Investigating tone and phonation interaction in languages of the
Americas (Heriberto Avelino, MPI Leipzig)
One of the most remarkable features in
the several languages of the Americas is the use of contrastive
laryngeal activity in vowels (Otomanguean, Mayan, Mixe-Zoquean. Maku,
Arawakan, Tucanoan), to signal tone and phonation categories. In
addition to contrastive pitch ranging from 2 to 8 tones, descriptive
grammatical studies of these languages use a variety of labels to
describe different phonation types, such as "rearticulated,"
"glottalized," and "aspirated," for instance. This usage,
although not intended to be phonetically technical, reflects the
impression of very skilled field linguists, in a way that strongly
suggests that the languages of the Americas might have between two or
three contrastive voice types, including modal, breathy, and creaky
voices. I will address the practical, methodological and theoretical
challenges entailed by the tone-phonation phenomena in these languages
and how the synchronic diversity can shed light on the evolution and
development of tone systems.
Documenting prosodic systems on the fly (Steven Bird, U Melbourne / U Penn)
Published linguistic descriptions may include a section that covers the prosodic system of a language. Although the notation and level of detail may vary, such discussions usually contain statements like: this language has three tones, high, low, and falling; falling tones only appear phrase finally; prefixes are toneless; tonal contrasts only appear on accented syllables; etc. Some of these statements are clearly phonological, while others are more phonetic in nature, and others are ambiguous: e.g. is falling tone just the reflex of high tone in phrase final position? The same ambiguity sometimes carries over into the transcriptions that are provided with a description: e.g. does something transcribed LLHLL have a hat or a hill contour? Viewed from the standpoint of a later analysis work, most such descriptions are simply incomplete, and a theoretical decision may hinge on a piece of information that was not answered in the description.
One option is to start over and produce a new and more detailed description of the language. However, suppose we don't have this luxury, and that instead we just have a few hours of access to a couple of speakers. What should we elicit? How should we prepare? In what form might we disseminate any new (still incomplete) materials? What should be archived? I will report on such an experience for the language Usarufa [usa], a language spoken by about 1200 people in the Eastern Highlands Province of Papua New Guinea. I will discuss the issues with the existing description, describe the new documentation that was collected, and evaluate its utility in supporting ongoing analytical work.
Finding a way into a family of tone languages: The story of the
Chatino Language Documentation Project (2003-) (Emiliana Cruz and Tony
Woodbury, University of Texas at Austin)
In 2001, Nora England
joined the linguistics department at the University of Texas and founded
there a Center for Indigenous Language of Latin America, dedicated to
offering doctoral training to speakers of Latin American languages in
linguistics and related disciplines. A year later one of us, Emiliana
Cruz, a speaker of San Juan Quiahije Eastern Chatino, joined as a
graduate student in anthropology. The other of us, Tony Woodbury, of the
UT linguistics faculty, volunteered to work with Emiliana on her
language, which we immediately realized had many, many tones (10
distinct short monosyllables in isolation; more if you consider
differences revealed by sandhi, as we now know). The next year
Emiliana’s sister, Hilaria Cruz, joined the linguistics department after
spending several weeks with us the previous summer as we attempted to
teach what (little) we had learned to groups of interested speakers in
San Juan Quiahije.
The Chatinos traditionally inhabit a small
part of the mountain area in Oaxaca, Mexico, that lies between the
valley of Oaxaca and the Pacific coast; and as long as a century ago,
Franz Boas realized that Chatino's endlessly various village varieties
could be classified into three main languages: Zenzontepec Chatino,
Tataltepec Chatino, and (as it were) all the rest, which we call Eastern
Chatino. Our immediate problem was that the only serious description of
any Chatino tone system--Jeff Rasch (2002) on Yaitepec Eastern
Chatino--looked little like what we encountered in San Juan Quiahije,
despite an otherwise obviously close relationship; and for both of our
parts we were nothing if not novices at tone description.
eight years later, there are eight of us: Emiliana, Hilaria Cruz, Jeff
Rasch, Tony, and four more graduate students. We work on all three of
the languages, and, within Eastern Chatino, on four different varieties.
Each of the six kinds of Chatino we study has a unique and
typologically surprisingly-different tone system. We believe that by now
for each one, we have an adequate representation of the lexically
contrastive tones and description (if not always an understanding) of
the phrasal sandhi processes; we have adult and adolescent trainees in
each location who are able to transcribe and teach these systems with
confidence; and we have posited the phonetic, phonological, and
morphological changes for each variety that allow us to explain the
pattern of cognate relations that gradually emerged to us among the
languages and subvarieties.
In this paper, we hope to chronicle
our path into Chatino tone systems. We will tell this story in part
individually and in part jointly, in order to reflect Emiliana’s
experience working on her own (and closely-related languages) “from the
inside” as a native speaker and trainer of native speakers; Tony’s
experience “from the outside” as an experienced field linguist but green
tonologist; and both of our experience with the practical and
intellectual dimensions of the project as it took shape. In particular
we want to describe what we think were the simple but diverse methods we
employed: we found minimal sets and like-sounding groups and subgroups
of word patterns; we had speakers create and judge groupings of word
patterns; we devised and improved sandhi test-frames and in turn
improved both our classifications and our understandings of phrasal
tonology; we "cheated" relentlessly in using the patterns of one
language to help us find cognate patterns in another; and we cheated
some more by building on our morphological knowledge and the
expectations it generated. Above all, however, we think we benefited
most, and at every turn, by the contributions of both speakers and
non-speakers in our core group; and on our emphasis on local training.
By going over the material slowly, again and again, with sharp,
inquisitive speakers, we never stopped learning ourselves.
Now you hear me now you don't: finding tone contrasts in unusual places in New Guinea (Mark Donohue, Australian National University)
Tone systems of New Guinea are poorly understood. A prime reason for this has been assumed to be the paucity of detailed description, combined with less training than might be hoped for on the part of
some researchers. Language descriptions about with comments such as "pitch is contrastive, but will not be marked on the examples that follow", or "the language speakers insisted that the words were
different, but they were in fact identical, with only the musical drift showing any differences." Recent work on Skou and related languages from north-central New Guinea provides another 'excuse' for
the state of tonal understanding in New Guinea: many languages for which tone melodies are lexically contrastive see those melodies disappear in the face of competition from melodies that are dictated
by the phonology of information structure. In Skou a number of tone melodies are found (H, L, HL, LH, and LHL), but LH and LHL may not associate on a single syllable. By contrast, examining running speech
we find that contrastive topics are always realised with a LHL contour, even if monosyllabic. Similarly (in terms of the irrelevance of lexical tone melodies), items in a list are always realised with a L contour. These observations based on a detailed analysis of Skou data have been attested in other languages as well, indicating that the investigation of tone in New Guinea faces methodological challenges not reported elsewhere.
Computing tone for theoretical and practical reasons (Dafydd Gibbon, U Bielefeld)
A computational linguistic and computational phonetic approach to the modelling of tone is proposed. This entails using formally explicit categorial and numeric data structures at phonological and phonetic levels for utterances (representations) and for generalisations over these (grammars), algorithms for processing these data structures, and operationalisation of these algorithms in software whose performance can be evaluated. This approach has both theoretical value in providing insights into details of the phonetic domains, and practical value both in providing high quality data resources and in providing a basis for socially useful speech technology applications.
The following metatheoretical assumptions are made. Phonological and phonetic approaches to tone analysis differ in their assumptions about empirical categories. Phonology takes the native speaker as a model and sharpens perceptual and analytic judgments into a metalanguage about sound patterns. Phonetics presupposes a phonological background, and adds physical models of the physiology of sound production and perception, and sound transmission. Natural phonologies and functional phonetics look for bridges between the two domains. Computational methods are essential in this bridging endeavour.
Specifically, two typologically and phylogenetically different tone languages will be described and modelled phonologically and phonetically and the models will be operationalised in a specimen speech synthesis implementation: Ibibio (Nigeria: Niger-Congo, Lower Cross) and Thadou (India: Tibeto-Burman). The speech synthesis implementation is intended to be used in perception tests, for theoretical purposes, and in practical speech technology applications. The Ibibio contribution is in part financed by the World Bank Step B project of Moses Ekpenyong and Eno-Abasi Urua, University of Uyo, Akwa Ibom, Nigeria.
Discovery Procedures for Tone and Intonation (Mark Liberman, U Pennsylvania)
This presentation will explore questions like these: Can
you tell whether an unknown language has lexical tone simply by
listening to an unanalyzed sample? Could we design an automated
procedure to do this? More generally, could there be inductive procedures for placing a
language in a typology of tone and intonation -- or perhaps even
specifying (aspects of) its prosodic grammar?
Answers require a
more precise specification of inputs and outputs. The input might be raw
audio, or pitch and amplitude contours alone, or either of these with
the possible addition of word boundaries, word identities, segment
boundaries, and phonetic types. Possible outputs include typological
distinctions, tonal inventories, estimation of relevant parameters,
specification of tonotactics, tone sandhi, and so on.
The idea of inductive discovery procedures has a long and varied
history in the study of speech and language, starting with the American
structuralists' positivist philosophy of science, flipping to emerge as
Chomsky's "Language Acquisition Device", and flipping again into the
contemporary application of machine learning to problems of ambiguity
resolution in speech and language technology.
As Jakobson famously pointed out to Chomsky, the fact of human
language acquisition shows that such inductive procedures must exist.
Given the new concepts, methods and tools now available to us, perhaps
it makes sense to return to the question as Zellig Harris posed it.
This presentation will work through a few radically simplified
version of the problem, for which data and algorithms will be made
available; and will raise for discussion the question of what resources
would be required to engage these questions in a more realistic way.
Rewards and Risks of F0 Analysis in Working on a Tone Language (Ian Maddieson, UC Berkeley)
Any linguist today can avail themselves of cheap, even free, software to produce an analysis of the fundamental frequency contour of an utterance. This makes it possible to obtain much objective data relevant to phonological and phonetic aspects of tone systems. Direction of pitch movements, size of pitch intervals, alignment of peaks and turning points and many other properties can be quantitatively studied. Both exploratory and confirmatory investigations can be carried out: the first to help develop a hypothesis, the second to test its predictions. However, there are limitations and pitfalls in the application of this technology which the user needs to understand. These start with being sensitive to how analysis settings influence the results obtained and when to have confidence in the output. Also important is to realize that a measurable difference is not necessarily a noticeable or linguistically-important difference. Careful manipulation of the signal to create perceptual tests can be employed to probe for what matters to the listener.
How tones and phonation types are transcribed in East and Southeast Asia (James A. Matisoff, UC Berkeley)
The complex tone systems of East and Southeast Asian languages have taxed the ingenuity of linguists and educators. An instructive example is the Black Lahu dialect of the Central Loloish language Lahu, where no fewer than four systems have been used to distinguish the 7 tones: Tone Equivalences in Lahu Orthographies
Catholic Baptist Chinese Matisoff mid caˍ ca ca ca high rising ca- caˉ caq cá high falling ca͆ caˇ cad ca^ low falling caˬ caˬ cal ca` very low ca caˍ cal cā high checked caˆ caˆ cat ca^ʔ low checked caꞈ caꞈ car ca`ʔ
• In both missionary orthographies, the tone-marks are placed one space after the vowel, which makes it impossible to run syllables of the same word together, and unesthetic to join syllables within a word with hyphens, so that there is no way to tell where one word ends and the next one begins.
• In my orthography the tone-marks appear above the vowel. In the case of the two checked tones, part of the tone-mark is a post-vocalic glottal stop, but this offers no impediment to hyphenization. I regard the checked tones as forming a distinct subsystem from the open ones (they derive from proto-syllables with final stops), so that no synchronic or diachronic connection is implied between the high falling and high checked tones, or between the low falling and low checked tones.
• The missionary systems use both superscript and subscript tone-marks, which makes them hard to type and easy to confuse with each other.
• The tone-marks in the Catholic system are particularly cumbersome, since several of them are not standard diacritics at all, but rather curved lines or hooks meant to suggest mnemonically the actual contours of the tones.
• The Chinese system of marking tones by arbitrary post-vocalic consonants has much to recommend it. Such transcriptions were popular in the U.S. in the 1940’s and 1950’s for languages like Thai and Burmese, and similar romanizations are now in use for Hmong and Mien. It makes it possible to write polysyllabic words with no spaces between the syllables, usually without ambiguity as to the syllable boundary. It is eminently printable, typeable, and word-processible. In its earlier version it was seriously flawed by writing the low falling and very low tones with the same letter ‑l. This has now been rectified by using ‑f for the latter, as urged in Matisoff 1984.
The transcription of phonation types (clear, breathy, creaky, etc.) poses special problems of its own.
Several indigenous systems of distinguishing tones have been developed, as in the traditional Yi syllabary of 800+ characters, where the tones are “built into” each graph.
Phonetic Cues to Phonological Tone (Martine Mazaudon, LACITO-CNRS)
I would like to speak of the particular kind of complexity that is met when studying emergent tone systems in Tibeto-Burman Himalayas. These systems do not have the complicated tone rules that we see in well entranched African type tone systems, whose origin goes back to before the Flood. In fact we have none of these, not even sandhi. The complexity we meet is in the untangling of different phonetic cues to a single phonological entity or tone, which leads to competing phonological hypotheses at the time of fieldwork (tone, stress and tone, tone and phonation, tone and length...) In the interest of good fieldwork, the notion of the “conditioning” of “redundant” features by “distinctive” features has to be clearly distinguished into the two, diachronically opposite, concepts it covers : the automatic effect deriving from coarticulation, in which “the
more of x, the more of y”, versus the phonological “conditioning” of the presence of one cue (the older one) by the presence of another cue (the newer one), in which we may have, but not necessarily have, the relation “the more of x the less of y”. Of course it is not a matter either of only a tandem of cues being involved.
On the Description and Analysis of Mixed Tone-Stress Systems (Lev Michael, UC Berkeley)
Prosodic systems that exhibit characteristics of both tone and stress systems present a number of interesting methodological and theoretical challenges. This talk examines these issues in light of three Peruvian Amazonian languages with mixed tone-stress systems, Iquito (Zaparoan), Kakataibo (Panoan), and Máíh+ki (Tukanoan). Systems of this type present at least two types of perceptually salient forms of prosodic prominence, which can lead to divergent analyses of the same systems when analyzed by different linguists, as in the cases of Iquito (compare Eastman and Eastman (1976), Sullón (2005), and Michael (in press)), Kakataibo (Shell (1950); Zariquiey and Michael (in prep.)), and Secoya (Johnson and Peeke 1962, Johnson and Levinsohn 1990), which is closely related to Máíh+ki. I discuss several methodological and analytical questions in the analysis of systems of this sort and consider how their analysis contributes to clarifying aspects of the debate over 'pitch-accent' systems.
Some techniques for the study of level-tone systems in Asian languages (Alexis Michaud, LACITO-CNRS)
Much headway is currently being made in the study of level-tone systems in Asian languages (see Evans 2008), largely thanks to researchers with a knowledge of level-tone systems of other areas such as Japan and Subsaharan Africa (in particular Hyman 2007a, b). I will report on challenges
encountered in fieldwork on the Naish branch of Sino-Tibetan (languages: Naxi, Na and Laze), and on the techniques employed to solve them. In addition to tonal analysis, simple ways to approach intonation (i.e. suprasegmental cues to phrasing and information structure) will be presented,
arguing that analyses of tones and intonation usefully complement each other.
Evans, Jonathan. 2008. "‘African’ tone in the Sinosphere." Language and Linguistics 9:3.463-490.
Hyman, Larry M. 2007a. "Elicitation as experimental phonology: Thlangtlang Lai tonology." Experimental Approaches to Phonology ed. by P. S. Beddor, M. J. Solé and M. Ohala, 7-24. Oxford: Oxford University Press.
Hyman, Larry M. 2007b. Kuki-Thaadow: an African tone system in Southeast Asia. Berkeley: UC Berkeley. UC Berkeley Phonology Lab Annual Report. 1-19.
Getting the balance right – Combining qualitative and quantitative
approaches in the study of tone and other suprasegmental phenomena (Bert Remijsen, U Edinburgh)
I have developed my expertise on tone through the study of languages
that present two analytic challenges. On the one hand, I have
investigated tone systems that have rich inventories (Matbat
[Austronesian] with a six-way tone contrast; Shilluk [Nilo-Saharan] with
seven tone categories). On the other hand, I have investigated
languages that have rich suprasegmental systems, involving one or more
suprasegmental contrasts alongside tone. In this way, Ma`ya
[Austronesian] and Papiamentu [Caribbean Creole] have distinctive stress
and tone, and Dinka and Shilluk [Nilo-Saharan] have vowel length and
voice quality distinctions alongside word-level tone. In the study of
these prosodic systems, it was not always evident from the outset which
suprasegmental patterns were to be analysed in terms of tone, and which
to other suprasegmental distinctions, such as stress, length or
Using examples from my past and current research, I will show how these
challenges can be tackled through the judicious combination of
qualitative and quantitative approaches. An important part of my paper
will relate to the division of labour between qualitative and
quantitative methods. In my experience, many students of prosody find it
difficult to get the optimal balance here, sticking to their
qualitative or quantitative comfort zone, even when a different approach
would be highly effective at the stage of analysis at hand. My
discussion will be specific, providing numerous practical
recommendations on how to make optimal use of both qualitative and
quantitative methods. I would also like to include some anecdotes, e.g.
on the discovery of tone in Matbat.
Studying tone in Athabaskan languages (Keren Rice, U Toronto)
I plan to look at tone in Athabaskan languages, beginning with some of
the challenges I encountered in my first fieldwork on an Athabaskan
language, in Dene (Slavey) as spoken in Fort Good Hope, Northwest
Territories, and figuring out tone systems in later fieldwork on other
dialects of the language. Tonal Athabaskan languages are generally
reported as having two tones with limited tone processes, so one might
expect a study of tone to be quite straightforward, but there are
numerous challenges to the beginning fieldworker. Early issues that I
encountered involved learning to distinguish the tones and understanding
the tone system of the verb; with later fieldwork on other dialects,
phrasal issues also became apaprent. I will then extend this to a more
general discussion of work on tones in Athabaskan languges, with
particular attention to work by John Alderete on Tahltan (tone does not
appear where expected, but something seems different) and Willem de
Reuse on San Carlos Apache, where mid tones are written (effect of the
orthography of one language on choices made for another closely related
On discovering contrastive tone melodies (Keith Snider, SIL International)
Most linguists are familiar with the Mende data set that so richly supports the notion that the distribution of Mende tones is better explained by assuming five underlying tone melodies that associate to root TBUs (up to three syllables in length) in a predictable manner than assuming five underlying tonemes that associate to TBUs in an, at best, perplexing manner. For many languages, however, the maximum number of syllables in roots is no more than two, so the distributional evidence for tone melodies in such languages is much less convincing than it is in Mende. Distributional phenomena, however, are only one type of evidence for the existence of tone melodies, and a lack of such evidence is mainly only an artifact of analyzing root structures that are comprised of less than three syllables.
In this paper, I will present:
- reasons why the underlying tonal make-up of roots in all tone languages should be analyzed as melodies rather than as individual tones, regardless of the number of tones in the melody or the number of syllables in the root,
- advantages that arise from analyzing tone languages from the perspective of their having tone melodies, and finally
- a methodology for analyzing tone that controls for non-tonal variables that affect the phonetic realization of underlying melodies (e.g., noun and verb class morphology, root syllable profiles, etc.). Once these variables are controlled, the surface pitch differences that emerge between forms that are the same in all other respects must be due to their having different underlying melodies.
How to study how tones are learned (Kristine Yu, UCLA)
We consider how lexical tone categories in tonal languages are represented in the speech signal in order to understand how to set up the learning problem for computational modeling of L1 lexical tone acquisition. Because infants begin as "universal citizens", in being able to distinguish (almost) all speech sound contrasts of the world's
languages, the infant learner is faced with the problem of discovering what dimensions--what parameters--in the speech signal are relevant for lexical tonal contrasts, and how lexical tone categories are distributed over these dimensions. To gain insight into the acquisitional process, it is possible to use computational modeling techniques to perform a discovery procedure for tonal categories based on clustering acoustic exemplars of tones.
In this talk, we emphasize the importance of setting up the learning problem appropriately, by considering how tonal exemplars should be represented (how they should be parameterized) to the learner, what the target of learning is, and what inductive biases may constrain the hypothesis space. We discuss these issues based on our work from: (1)
a corpus of tonal production data from cross-linguistic fieldwork (Bole, Igbo, Mandarin, Cantonese, White Hmong, to date), (2) tonal perception experiments in Cantonese, and (3) computational modeling of the tonal production data. Some methodological issues we emphasize are that: (1) work on cross-linguistic data is necessary for understanding tonal acquisition, (2) the parameterization of tonal exemplars, even if f0-based, is not simple, and (3) modeling assumptions can be grounded in cognitive reality with an interplay of psychological experimentation and computational modeling.
The duration aspect of tones
(Jiahong Yuan, U Pennsylvania)
Tones are primarily realized by fundamental frequency (F0). The duration characteristics of tones have not been paid much attention in the
literature. In this talk I will emphasize how and what we can learn about tones from the perspective of duration, by taking the example of Mandarin Chinese. My previous study found that intonation has different effects on the duration patterns of different tones, e.g., the final falling
tone is longer in question intonation than in statement intonation, whereas the
final rising tone has similar duration for the two intonation types. In this study, the effects of intonation on the duration of tones with different speaking rates, i.e., normal, fast, and slow, will be examined. The goal is to demonstrate how these tone-dependent effects may inform a full understanding of the underlying representation of tone.