fold tension. Recall that CQ is a measure of the relative 416 duration vocal fold contact during a vibratory cycle, and an increase in contact quotient is typically 417 interpreted as increased vocal fold adduction or contraction of the thyroarytenoid (Herbst, 2020) 418 Mean CQ values for each vowel category can be seen in Table V  As previously stated, the tongue-pull, tongue-compression, and jaw height hypotheses state that 427 intrinsic F0 of vowels is caused by mechanical coupling of the tongue and larynx. Under the tongue-428 pull and tongue-compression hypotheses, tongue movement induces changes in vocal fold tension, 429 which in turn lead to F0 perturbations. Based on this, we would predict that vocal fold tension 430 would differ even within vowel categories for the tongue-pull or tongue-compression hypothesis. 431 The effects of F1 and F2 on CQ were examined independent of vowel category to investigate 432 changes in CQ as a result of tongue movement. The relationship between F1, F2, and CQ was 433 examined with a linear mixed effects model including participant as a random intercept and F1 and 434 F2 as fixed effects. Table VIII shows the summary of the full model with coefficients and standard 435 errors of fixed effects. Nested model comparison revealed the interaction between F1 and F2 to be 436 significant (X 2 (1) = 9.9346, p < 0.05); however, the significance was due to improved model fit 437 without the interaction term compared to with. The effect of F2 was found to be not significant (X 2 438 (1) = 0.8707, p = 0.3508). Similarly, was the effect of F1 found to be not significant (X 2 (1) = 439  The purpose of the study was to examine IF0 of Amharic vowels and compare predictions 450 of different articulatory hypotheses that have been proposed in the literature. Predictions of each 451 hypothesis are outlined in Table II. Under the tongue-pull hypothesis, high vowels in Amharic were 452 expected to have the highest F0, followed by mid and low vowels, which were expected to pair 453 together. According to the tongue-compression hypothesis, high and mid vowels were expected to 454 pair together, and low vowel /a/ was predicted to have the lowest F0. If both tongue-pull and 455 tongue-compression both play a role in IF0, high, mid, and low vowels were expected to differ in 456 F0. Additionally, since both hypotheses relate the cause of IF0 to differences in vocal fold tension, 457 differences in contact quotient were expected to follow the predicted F0 patterns. Since testing the 458 horizontal-vertical pull hypothesis requires measuring muscle activation (e.g., with 459 electromyography), predictions for this hypothesis were not described. Similarly, the current study 460 did not directly assess the contribution of jaw height to F0, though F0 was expected to vary 461 generally by vowel height based on previous studies. 462 The study found that F0 varied as a factor of vowel category due to differences in vowel 463 height. High vowels had the highest F0, followed by mid vowels, and finally, low vowels. No 464 significant differences were found within height groups (e.g., /i/ was not different from /u/). 465 Similarly, when examining the effect of F1 and F2 on F0, the study found that F1 is predictive of F0 466 differences, while F2 is not. F0 varied by vowel height but not by backness. With respect to CQ, the 467 study revealed no effect of vowel category on CQ, nor was there an effect of F1 or F2 on CQ 468 independent of vowel category. 469 470 articulators. Indeed, the general trend of F0 varying by height in this study seems to support the 499 hypothesis that IF0 is a result of mechanical coupling. On the other hand, the phonological 500 grouping of F0 despite differences in F1 suggests that other factors contribute to F0, such as 501 auditory enhancement or forming part of the articulatory goal for vowel production (Sapir's 502 horizontal-vertical pull hypothesis). The implications of these findings as they relate to the proposed 503 causes of IF0 are discussed below. 504 The tongue-pull hypothesis cannot account entirely for the results of the study for three 505 reasons. First, F0 of mid and low vowels did not pair together. Instead, F0 was found to be gradient 506 with different values for high, mid, and low vowels; this finding was expected based on results of 507 previous studies. Second, the study found that high vowels, /i/ and /ɨ/, and mid vowels /e, ə, o/ 508 differed significantly with respect to F1 yet did not differ in F0. In other words, differences in F1 did 509 not correlate with differences in F0 for these groups of vowels, which does not support the 510 prediction that higher tongue position necessarily leads to increased F0. Instead, a phonological 511 grouping of F0 was found between high and mid vowels. Finally, there were no differences found in 512 CQ, suggesting no changes in vocal fold tension. Therefore, the tongue-pull hypothesis does not 513 appear to be the best account of IF0. It is possible that tongue-pull contributes to the effect for high 514 vowels, but tongue-pull alone does not account for the data. 515 Likewise, the tongue compression hypothesis cannot fully account for this study's findings 516 for the reasons cited above: namely, the gradient effect and lack of difference in CQ for /a/ 517 compared to other vowels. Tongue retraction may contribute to laryngeal changes that cause 518 reduced F0, but the laryngeal changes due to tongue retraction do not seem to be those predicted in 519 the hypothesis: thick, slack vocal folds. Instead, we see that vocal fold tension, as measured via CQ, 520 during low vowel /a/ is similar to other vowels. Interestingly, the CQ results did not reveal There are possible explanations for the lack of difference in CQ between vowels. Recall that 523 EGG measures vocal fold contact area and increased contact is typically a result of increased 524 laryngeal constriction or thyroarytenoid contraction. Previous studies that use CQ as a measure of 525 vocal fold tension compared very different phonation patterns, such as normal and disordered 526 voicing (Childers, 1990), hypophonic and hyperfunctional voicing (Szkiekowska,Krasnodebska,527 Miaskiewicz & Skarzỳnśki, 2018), or modal and non-modal phonation (Scherer, 1987;Kochetov, 528 2020;Herbst, 2020). Participants in the present study were judged to have normal voice quality, 529 reported no history of voice or speech disorders, and produced speech with modal phonation during 530 the study task. Therefore, it might be the case that there are small changes in vocal fold tension, but 531 they were not captured in this study because there were not significant changes in vocal fold contact 532 area that are typically seen with different types of phonation. In this case, CQ might not be 533 appropriate to measure small differences in tension with modal voicing, or the sample size needs to 534 be much larger to capture the very small effect size with modal voicing. Of course, it is also possible 535 that there were no changes in vocal fold tension that caused the F0 changes seen in the IF0 effect. If 536 the mechanism behind IF0 causes a passive stretching of the vocal folds without activation of the 537 cricothyroid or thyroarytenoid, there would be no increase in thyroarytenoid activation and 538 potentially no laryngeal constriction causing changes in vocal fold contact area. In this case, the CQ 539 results correctly captured the null effect. 540 Finally, the F0 findings may be consistent with the jaw height hypothesis. Recall that though 541 F1 is thought to reflect tongue height, Chen et al. (2019) found that jaw height was a better predictor 542 of F0 than tongue height. Therefore, differences in F0 between high, mid, and low vowels, may 543 reflect different jaw heights. For the mid vowels, tongue height might differ while jaw height is 544 similar, explaining the finding of different F1 values for mid vowels while they had similar F0 values. 545 In addition, as previously stated, tongue compression may play a role in further lowering F0 beyond 546 what is expected by jaw height alone. Therefore, IF0 may be an effect of multiple mechanisms 547 including lingual and mandibular movement. Additionally, auditory enhancement cannot be ruled 548 out as a contributing factor in the F0 results of this study. The F0s values may be a result of auditory 549 enhancement of the phonological grouping of the vowels to contrast high, mid, and low vowels in 550 Amharic. Sapir's vertical-horizontal pull hypothesis also cannot be ruled out as a contributing factor 551 as the phonological grouping of F0 might be due to differences in the articulatory goal of the 552 vowels. While no independent evidence from Amharic explains why these phonological groupings 553 emerge, findings for Dutch in Turner & Verhoeven 2011 indicate that the phonological grouping of 554 F0 is not unique to Amharic and warrants further cross-linguistic investigation. 555

V. CONCLUSION 556
The cause of intrinsic fundamental frequency of vowels has been long studied yet still 557 remains unknown. Proposed hypotheses that account for the universal phenomenon primarily relate 558 IF0 to mechanical coupling between oral vocal tract structures and the larynx, where movement of 559 the oral structures causes changes in vocal fold tension, which lead to perturbations in F0. The 560 purpose of the present study was to investigate IF0 in Amharic vowels in an effort to compare 561