Intonational Analysis and Prosodic Annotation of Greek Spoken Corpora

Abstract

This chapter provides an analysis of the prosodic and intonational structure of Greek within the autosegmental/metrical framework of intonational phonology, and presents Greek ToBI (GRToBI), a system for the annotation of Greek spoken corpora based on this analysis. Both the analysis and the annotation system have largely been developed on the basis of a corpus of spoken Greek. The analysis posits five pitch accents (H*, L*, H*+L, L*+H, L+H*), and two levels of phrasing, the intermediate phrase (ip) and the intonational phrase (IP), which are tonally demarcated by three types of phrase accent (H-, L-, !H-) and three types of boundary tone (H%, L%, !H %) respectively. Unlike the original ToBI, GRToBI has five tiers: the Tone Tier, the Words Tier, the Break Index Tier, the Miscellaneous Tier, and the Prosodic Words Tier (a phonetic transcription of prosodic words).

Analysis of inter-transcriber consistency in the Cat_ToBI prosodic labeling system

“…According to this, Catalan has six basic pitch accents H*, L+H*, L+>H*, L*, L*+H and H+L*, with the following upstepped and downstepped pitch counterparts (i.e., scaled higher or lower than the previous pitch accent): !H*, H*, L+!H*, L+¡H* and !H+L*. With respect to the use of the symbol '>', the same convention used in MAE-ToBI [3] and in Gr ToBI [9] is adopted: if the maximum F0 peak does not actually occur within the syllable nucleus, the late F0 event is marked by putting the symbol '>' before the H.…”

“…This framework has also been applied in automatic speech processing and database annotation to yield ToBI (TOnes and Break Indices) a prosodic labelling standard for speech databases that is based on Pierrehumbert's thesis [2]. ToBI-based systems have been developed to label oral databases for many languages such as English [3], Spanish [4,5,6], German [7], Japanese [8], Greek [9], Korean [10] and Catalan [11,12] among others.…”

Focus in Greek wh-questions

“…In the covert indirect questions there are only two ips, since the matrix negative clause is absent. All ips in the experiment end in a H-phrase accent which is a typical tone for a continuation rise in Greek (Arvaniti and Baltazani 2005). One noteworthy detail is that the tone used for the NPA aligned with the wh-word is a H*, whereas in direct wh-questions it is usually a L*+H.…”

Integrating Contrast in a Framework for Predicting Prosody

“…The four pragmatic conditions were compared on the basis of both phonological and phonetic criteria. In the first case, utterances were annotated for pitch accent type based on the GRToBI annotation scheme [1]. In the second case, measurements were taken of mean F0 (vowel), vowel duration and mean intensity (vowel).…”