Could L1 intonation patterns be applied in teaching Mandarin tones to atonal learners of Chinese? – An acoustic phonetic study

1.1 The perception of fundamental frequency and musical tones

Due to the fact that the production of both lexical tones and intonation are correlated to the modulation of f ₀, a brief musical introduction is required on the relevant terms appearing in the acoustic characterization of the analyzed MC and Hungarian tunes. Correspondingly, the following description serves as a basis of the acoustic analysis, since it sheds light on how human perception processes and normalizes inter-speaker differences of f ₀ variation. As regards human perception, the relationship between the perception of pitch variation and the actual f ₀ frequency cannot be described by a mere linear relationship. This means that the difference between two frequencies is only perceived as the same pitch interval, if the ratio between the frequencies is the same. Thus, the relationship between the perceived pitch and the corresponding frequencies is rather described based on log₂ function, as demonstrated below in Figure 1. The ratio between two musical notes is referred to as a musical interval, and an octave represents a distance of eight musical notes (see Figure 1). In this example, the Aʹ musical note is chosen as a reference point, with a frequency value of 440 Hz (a conventional reference point commonly used in instrument tuning) (Bolla 1995; Kesztler 1952). The frequency of the musical note Aʺ, (which is located an octave higher compared to Aʹ), is characterized by 880 Hz, while the A note (positioned one octave lower relative to Aʹ) features 220 Hz (Figure 1). In sum, if we compute f ₀ values of A notes, beginning from 220 Hz, we can observe a pattern that the neighboring octave is always obtained by raising 2 to the power of x. This means that the 440 Hz (Aʹ) is twice (2¹) the A’s 220 Hz. Correspondingly, the next A note positioned at an octave higher (i.e., Aʺ) is four times (2²) 220 Hz (equal to 880 Hz), and the subsequent A note located even higher (i.e., A‴) features a frequency value of 1,760 Hz, which is eight times (2³) the original 220 Hz.

Figure 1:

Frequency values of A notes and their positions on the octave scale (based on Bolla 1995; Kesztler 1952).

Based on this example, it can be concluded that comparing absolute frequencies is not suitable for contrasting speech produced in different frequency ranges by different individuals. For instance, a 50 Hz frequency difference in a male speaker’s production, (e.g., f ₀ rising from 100 to 150 Hz), can be equivalent in perception to an f ₀ change of 90 Hz in a female speaker’s production (in which case f ₀ rising from 180 to 270 Hz). However the frequency ratio in both cases is 1:1.5 (Hart et al. 1990: 24), thus, based on the same ratio of the two f ₀ changes, the two tunes would be perceived as having the same magnitude. Therefore, f ₀ values are often converted into logarithmic units, such as semitones, to normalize f ₀ changes occurring in different frequency ranges (Nolan 2003). The difference of one semitone corresponds to one-twelfth of an octave, meaning an octave can be divided into 12 equally spaced semitones (Bolla 1995). In musical theory, the musical interval of one semitone (in perceived pitch difference) is referred to as a minor second (e.g., between E-F and B-C), while a major second equals to two semitones (which surfaces between all other adjacent musical notes, except for E-F and B-C). In the case of thirds, a minor third represents a three-semitone difference, while a major third spans four semitones. In later parts of this study, other musical intervals may arise, such as fourths (representing five musical notes), fifths (seven musical notes), and sixths (eight or nine musical notes) (Kesztler 1952).

Furthermore it is essential to note that apart from the f ₀ range of the tonal contour, the shape of the f ₀ curve is also an important factor in speech perception as well as production. For instance, perceptual experiments have shown that in American English the shape of the intonation curve and the timing of f ₀ variation play a significant role in distinguishing different intonation melodies (Barnes et al. 2010). Thus, it can be concluded that the f ₀ contour’s shape is also a significant acoustic feature, and so will be analyzed in this experiment. However, prior to the acoustic analysis, the analyzed tunes’ acoustic characteristics are presented in the next section.

1.2 The analyzed tunes and the motivation of the study

The present acoustic analysis is focusing on the comparison of MC tones and the corresponding Hungarian monosyllabic intonation contours, that is, MC rising Tone 2 is compared to two Hungarian rising contours: i) monosyllabic yes/no interrogative contour and ii) the rising phase of the f ₀ pattern in alternative questions. MC falling Tone 4 is likewise compared to Hungarian i) declarative and ii) imperative f ₀ curves. Thus, in both cases, one MC tone was compared to two Hungarian monosyllabic intonation contours, resulting in six different f ₀ patterns being analyzed.

The motivation for this study arises from the fact that Hungarian teachers of MC often presume that the analyzed rising and falling Hungarian and Chinese tunes have an identical acoustic structure, thus instructing Hungarian MC learners to substitute MC L2 tonal patterns with the monosyllabic realization of a Hungarian L1 intonation contour. In particular, among Chinese teachers, it is a popular technique to regard the MC Tone 2 and the Hungarian monosyllabic yes/no interrogative intonation contour identical. Likewise the MC Tone 4 is considered equal to the Hungarian imperative intonation contour. However, the application of these teaching strategies is merely based on impressions, since up to this point there is no previous empirical analysis on the acoustic similarity of the above-mentioned MC and Hungarian tunes.

The validity of the above-mentioned analogies and teaching methodology must also be questioned from the viewpoint of L2 learning: if the acquisition of L2 segmental and suprasegmental levels are concerned, L1 influence is fundamentally present and assumed to bias both perception and production in L2, since the two are presumed to be intertwined, or to develop simultaneously (Flege and Bohn 2021). The perceptional (and therefore also the productional) bias exerted by the L1 often induces perceptual assimilation of the non-native L2 pattern to its closest L1 category. In this particular case (according to the Perceptual Assimilation Model for suprasegmentals – PAM-s) this means that native speakers of a non-tonal language are hypothesized to assimilate the foreign (L2) lexical tones into their native prosodic (i.e., intonational) categories (So and Best 2014). In particular, similar L2 lexical tones and L1 monosyllabic intonation patterns are expected to be perceived as identical, even though there may be observable and apparent dissimilarities between the two realizations. This means that these acoustic differences are expected to get filtered out from the perception of the L2 learner as perceptual assimilation takes place (labelled as ‘equivalence classification’ by Flege (1995) and Flege and Bohn (2021)), since the acoustic difference at hand is not considered contrastive in the L1. Therefore, to summarize, the investigation of acoustic similarities of MC lexical tones and Hungarian monosyllabic intonation patterns could possibly lay the foundation for discovering both teaching and learning difficulties among Hungarian teachers and learners of MC. To be more precise, a question arises regarding what kind of influence could be expected from the usage of L1 patterns in MC speech, if the hypothesis outlined above on perceptual assimilation surfaces in the case of Hungarian speakers of MC. This hypothesized perceptual bias could appear at two different levels of MC acquisition and teaching. Firstly, the above-mentioned analogies of MC lexical tones and Hungarian intonation patterns might be induced by a simple case of perceptual assimilation where teachers themselves are deceived by their own L1 (Hungarian) perception, identifying two similar, albeit acoustically distinct, realizations as identical, and as a result, encouraging students to apply the analogies between the two languages, which may then lead to foreign-accented speech among L2 learners. Secondly, even if L2 learners receive proper guidance on producing MC tones, their own perception may happen to get biased by the L1 patterns, so similar outcomes are expected in this case, too, leading to foreign accented speech.

Consequently, if the above-mentioned analogies between MC tones and Hungarian intonation patterns turn out to be false and misleading – and also if L1 effect is presupposed in the case of L2 learners and differences between L1 and L2 patterns are filtered out from learners’ perception (and as a result, production) – the results of the present paper could possibly predict how L1 intonation patterns influence and reshape MC tones in Hungarian learners’ production. In contrast, if the analyzed L1 intonation patterns are proven to be acoustically identical to MC lexical tones, then the substitution of L2 lexical tones by the L1 intonation patterns (i.e., the direct transfer of L1 patterns to L2 production) could be directly utilized in teaching MC tones for Hungarian speakers. This teaching methodology could provide a simple and genuine technique aiding the proper acquisition of MC tones, as opposed to previous methods, which often hold out prerequisites for previously acquired knowledge or skills, as they are mostly built upon using music as the primary tool when introducing MC tones to atonal speakers.

For instance, Chao (1948/1963) registers MC tones within a scale of 5 degrees (see Figure 4), while Yu (2007) proposes that the concept of music could help learners understand how tonal melodies change. The importance of music also surfaces in more recent papers – see the results of Han and colleagues (2019) who shed light on direct interaction between accuracy of Mandarin tone perception and musical training. In addition, teaching lexical tones often involves visual and/or hand movement techniques, however, according to Guan (2000), these techniques often do not achieve the desired effect, since they do not provide proper introduction to how these tonal curves are produced. In sum, all of the above listed examples require the perceptual skills to acoustically detect and identify dissimilarities between melodies, especially in those widely used imitation techniques where students need to learn lexical tones while barely relying on what they hear and perceive. This means that even though there are various techniques for teaching MC lexical tones, if the perceptual basis on detecting the acoustic keys of different tones is absent, then without proper acoustic feedback, the development of the native-like production of lexical tones may be delayed. As opposed to the methods listed above, the usage of L1 patterns transferred to L2 production might provide a more direct path to tonal acquisition of MC, since L1 patterns could be accessed by L2 learners directly.

To summarize the motivation of the present paper, the empirical results presented here could serve as an insight into how the application of L1 patterns influences Hungarian MC learners’ production and could confirm (or eventually disprove) the validity of the teaching techniques involving L1 intonation patterns. In other words, the results could contribute to and develop the teaching methodology concerning MC lexical tones by drawing attention to the (in)appropriate application of L1 intonation patterns. Furthermore, this analysis may serve as a basis for further acoustic comparisons in the future which could draw conclusions on how and what L1 patterns could be applied in MC teaching in languages other than Hungarian.

In the following subsections, the acoustic characteristics of the presumably similar MC and Hungarian tunes are presented. Moreover, as was mentioned above, the analysis is also complemented by two additional Hungarian tunes: a rising and a falling one – specifically, the rising melody of the first (monosyllabic) conjunct of an alternative question and the intonation contour of monosyllabic declaratives. These tunes are analyzed in order to provide more reference on the realization of MC lexical tones relative to Hungarian intonation patterns. It must be noted that this paper exclusively deals with MC tones produced in isolation, disregarding the problem of tonal realization in tone sequences, the effect of tonal coarticulation, and the influence of intonation exerted on tonal curves (Shen 1990; Xu 1997; for the interaction of tones and intonation in the production of Hungarian learners of MC, see Juhász and Bartos 2023). The reason for this decision is that the usage of L1 patterns in teaching MC tones is applied exclusively in the beginning phase of tonal acquisition, when lexical tones are most probably anchored to monosyllabic L1 patterns in the case of atonal speakers, since MC lexical tones are introduced to L2 learners in isolation.

1.2.1 The rising tunes analyzed

The two types of analyzed Hungarian rising tunes are two interrogative patterns: the monosyllabic yes/no question and the first (monosyllabic) unit of the alternative question. According to the description given by Fónagy and Magdics (1967), Hungarian monosyllabic yes/no questions do not exhibit the standard rising-falling interrogative pattern, instead, owing to the truncation of the underlying L*HL pattern (Varga 2002), the f ₀ curve only features a rising phase in which the f ₀ rises a minor third (Fónagy and Magdics 1967). Based on Markó’s (2007) measurements of spontaneous speech, the pitch change can also be equivalent to a major second, a minor third, or even a perfect fifth, and the f ₀ contour of this tune is described as a convex pattern (see Figure 2).

Figure 2:

F ₀ contour of the Hungarian monosyllabic yes/no interrogative Én? [eːn?] ‘Me?’ (Markó 2007: 64, left, and Olaszy 2002: 90, right), (in the latter case, values are normalized to the f ₀ value at the beginning of declarative statements (equals to 100 %)).

The second analyzed Hungarian rising intonation pattern is the first (monosyllabic) conjunct of an alternative question. For example, in the question Én vagy ő? [eːn vaɟ øː?] ‘ Me or her?’, the first syllable of Én ‘Me’ features a rising pattern, argued by both Olaszy (2002) and Markó (2017) (see Figure 3).

Figure 3:

F ₀ pattern of the alternative question Én vagy ő? [eːn vaɟ øː?] ‘Me or her?’ (Olaszy 2002: 95).

According to Olaszy’s (2002) measurements, we can draw the following conclusions in the comparison of these two Hungarian rising intonation contours. Relative to the monosyllabic yes/no question, the rising phase of the alternative question is initiated from a higher f ₀ register and its f ₀ range is also compressed compared to the monosyllabic yes/no question pattern. However, concerning maximal f ₀, the rising phase of alternative questions reaches a higher f ₀ compared to the monosyllabic question.

As is well-known, in Mandarin Chinese, lexical tone is an argument of a syllable and defines the syllable’s meaning on the lexical level (as far as homophonous syllables are not considered). The standard variation of Mandarin Chinese distinguishes 4 (+1) tones (see Figure 4), however in this particular analysis, only the rising Tone 2 and the falling Tone 4 are addressed because only these contours have a monosyllabic counterpart in the Hungarian intonation system.

Figure 4:

Introductory musical notes of MC tones (top row) and their schematic tone contours (middle row) (based on Chao 1948/1963: 85), along with the Chinese syllables realized with the respective tone (in Chinese characters and Pinyin transcription, and their meanings).

According to Chao’s description (1948/1963), in the case of the rising MC Tone 2, the f ₀ rises a major third, equivalent to 4 semitones, while measurements by Moore and Jongman (1997) index an approximate rising phase of about 3.5 semitones. Furthermore, MC Tone 2 is characterized by a convex f ₀ contour, similar to the Hungarian monosyllabic yes/no question intonation pattern (see Figure 5).

Figure 5:

F ₀ contours of Mandarin Chinese tones (Moore and Jongman 1997: 1872).

1.3 Research questions

1. The present study aims to empirically investigate whether the presumed similarities between the Hungarian intonation patterns and MC lexical tones (namely, the intonation pattern of Hungarian monosyllabic yes/no questions versus the rising MC Tone 2, on the one hand, and the Hungarian imperative intonation pattern versus the falling MC Tone 4, on the other) are confirmably strong, and whether their application in Mandarin Chinese L2 teaching is valid and promotes L2 learners’ tone production.

2. Furthermore, in view of the fact that both lexical tone and intonation production is generated by the same physiological process and both arising by the modulation of vocal fold vibration rate, we explore what acoustic features differentiate lexical tones and monosyllabic intonation patterns in atonal and tonal languages.

2.1 Participants

In the recording session the speech production of two speaker groups were analyzed: i) a group of Hungarian native speakers (mean age: 26.3 ± 2.81 years) and ii) a group of Mandarin Chinese native speakers (mean age: 25.8 ± 1.86 years), in each group seven female university students were recorded with 14 participants in total. The Hungarian speakers all spoke English daily and had no experience learning Mandarin Chinese or any other tonal language. The Chinese native speakers were all from Beijing and had been in Hungary for an average of 3.8 years. All seven Chinese native speakers spoke English weekly on average, and four of them reported speaking Hungarian on a daily basis. The speakers’ speech was recorded in 16-bit/44.1 kHz using an external sound card and an omnidirectional condenser head microphone in a sound attenuated room.

2.2 Recording material

The recorded material consisted of five Hungarian and five Chinese CV-structured, monosyllabic words that were read five times in a randomized order in the following tune types. For Hungarian utterances, the speakers produced four different sentence types: for rising tunes, monosyllabic yes/no questions and alternative questions were recorded, and for falling tunes, declarative and imperative monosyllabic utterances were produced. Chinese monosyllabic words featured either the MC rising Tone 2 or falling Tone 4. In sum, a total of six different tunes were examined in this experiment, encompassing four different Hungarian intonation patterns and two different MC lexical tones. Hungarian native speakers exclusively read Hungarian utterances, likewise the Chinese native speakers only pronounced the Chinese monosyllabic words. This means that (5 repetitions × 5 words × 7 speakers =) 175 tokens for each of the six tunes were recorded, that is, in sum, 1,050 realizations were analyzed in the present experiment. During the experiment, the analyzed Hungarian monosyllabic utterances were embedded into short texts to facilitate the standard production of the required intonation pattern. In the case of the Chinese words, context was not necessary, as the realization of the given MC tone was induced by the lexical meaning, thus, Chinese words were recorded in isolation and displayed both Chinese characters and pinyin transcription (for examples, see Table 1; for the entire material, see Tables A and B in the Appendix).

The Chinese utterances were recorded in isolation and in randomized order. In the acoustic analysis, the vocalic phase of monosyllabic words was analyzed, which in each case was a central vowel, either a front unrounded [øː] for Hungarian words, or a central vowel, back unrounded [ɤ] for Chinese utterances. Despite the different articulatory features of these two vowels, they are acoustically very similar: both vowels are positioned in the middle of the acoustic vowel space, since the place of articulation (in this particular case: horizontal tongue position) and lip rounding have opposing effects on the acoustic structure of vowels. The labiality of the front [øː] in Hungarian lowers the second formant frequency (F ₂) (compared to other illabial front vowels), while the illabiality of the back [ɤ] in MC raises the F ₂ (compared to other labial back vowels). In this manner both vowels are positioned in the middle of the acoustic vowel space (Juhász 2020).

Additionally, the consonant context of each vowel was controlled. Syllable nuclei were always preceded by a voiceless unaspirated obstruent onset in the interest of avoiding distinct coarticulatory effects of consonant contexts on f ₀ values. It must be noted that both MC and Hungarian are characterized by a two-way obstruent opposition: in Hungarian, voiced unaspirated (lenis) and voiceless unaspirated (fortis) segments are in contrast, while in MC, voiceless unaspirated (lenis) and voiceless aspirated (fortis) obstruents are in opposition (for detailed comparison, see Juhász 2021). To elaborate, Hungarian fortis and MC lenis obstruent categories overlap, that is, segments in both languages can be described by one single phonetic category concerning the acoustic feature of voicing: the voiceless unaspirated category. All MC and Hungarian obstruents recorded in this experiment belong to the above-mentioned voiceless unaspirated phonetic category. Note that in the descriptions of Ye and Bartos (2017), the voiceless unaspirated plosives are denoted by devoiced IPA characters, whereby denotation is also applied in the present paper (e.g., [d̥, g̥]).

It should further be noted that, due to the restricted syllable structure variability and monosyllabic nature of MC, there were limited options to find Hungarian words to stand as equivalents to the Chinese syllables in all of the key features, namely, syllable structure, meaning, and the formation of minimal pairs. Thus, we were forced to also record MC Tone 2 syllable “sé [sɤ]”, which is a well-formed sequence in terms of Chinese phonotactic rules but happens to represent an accidental gap – Tone 2 is not associated with this sound sequence in any MC lexeme. That is, this sequence does not exist in Mandarin Chinese and has no associated meaning or Chinese character. Since our options were limited, we chose to nevertheless include this syllable in the analysis, and we found that it was not problematic for MC speakers to produce it. Besides, although the recorded Hungarian dialogues might appear relatively rarely in spontaneous speech, the meaning of the recorded utterances were apparent, and so Hungarian speakers could easily produce the expected intonation contours.

2.3 Acoustic analyses

Audio recordings were labelled and analyzed using Praat software (Boersma and Weenink 2019). All analyses were performed exclusively on the vocalic phases featuring a quasi-periodic signal. This interval was also considered as the duration of the segment since this is the voiced phase carrying the tonal contour (owing to the fact that voiceless obstruent onsets are not voiced). The extracted f ₀ values were converted to semitones using the hqmisc package (Quené 2014) in R (R Core Team 2019), with a reference value of 50 Hz in all cases. The vocalic phase of the Hungarian and Chinese utterances was analyzed using four static features (i. – iv.) and a dynamic one (v.), which are listed below:

1. Maximal f ₀ (f _0max, in semitones).

2. Minimal f ₀ (f _0min, in semitones).

3. F ₀ range (which was calculated as the difference between maximal and minimal f ₀, in semitones).

4. The duration of the vocalic phase.

5. The dynamic analysis of the f ₀ contours (in order to obtain f ₀ contours, f ₀ was extracted by each 5 ms interval automatically).

In the statistical analysis of static acoustic features, linear mixed models were applied in R (R Core Team 2019) using the lme4 package (Bates et al. 2015). Satterthwaite approximation was used to calculate p- and F-values, which is available in the lmerTest package (Kuznetsova et al. 2017). The levels of variables were compared pairwise using post-hoc tests with the emmeans package (Lenth 2020). The random variable was defined as the speaker in all mixed models (random intercept). Rising and falling tunes were analyzed separately, meaning that altogether, eight models were composed for the four static dependent variables, namely maximal f ₀ , minimal f ₀ , f ₀ range, and duration. In these models, the independent variable was always the tune type, which had three levels for rising tunes (MC Tone 2, Hungarian monosyllabic yes/no question, and Hungarian alternative question) and three levels for falling tunes (MC Tone 4, the Hungarian imperative, and the Hungarian declarative question). The post hoc test was carried out by using the emmeans package (Lenth 2020), computing pairwise post-hoc comparisons adjusted to the Tukey test. The graphs were created using the ggplot2 package in R (Wickham 2016).

For dynamic analyses, generalized additive models (GAMMs) were applied for rising and falling tunes, respectively. GAMM is a statistic modelling used for analyzing non-linear data and calculates an estimated curve using combinations of basic functions and associated weights. In the basic GAMM models, the effect of the normalized duration was examined on the dependent variable f ₀ (obtained every 5 ms and converted into semitones), that is, to observe how the f ₀ changes dependent on the normalized duration of the vocalic phase. The GAMMs were further extended by the inclusion of a three-level ordered factor variable of tune type (as a parametric term), defining the reference curve and the difference curves measured from this reference. In the models of both rising and falling tunes, the reference curve was set to the MC Tone, and difference curves were computed for the Hungarian tunes. Random smooth function was applied by each f ₀ trajectory. The results of GAMMs were calculated in R using the mgcv package (Wood 2017) and the estimated curves were visualized by using the itsadug package (van Rij et al. 2020).

3.1 Rising tunes

3.1.1 Static analyses of rising tunes

The tune type independent variable significantly influenced the maximal and minimal f ₀, as well as f ₀ range of rising tunes (f _0max: F (2, 25) = 113, p < 0.001; f _0min: F (2, 25) = 163, p < 0.001; f ₀ range: F (2, 25) = 426, p < 0.001). Based on pairwise post-hoc tests, the Hungarian monosyllabic yes/no question had a significantly higher f _0max mean compared to the alternative question (p < 0.001) and the Chinese Tone 2 (p < 0.05). In terms of minimal f ₀, the Hungarian alternative question featured a significantly higher f _0min mean compared to the MC Tone 2 (p < 0.05) and the Hungarian monosyllabic yes/no question (p < 0.001) (see Figure 6, left). As regards f ₀ range, MC Tone 2 was realized with a more compressed f ₀ range compared to the Hungarian monosyllabic yes/no questions (p < 0.01), but featured a wider mean range compared to the alternative questions (p < 0.05). In the rising phase of MC, Tone 2 f ₀ rose an average of 5.5 semitones (approximately a perfect fourth), while the Hungarian alternative question featured a rise of three semitones (approximately a minor third), and the f ₀ of Hungarian monosyllabic yes/no questions increased by 8.9 semitones (a sixth) (see Figure 6, right).

Figure 6:

Minimal and maximal f ₀ (mean and SD, in semitones) of the three analyzed rising tunes (MC Tone 2 and the Hungarian monosyllabic and alternative questions) on the left, and f ₀ range of the utterances (in semitones) on the right.

3.1.2 Dynamic analysis of rising tunes

In the basic GAMM model comparing rising tunes, the f ₀ change was analyzed throughout the normalized duration of the vocalic section of the monosyllabic utterances. The model was complemented by the ordered three-level factor variable tune type (a parametric term), which allowed us to obtain a better fit than the basic model (χ ² (8.00) = 2,632, p < 0.001). The model was further complemented by a random smooth function applied by each f ₀ trajectory. As a result of incorporating the parametric term and random smooth functions, the model could explain 99.6 % of the variance in the data. The estimated curve of the MC Tone 2 was characterized by a concave f ₀ curve (edf = 8.8; p < 0.001), featuring its lowest f ₀ values positioned at the 20–40 % interval of normalized duration of the vocalic phase (see Figure 7). The f ₀ curve of Hungarian monosyllabic questions was characterized by a significantly steeper rising pattern (edf = 8.6; p < 0.001) compared to the MC Tone 2, featuring an intersection of both the MC Tone 2 and the Hungarian alternative question (edf = 8.3; p < 0.001) (see Figure 7). MC Tone 2 featured exclusively one single negative excursion and showed a rising pattern beginning at approximately 40 % of the normalized duration. In contrast, the shape of both Hungarian question contours was realized with a more linear pattern containing two excursions: one negative excursion at 20 % and one positive excursion at 90 % of the normalized duration. Thus, in both Hungarian interrogative contours, the rising phase starts from 20 % of the normalized duration, as opposed to MC Tone 2, where it starts at approximately 40 %. This means that compared to the two Hungarian question intonation patterns, where the rising phase dominates 80 % of the vocalic section, the MC Tone 2 approximates a falling-rising pattern. The intervals of normalized duration in which the three f ₀ curves significantly differ are shown in Table 2.

Figure 7:

The f ₀ curves of the analyzed rising tunes (red solid line = MC Tone 2, blue dashed line = Hungarian monosyllabic yes/no question, green dotted line = Hungarian alternative question).

Table 2:

The significantly different intervals of normalized duration of rising tones.

MC Tone 2 – Hungarian monosyllabic yes/no question	0–36 %	58–100 %
Tone 2 – Hungarian alternative question	0–78 %	80–100 %
Hungarian monosyllabic yes/no question – Hungarian alternative question	0–89 %

3.2 Falling tunes

3.2.1 Static analyses of falling tunes

As regards falling tunes, maximal f ₀ and minimal f ₀, as well as the f ₀ range dependent variables were all influenced significantly by the tune type independent variable (f _0max F (2, 25) = 188, p < 0.001; f _0min F (2, 25) = 22, p < 0.0001; f ₀ range: F (2, 25) = 34, p < 0.001). Post-hoc tests regarding the maximal f ₀ did not show differences between MC Tone 4 and the Hungarian imperative intonation, but the Hungarian declarative tune was characterized by a significantly lower maximal f ₀ mean compared to the other two tunes (p < 0.001, see Figure 8, left). Concerning minimal f ₀, the Hungarian imperative tune featured a significantly higher mean compared to the Hungarian declarative pattern. On the other hand, MC Tone 4 did not differ from either of the Hungarian falling tunes, which can be attributed to the relatively high standard deviation in the data (see Figure 8). In terms of f ₀ range, the Hungarian declarative tune surfaced with a significantly narrower, compressed f ₀ range compared to MC Tone 4 (p < 0.05) and the Hungarian imperative tune (p < 0.001). It is important to note that MC Tone 4 featured a higher average pitch range, as a tendency, compared to the Hungarian imperative tune, and the data for the Chinese Tone 4 also showed more variability, possibly explaining the lack of significant difference between MC Tone 4 and the Hungarian declarative tune. F ₀ range in MC Tone 4 descended by approximately 9.2 semitones (equivalent to a major sixth), while the Hungarian imperative tune featured a falling phase of approximately 8.3 semitones (corresponding to a minor sixth), and the Hungarian declarative tune descended approximately 5.9 semitones (a fifth, see Figure 8, right).

Figure 8:

Minimal and maximal f ₀ (mean and SD, in semitones) of the three analyzed falling tunes (MC Tone 4 and the Hungarian imperative and declarative tunes) on the left, and f ₀ range of the utterances (in semitones) on the right.

3.2.2 Dynamic analyses of falling tunes

In the dynamic analysis of falling f ₀ contours, the basic GAMM model (examining the f ₀ change within the normalized duration) was extended with the parametric term of tune type (which was an ordered factor variable) to achieve better fitting to the data (χ ² (8.00) = 7,789, p < 0.001). Similarly to rising tunes, a random smooth function was applied by each f ₀ trajectory. In this manner, the final model explained 99.7 % of the variability found in the data. The results of the GAMM showed that the Hungarian imperative f ₀ contour differed significantly from the MC Tone 4’s curvature (edf = 6.3; p < 0.001). Similarly, the f ₀ pattern of the Hungarian declarative tune also featured significantly different realization compared to the MC Tone 4 (edf = 8.1; p < 0.001). These differences are also apparent in Figure 9, where curvatures of the three falling tunes are all well separated from each other (see also Table 3). MC Tone 4 was realized with a slightly convex f ₀ curve, initiated from a relatively high f ₀ level. In contrast, the Hungarian declarative f ₀ curve featured a concave shape, positioned to a relatively lower f ₀ range than the other two curves. The f ₀ pattern of Hungarian imperatives was located between the other two descending patterns, approximating a quasi-linear pattern.

Figure 9:

The f ₀ curves of the analyzed falling tunes (red solid line = MC Tone 4, blue dashed line = Hungarian imperative tune, green dotted line = Hungarian declarative pattern).

Table 3:

The significantly different intervals of normalized duration of falling tones.

Tone 4 – Imperative	0–100 %
Tone 4 – Declarative	0–100 %
Imperative – Declarative	0–100 %

3.3 The duration of rising and falling tunes

Tune types also showed significant difference across their duration (rising tunes: F (2, 25) = 259, p < 0.001), falling tunes: F (2, 25) = 17, p < 0.001)). Among rising tunes, all three tune types differed significantly (p < 0.001), with the MC Tone 2 featuring the longest mean duration and the alternative questions being the shortest. Consequently, Hungarian monosyllabic questions had an intermediate duration compared to the two previously mentioned rising tunes. As for the analyzed falling tunes, only the realization of Hungarian imperative and declarative tunes differed significantly (p < 0.001), with the imperative utterance featuring a longer mean duration compared to the declarative realization. This also means that no significant difference was found in terms of duration between the MC Tone 4 and the two Hungarian falling intonation contours, presumably because the average duration of the MC Tone 4 fell between the mean durations of the two Hungarian falling tunes (see Figure 10). Furthermore, we should note that considering only mean durations, MC Tone 2 has a strikingly long realization compared to MC Tone 4 as well as to each of the Hungarian tunes, yielding a length of approximately 200 ms.

Figure 10:

The duration (mean and SD) of the vocalic phases of the analyzed rising and falling tunes (dark grey bar plots = rising tunes, light grey bar plots = falling tunes).