Phrasing – boundaries and internal declination

Firstly, the sentence has four divisions as seen by the breaks in the pitch lines (marked with the number 1). These are, in this case, good indications of the way in which the sentence was phrased. It is important to note, however, that not all phrases are separated by a pause. In many cases the analyst has to look for other subtle signals, including pitch discontinuity and changes of loudness and tempo (“final lengthening”) and increased vocal fry (“creak”) to find acoustic evidence of a perceived boundary.

There is a second element of intonation present in this sentence, and that is the tendency of voice pitch to start high in a tone group and move lower as the speaker moves through the tone unit. This is known as declination and is clearest in the second tone group, which starts relatively high and ends relatively low in pitch. Related to this is the noticeable reset in pitch at the beginning of the next tone group (marked with 2). The only phrase to reverse this is the final phrase, which is a tag question. Tag questions can be realized with a fall or a rise, depending on their function. The rising contour here suggests that it is closer to a real question than simply a request for confirmation.

Prominence

The next linguistic use of pitch in English is the marking of certain syllables as prominent. In the AM system, these prominent syllables are marked as having starred pitch accents. Pitch accents (or peak accent; see Grice 2006) are, at their most basic, marked with either High pitch (H*) or Low pitch (L*).

This corresponds to other terminology including (in the British system): tonic (Halliday 1967), nuclear stress (Jenkins 2000), sentence stress (Schmerling 1976), primary phrase stress (Hahn 2004), focus (Levis and Grant 2003), prominence (Celce-Murcia, Brinton, and Goodwin (2010), highlighting (Kenworthy 1987), and selection (Brazil 1995), among others. The perception of prominence is triggered primarily (in English) by a pitch excursion, upwards or occasionally downwards. Again, pitch works together with other phonetic features in English to signal prominence, especially syllable lengthening and fuller articulation of individual segmentals (vowels and consonants), but pitch plays a central role in marking these syllables. The pitch excursions are often visible in the F0 contour – in Figure 8.1 they are aligned with the accented syllables; I’d, Jim, Jane and the second syllable of concerned all have H* pitch accents and are marked with’ while do has a L* pitch accent – the beginning of a rising contour that reflects the questioning function of do they.

In the AM system, all pitch accents are of equal status, but the British system of nuclear tones reserves special significance for the last pitch accent in a phrase (or tone group). In the sentence in Figure 8.1, the second tone unit has two pitch accents, Jim and Jane. Jane would carry the nuclear accent, while the accent on Jim would be considered part of the Head.

The nuclear syllable is associated with the nuclear tone or pitch contour, described in holistic terms as a fall, a rise, or fall-rise, for example. The tone extends across any subsequent unstressed syllables up to the end of the tone group. In Figure 8.1 the nuclear fall beginning on I’d extends over be worried. This nuclear tone drops from the H* to an L pitch and then rises to the end of the tone group, with a final H% (the % means the final pitch level of a tone group). This is the kind of intonation that, when spoken phrase finally, “has a ‘but’ about it” (Cruttenden 1997; Halliday 1967: 141). That beginning on concerned extends over about it. Less easy to determine is the contour beginning on Jane: it could be a falling tone that flattens out over you know or, if the slight rise visible at the end of the phrase is audible, it could be another falling-rising contour (H*LH%, as marked) or a falling contour (H*LL%) extending across the three syllables Jane you know. Alternatively, the tone group can be seen as two tone groups, the first, but neither Jim or Jane followed by a separate phrase you know, particularly if the final item such as a discourse marker is separated by a slight pause – not the case here. In this analysis, you know would be an anomalous phrase with no nucleus and spoken at a low, level pitch with a slight rise or with a fairly flat contour. This kind of parenthetical intonation patttern was discussed by Bing (1980) and others (e.g., Dehé and Kavalova, 2007).

English also has other pitch accents that are characterized by the way that the prominent syllable aligns with the pitch accent. In the currently most fully developed system for transcribing intonation, the ToBI system, based on the work of Pierrehumbert (1980) and Beckman and Pierrhumbert (1986), English pitch accents can also be described as L+H*, L*+H, and H+!H*. These somewhat intimidating diacritics simply mean that the pitch accent is not perfectly aligned with the stressed syllable that is accented. In H* and L*, the vowel that is accented is aligned with the peak or lowest point of the pitch accent. This misalignment of pitch accent and stressed syllables is linguistically meaningful in English (see Ladd 2008; Pierrehumbert 1980; Pierrehumbert and Hirschberg 1992 for more information). Figure 8.2 shows the difference between the H* and L*+H pitch accent, which starts low on the word I but continues to a high pitch on the same syllable.

Discourse meaning

The sentence in Figure 8.1 has other features that are important in understanding English intonation. The first pitch accent, on I’d, is noticeably higher than the other pitch accents. This may be because it is first and because pitch declines across an utterance unless it is fully reset (Pierrehumbert 1980). An extra-high reset may also be connected to topic shifts (Levis and Pickering 2004; Wichmann 2000). In this case the expanded pitch range is most likely to be the result of contrastive stress – emphasizing I’d, presumably in contrast to someone else who is worried. The default position for a nucleus is the last lexical item of a tone group, and this means that the neutral pronunciation of I’d be worried would be to have the accent on worried. Here, however, the accent has been shifted back to I’d and worried has been de-accented. This has the effect of signaling that “being worried” is already given in that context and that the focus is on I’d. This is a contrastive use of accent and a de-accenting of given or shared information. The contrastive stress may also contain an affective element – emphasis is hard to separate from additional emotional engagement. However, we know that perceptions of emotion are not a function of intonation alone (Levis 1999).

In summary, linguistic uses of intonation in English include:

1. The use of pitch helping to mark juncture between phrases.
2. Pitch accents marking syllables as informationally important.
3. De-accenting syllables following the final pitch accent. This marks information as informationally unimportant.
4. Final pitch movement at the ends of phrases providing general meanings of openness or closedness of content of speech. These include pitch movement at the ends of intermediate and final phrases in an utterance.
5. Extremes of pitch range marking topic shifts or parenthetical information.

Applications in applied linguistics

Acoustic analysis has been used to examine in fine phonetic detail some of the prosodic differences between languages, with important implications for both clinical studies and also the study of second language acquisition (e.g., Mennen 2007). An example of cross-linguistic comparison is the work on pitch range. It is commonly claimed that languages differ in the degree to which their speakers exploit pitch range and that these differences are cultural (e.g., Van Bezooijen 1995), in some cases giving rise to national stereotypes. However, such observations are often drawn from studies whose ways of measuring are not necessarily comparable. Mennen, Schaeffler, and Docherty (2012) examined the pitch range differences between German and English using a variety of measures and found that global measures of the F0 range, as used in many other studies, were less significant than measures based on linguistically motivated points in the contour. The claim, therefore, is not just a reflection of cultural differences but that “f0 range is influenced by the phonological and/or phonetic conventions of the language being spoken” (2012: 2258). Such findings have important implications for L2 acquisition: while some learners may indeed resist an F0 range if it does not accord with their cultural identity, the results of this study show that such cross-language differences in F0 range may also arise from difficulty in acquiring the intonational structure of the other language (2012). This may also be the case for disordered speech, where pitch range has been thought to be symptomatic of certain conditions. Here, too, it may be a case of inadequate mastery of the intonation system rather than its phonetic implementation.

Another area of study made possible by instrumental techniques is the close analysis of the timing of pitch contours in relation to segmental material. Subtle differences in F0 alignment have been found to characterize cross-linguistic prosodic differences. Modern Greek and Dutch, for example, have similar rising (pre-nuclear) contours but they are timed differently in relation to the segmental material (Mennen 2004). These differences are not always acquired by non-native speakers and contribute, along with segmental differences, to the perception of a foreign accent. Variation in alignment can also be discourse-related. Topic-initial high pitch peaks often occur later in the accented syllable – even to the extent of occurring beyond the vowel segment (Wichmann, House and Rietveld 2000), and our perception of topic shift in a spoken narrative may therefore be influenced not only by pitch height but also by fine differences in peak timing.

Experimental methods are now widely used to examine the interface between phonology and phonetic realization. This includes studies of timing and alignment, as described above, studies to establish the discreteness of prosodic categories underlying the natural variation in production, and also investigations into the phonetic correlates of perceived prominence in various languages. Such experiments sometimes use synthesized stimuli, whose variation is experimentally controlled, and sometimes they rely on specially chosen sentences being read aloud. Laboratory phonology, as it is called, has been seen as an attempt at rapprochement between phonologists who deal only with symbolic representation with no reference to the physics of speech, and the engineers who use signal- processing techniques with no reference to linguistic categories or functions (Kohler 2006). There are many, however, who criticize the frequent use of isolated, noncontextualized sentences, with disregard for “functions, contextualisation and semantic as well as pragmatic plausibility” (Kohler 2006: 124). Those who study prosody from a conversation analysis perspective have been particularly critical of these methods, rejecting both the carefully designed but unnatural stimuli and also the post hoc perceptions of listeners. The only valid evidence for prosodic function in their view is the behavior of the participants themselves. The experimental approach therefore, however carefully designed and however expertly the results are analysed, has from their perspective little or nothing to say about human interaction.

Attitudinal meaning

We know intuitively that intonation can convey emotions and attitudes, but what is more difficult is to ascertain how this is achieved. We should first distinguish between attitude and emotion: there have been many studies of the effect of emotion on the voice, but the accurate recognition of discrete emotions on the basis of voice alone is unreliable. According to Pittham and Scherer (1993), anger and sadness have the highest rate of identification, while others, e.g., fear, happiness, and disgust, are far less easily recognized. If individual emotions are hard to identify, it suggests that they do not have consistent effects on the voice. However, it does seem to be possible to identify certain dimensions of emotion in the voice, namely whether it is active or passive, or whether it is positive or negative (Cowie et al. 2000). As with emotions, there are similar difficulties with identifying attitudes. There is a plethora of attitudinal labels – all familiar from dialogue in fiction (Yes, he said grumpily; No, she said rather condescendingly…, etc.). Yet experiments (e.g. Crystal 1969), show that listeners fail to ascribe labels to speech samples with more than a minimum of agreement. O’Connor and Arnold (1973) and Pike (1945) attempted for pedagogical reasons to identify the attitudinal meanings of specific melodic patterns, but one only has to change the words of the sample utterance to evoke an entirely different meaning. This suggests that the meaning, however intuitively plausible, does not lie in the melodic pattern itself. According to Ladd (1996), the elements of intonation have very general, fairly abstract meaning, but these meanings are “part of a system with a rich interpretative pragmatics, which gives rise to very specific and often quite vivid nuances in specific contexts” (1996: 39–40). In other words, we need to look to pragmatics, and the inferential process, to explain many of the attitudes that listeners perceive in someone’s “tone of voice”.

Pragmatic meaning

In order to explain some of these pragmatic effects we first need an idea of general, abstract meanings, which, in certain contexts, are capable of generating prosodic implicatures. The most pervasive is the meaning ascribed to final pitch contours: it has been suggested, for example, that a rising tone (L*H) indicates openness (e.g. Cruttenden 1986) or nonfinality (e.g. Wichmann 2000), while falling contours (H*L) indicate closure, or finality. This accounts for the fact that statements generally end low, questions often end high, and also that nonfinal tone groups in a longer utterance also frequently end high, signaling that there is more to come. We see the contribution of the final contour in English question tags; they either assume confirmation, with a falling tone as in: You’ve eaten, haven’t you, or seek information, with a rising tone, as in: You’ve eaten, /haven’t you? The “open”–“closed” distinction also operates in the context of speech acts such as requests: in a corpus-based study of please-requests, of the type ‘Can/could you … please’ (Wichmann 2004), some requests were found to end in a rise, i.e., with a high terminal, and some in a fall, i.e., with a low terminal. The low terminal occurred in contexts where the addressee has no option but to comply and was closer to a polite command, while the rising version, ending high, sounded more tentative or “open” and was closer to a question (consistent with the interrogative form). The appropriateness of each type depends, of course, on the power relationship between the speaker and hearer. If, for example, the speaker’s assumptions were not shared by the addressee, a “command”, however polite, would not be well received and would lead the hearer to infer a negative “attitude”.

The low/high distinction has been said to have an ethological basis – derived from animal signaling where a high pitch is “small” and a low pitch is “big”, and, by extension, powerless and vulnerable or powerful and assertive respectively. This is the basis of the Frequency Code proposed by Ohala (1994) and extended by Gussenhoven (2004), who suggests that these associations have become grammaticalized into the rising contours of questions and the falling contours statements, but also underlie the more general association of low with “authoritative” and high with “unassertive”.

Another contour that has pragmatic potential is the fall-rise, common in British English. It is frequently exploited for pragmatic purposes to imply some kind of reservation, and is referred to by Wells (2006: 27–32) as the “implicational fall-rise”. In the following exchanges there is an unspoken “but” in each reply, which leaves the hearer to infer an unspoken reservation:

Information structure

It is not only the choice of tonal contour – i.e., whether rising, falling, or falling-rising – but also its location that conveys important information. The placement of prominence in English conveys the information structure of an utterance, in other words how the information it contains is packaged by the speaker in relation to what the hearer already knows. Nuclear prominence is used to focus on what is new in the utterance and the default position is on the last lexical word of a phrase, or more strictly on the stressed syllable of that word. This default placement is the background against which speakers can use prominence strategically to shift focus from one part of an utterance to another. In most varieties of English, the degree of prominence relates to the degree of salience to be given to the word in which it occurs. If the final lexical item is not given prominence it is being treated as given information or common ground that is already accessible to the hearer, and the new information is signaled by prominence elsewhere in the phrase or utterance. In this way, the hearer can be pointed to different foci, often implying some kind of contrast. In the following exchange, the item “money” is being treated as given, but the word “lend” (probably with an implicational fall-rise) sets up an implied contrast with “give”:

This technique of indicating what is assumed to be given information or common ground is, like other aspects of intonation, a rich source of pragmatic inference.

Grammatical meaning

A further source of intonational meaning is the phrasing or grouping of speech units through the placement of intonation boundaries (IPs, tone-group boundaries). Phrasing indicates a degree of relatedness between the component parts, whether in terms of grammar, e.g., phrase structure, or mental representations (Chafe 1994). The syntax-intonation mapping is less transparent in spontaneous speech, but when written text is read aloud, phrase boundaries tend to coincide with grammatical boundaries, and the way in which young children read aloud gives us some insight into their processing of grammatical structures: when word by word reading (she – was – sitting – in – the – garden) changes to phrase by phrase (she was sitting – in the garden) we know that the reader has understood how words group to become phrases. Phrasing can in some cases have a disambiguating function, as in the difference between ||He washed and fed the dog|| and ||He washed | and fed the dog|| (Gut 2009). However, such ambiguities arise rarely – they are often cited as examples of the grammatical function of intonation, but in practice it is usually context that disambiguates and the role of intonation is minimal. One important point to note is that pauses and phrase boundaries do not necessarily co-occur. In scripted speech, there is a high probability that any pause is likely to co-occur with a boundary, but not that each boundary will be marked by a pause. In spontaneous speech, pauses are an unreliable indicator of phrasing, since they are performance-related.⁴

Discourse meaning

Texts do not consist of a series of unrelated utterances: they are linked in a variety of ways to create a larger, coherent whole. Macrostructures are signaled inter alia by the use of conjunctions, sentence adverbials, and discourse markers. There are also many typographical features of written texts that guide the reader, including paragraphs, headings, punctuation, capitalization, and font changes, all of which provide visual information that is absent when listening to a text read aloud. In the absence of visual information, readers have to signal text structures prosodically, and to do this they often exploit gradient phenomena including pitch range, tempo, and loudness. Pitch range, for example, is exploited to indicate the rhetorical relationships between successive utterances. “Beginnings”, i.e., new topics or major shifts in a narrative, often coinciding with printed paragraphs, tend to be indicated by an extra-high pitch on the first accented syllable of the new topic (see Wichmann 2000). In scripted speech this is likely to be preceded by a pause, but in spontaneous monologue there may be no intervening pause but a sudden acceleration of speech into the new topic, the so-called “rush-through” (Couper-Kuhlen and Ford 2004: 9; Local and Walker 2004). In conversation this allows a speaker to change to a new topic without losing the floor to another speaker.

If, in contrast, speakers wish to indicate a strong cohesive relationship between two successive utterances, the pitch range at the start of the second utterance is compressed so that the first accented syllable is markedly lower than expected. Expansion and compression of pitch range also play a part in signaling parenthetical sequences. Typically these are lower in pitch and slightly faster than the surrounding speech, but sometimes there is a marked expansion instead; in each case the parenthetical utterance is marked out as “different” from the main text (Dehé and Kavalova 2007). These prosodic strategies for marking macrostructures are also observable in conversational interaction, where they are combined with many more subtle phonetic signals that, in particular, enable the management of interaction in real time. This aspect of discourse prosody is the focus of much work in the CA framework (see Chapter 11 by Szczepek Reed in this volume).

The strategies described above are all related to the structure of the text itself, but there is a recent strand of research into intonational meaning that investigates how the pitch relationships across speaker turns can signal interpersonal meaning, such as degrees of rapport between speakers. This is seen as an example of a widely observed mirroring, or accommodation, between conversational participants. It occurs in many ways – in posture, gesture, accent, and in prosody. Meaning is made not by any inherent characteristics of individual utterances or turns at speaking but by the sequential patterning, in other words, how an utterance relates prosodically to that of another speaker. We know, for example, that the timing of response particles such as mhm, right, ok, etc., is important: if they are rhythmically integrated with the rhythm of the other speaker they are perceived to be supportive, while a disruption of the rhythm is a sign of disaffiliation (Müller 1996). Sequential pitch matching has similarly been found to be a sign of cooperativeness or affiliation: speakers tend to accommodate their pitch register to that of their interlocutor over the course of a conversation (Kousidis et al. 2009) and this engenders, or reflects, rapport between the speakers.⁵ This can have an important effect on interpersonal relations, as has been observed in a classroom setting (Roth and Tobin 2009): “in classes … where we observe alignment in prosody, participants report feeling a sense of solidarity …”(2009: 808).

In summary, at a very abstract level, the placement and choice of pitch accents and the way in which speech is phrased can convey grammatical and pragmatic meaning, such as speech acts, and also the information structure of an utterance. The phonetic realization of these choices, i.e., exploiting the range of highs and lows that the voice can produce, can convey discourse discontinuities, such as paragraphs or topic shifts, and, conversely, continuities in the cohesive relations between successive utterances. All these choices can also be exploited by speakers to generate pragmatic implicatures, which are often interpreted as speaker “attitudes”. Finally, the whole range of prosodic resources, including pitch, loudness, and timing are drawn on to manage conversational interaction, including ceding and taking turns, competing for turns, and holding the floor, and in the creation or expression of interpersonal rapport.