Linguist Forum
Specializations => Phonetics and Phonology => Topic started by: zaba on July 31, 2014, 08:47:19 AM
-
Have you used ToBI?
http://www.cs.columbia.edu/~agus/tobi/ (http://www.cs.columbia.edu/~agus/tobi/)
What are your experiences with it? I'm interested in how you used it and what you think about it. As for me, I can't make heads or tails of what the heck this is all abouts.
Any input would be appreciated!
-
It's a popular format for experiment based prosody research. But often many scorers are used to increase reliability.
I believe it's considered to be the best (at least current standard) impressionistic method for prosody.
-
ToBI is both a theory of prosodic structure and a framework for annotating that structure. Its theoretical commitments include, for example, that intonation should be understood in terms of high and low tone targets (rather than contours or more than two levels of tone), that prosodic phrases are strictly layered, and that there are two kinds of tones, pitch accents and boundary tones. It is a framework because more detail has to be provided to produce annotations for a particular language, such as the tone inventory of that language.
-
Thanks, guys. Does the ToBI frameowrk distinguish between e.g. different illocutionary acts or is it not used for that?
-
There's nothing in ToBI per se about speech acts. If there's some intonational regularity associated with speech acts, you could describe the regularity using ToBI labels (assuming ToBI makes the right theoretical commitments).
-
So ToBI labels describe the relative f0 (e.g. HIGH/LOW) and nothing else?
-
It's much more than a binary high/low because it uses curves. But, basically, yes.
It's roughly a graph of pitch over time, so it also gives you time data (how quickly an utterance was spoken).
-
I don't mean to be contrarian, but I don't think it's useful to think of ToBI as graph of pitch over time, or as using curves. We should distinguish between ToBI as a model of intonational structure, and ToBI as an annotation system. As a model of intonational structure, ToBI does not model how quickly an utterance is spoken; the only temporal information it cares about is the sequence of tones and break indices, and the alignment of those tones and break indices with the text. As an annotation system, ToBI will convey how quickly the utterance was spoken, because the tones are aligned with where the text was pronounced on the waveform.
ToBI models the tune as a sequence of tone targets, which are expressed as an opposition between high pitch and low pitch. So a L+H* pitch accent has a high target on the accented syllable, and is preceded by a low target. A H-H% boundary tone (e.g. for polar interrogatives) has a high target at the end of a prosodic phrase, and is preceded by a high target.
ToBI is also not about characterizing F0 traces. ToBI's focus is intonational structure, which it models as a shallow tree of prosodic phrases and a sequence of tones, called the tune. Remember the full name of ToBI: Tones and Break Indices. Each word boundary is associated with a break index which expresses the level of perceived juncture. Each level of juncture corresponds to the type of prosodic constituent that is ending. For example, in Mainstream American English (MAE) ToBI, a word that is followed by a break of 4 is at the end of a prosodic phrase, and a word that is followed by a break of 3 is at the end of an intermediate phrase.
To understand how this relates to F0, we should adopt the following ontology. F0 is a physical property of periodic signals. Pitch is a psychoacoustic percept that is largely a reflex of F0. Tone is the linguistic interpretation of pitch. F0 traces might do many things, but there will be ToBI tones only at pitch accents, which align with the primary stress-bearing syllable of a word (in English), and boundary tones, which occur at the ends of prosodic phrases. Prosodic phrasing, however, is correlated with many things besides F0, such as duration, syllable weight, and loudness. The tune is not just a coarse encoding of F0.
For a more detailed practical description, have a look at the versions of ToBI developed for different languages (http://www.ling.ohio-state.edu/~tobi/), or the labeling guidelines for English (http://www.ling.ohio-state.edu//~tobi/ame_tobi/labelling_guide_v3.pdf). For a more involved discussion of the justification for the theoretical choices, see Beckman and Pierrehumbert (1986) (http://faculty.wcas.northwestern.edu/~jbp/publications/intonation_japanese_english.pdf) or Pierrehumbert (2000) (http://faculty.wcas.northwestern.edu/~jbp/publications/tonal_elements.pdf).
-- EDIT
zaba, based on your recent questions, I would really recommend carefully reading Pierrehumbert (2000) linked above. It contrasts the ToBI view with other views, clearly states and argues for the theoretical commitments (e.g. ``there is no such thing as an intonation pattern without a prosodic context''), and provides lots of concrete examples that demonstrate and justify those commitments.
-
More accurate than my post, thanks.
(It does however contain some timing information but you're right about it being based primarily on whatever it is aligned to.)