Handout #5TOBITOBI• TOBI (or ToBI) stands for Tones and BreakIndices (Silverman et al. 1992).• It is a transcription system for prosody in some“similar” dialects of English: General American,Australian, and Southern English.• The purpose was to develop a common annotationsystem for large shared corpora, to be used inprosodic research and in speech technology.The Triumph of TOBI• TOBI has become the dominant frame ofreference for discussion of intonation inEnglish (Ladd 1996).• Analogous TOBI systems have beenformulated for other dialects and otherlanguages (e.g. K-TOBI for Korean: Jun2000).Components of a TOBItranscription• A TOBI transcription includes:• (1) a tone tier, consisting of intonational tonecategories• (2) an orthographic tier, consisting of anorthographic transcript• (3) a break indices tier, marking degrees ofdisjunction between adjacent items• (4) a miscellaneous tier, consisting of commentsTone tier• The inventory of tone categories are a slightlyrevised version of that in Pierrehumbert (1980).• It includes:– Pitch accents (T*), associated with prominentstressed syllables– Phrase accents (T-), associated with the end ofan intermediate phrase– Boundary tones (T%), associated with theedges of an intonational phrasePitch accents• H*: High tone aligned with the prominent syllable• L*: Low tone aligned with the prominent syllable• !H* : A downstepped high tone at a lower pitchrange than a preceding high tone in theintermediate phrase.• L*+H: L* followed by a high tone• L+H*: H* preceded by a low tone• H+!H*: !H* preceded by a high tonePhrase accents• L-: A low tone extending from just after the lastpitch accent in an intermediate phrase to the endof that phrase.• H-: A high tone extending from just after the lastpitch accent in an intermediate phrase to the endof that phrase.• !H-: A downstepped version of H-.• Intermediate phrase is the domain of downdrift inEnglish (Beckman and Pierrehumbert 1986).• It is a prosodic unit intermediate between theprosodic word and the intonational phrase.Boundary tones• L%: Low tone at the end of an intonationalphrase• H%: High tone at the end of an intonationalphrase• %H: High tone at the beginning of anutterance-initial intonational phrase.Phonetic implementation• The critical innovation of Pierrehumbert(1980) was to limit the tone categories totwo tone levels, while positing a richlanguage-specific mapping from tonecategories to f0.• She called this mapping phoneticimplementation, and she considered it partof the grammar of English.Downstep• Downstep is a rule of phonetic implementationthat reduces the f0 value of H relative to apreceding H (Pierrehumbert 1980: 146).• In H+L Hi and H L+Hi, the f0 value of Hi is ktimes the f0 value of H, where k is a constantbetween 0 and 1.• This function is explored and modelled inPierrehumbert (1979) and Liberman andPierrehumbert (1984).• The domain of this rule is the intermediate phrase(Beckman and Pierrehumbert 1986).Liberman and Pierrehumbert (1984)Liberman and Pierrehumbert (1984):Mean f0 values for one speakerLiberman and Pierrehumbert (1984)Liberman and Pierrehumbert (1984):Mean f0 values for one speakerLiberman and Pierrehumbert(1984)• Liberman and Pierrehumbert (1984) modelled thef0 downtrend quantitatively for two sets ofsentences.• The downtrend was quite regular.• They argue that it consists of:– Catathesis (i.e. downstep) - a relation betweensuccessive H tones in a phrase.– Final lowering - an extra degree of loweringfor H in the neighborhood of phrase-finalposition.Crosslinguistic work• Downtrends were originally observed inAfrican languages (e.g. Meyers 1976).• They have been explicitly modelled in anumber of languages besides English, e.g.:– Japanese (Pierrehumbert and Beckman 1988)– Igbo (Liberman et al. 1993)– Spanish (Prieto et al. 1996)– Chichewa (Myers 1996, 1999)Upstep• Pierrehumbert (1980: 146) also posited a phoneticimplementation rule of Upstep, raising boundarytones after H-.• In the sequence H- H%, this raises the f0 value ofH% to an extra-high level as high above H- as thevalue of H- is above that of L (a high-risingphrase-final pattern as in yes-no questions)• In the sequence H- L%, this rises the f0 value ofL% to the same level as that of the preceding H- (afinal high plateau).Upstep• This posited rule of phonetic implementation wasstated in categorical, not quantitative, terms.• It has never been backed up by any experimentalor quantitative work.• It does not correspond in conditioning factors or ineffect to patterns of upstep that have beenmodelled in other languages, such as Mandarin(Xu 1999) and Yoruba (Laniran and Clements2003).Tone Spread• The last rule of phonetic implementation fromPierrehumbert (1980) that is assumed in TOBI isher rule of Tone Spread (p. 220).• This rule spreads a phrase accent T- rightward tothe end of the phrase, as long as there is a tonefollowing it in the phrase with an f0 value greaterthan or equal to that of T-.• This results in level f0 sequences extending fromthe last pitch accent in the phrase up to the T%boundary tone.Tone Spread• This is another non-quantitative phoneticimplementation rule.• It has never been supported by experimental orquantitative studies.• It is analogous to tone spread in tonal phonology,but it is stated as a phonetic implementation rulebecause it depends on the f0 value.• It is quite unlike tonal coarticulation, its closestphonetic analogue, in conditioning factors andeffect (Bruce 1977, Pierrehumbert and Beckman1988, Xu 1994, Potisuk et al. 1997).The categories of Pierrehumbert(1980)• Pierrehumbert’s (1980) tone categories were builtinto the surface phonological representation.• They are thus supposed to be the psychologicalspeech categories formed by English speakers.• The phonetic implementation rules are then meantto express the speaker’s knowledge of how thesecategories relate to f0 in actual speech.• But there has been very little work testing theposited categories, with the notable exception ofPierrehumbert and Steele (1989).Pierrehumbert and Steele (1989)• Pierrehumbert and Steele (1989) tested thehypothesis that the timing distinction betweenL*+H and L+H* is categorical.• They synthesized a continuum of stimuli differingin the timing of the f0 trough and following f0 peakwith respect to the stressed syllable in Only aMILLionaire.• Listeners repeated the