Tonal Constituents and Meanings of Yes-No Questions in American English Nancy Hedberg, Juan M. Sosa & Lorna Fadden Department of Linguistics Simon Fraser University, Burnaby, B.C., Canada {hedberg; sosa; fadden}@sfu.ca Abstract We analyzed the different meanings associated with the tonal contours of 104 positive yes-no questions from the CallHome Corpus of American English. We take into consideration such broad constituents as the head, nucleus and tail of intonational phrases, as well as ToBI sequences of pitch accents, phrase accents and boundary tones. The meaning of a question as unmarked or marked in a variety of ways is shown to depend upon the intonational contours associated with these broad constituents, and even with the contour associated with the question as a whole. 1. Introduction We first outline our method of reliably arriving at a ToBI notation for American English yes-no questions, and its correlates in terms of higher constituents such as head, nucleus, and tail. We then discuss and exemplify the meanings we have found to correlate with these categories. Finally we consider whether it is the intonational pattern stretching across the entire question that is most relevant for characterizing basic aspects of the meaning of the question, such as whether it is marked or unmarked in particular ways. 2. Methods 2.1. Reliability Study We selected 48 positive yes-no questions for the training phase of the study from the CallHome corpus [1], [2], a corpus of 30-minute recorded telephone calls between people who know each other. The purpose of the training phase was to learn to reliably code ToBI categories [3]. Together we listened to each example and examined pitch tracks and came to a consensus on the appropriate tonal notation for each. We took notes on how decisions were made so that we could refer back to them later. The second phase involved each of the three members’ independent coding of 56 positive yes-no questions. We calculated transcriber-pair-word agreement by comparing the labels of each transcriber against the labels of every other transcriber. If all three pairs agreed, we gave the word a score of 1. If two out of three pairs agreed we gave it a score of .33. If none agreed, we gave it a score of 0. Agreement was calculated three different ways. First, we calculated whether we agreed on pitch accents and terminal contours, and we agreed in 72% of the cases. Second, we calculated the presence and type of pitch accent, and agreement was 75.7%. Third, we calculated the agreement of the presence or absence of pitch accent, and agreement was 85.3%. Our results compare about equally with previous studies reported in the literature. Using a variety of read and spontaneous speech, [4] report 68% agreement on particular pitch accent or no pitch accent, and 80.6% agreement on presence or absence of pitch accent. In a study on German ToBI, [5] report 71% agreement on particular pitch accent or no pitch accent, and 87% on presence or absence of pitch accent. [6] calculated agreement for one professional male and one professional female speaker. They reached 71% and 72% agreement for the female and male speaker, respectively, for particular pitch accent or no pitch accent. They report 92% and 91% agreement for the female and male speaker for presence or absence of pitch accent. In a study using ToBI-Lite on samples from the Switchboard Corpus [7], [8] report agreement of 85.6% for particular pitch accent or no pitch accent, and 86.6% for presence or absence of pitch accent. In [9], reliability on declarative questions in the Santa Barbara Corpus of American English [10], [11] was 47% agreement for particular pitch accents using full ToBI, and 87% using ToBI-Lite. Once pairwise agreement had been calculated, we worked collectively to arrive at a consensus for the final coding of the test questions. After that, we recoded the training data for inclusion in our study. By that time, we felt confident about the reliability and validity of our coding categories. 2.2. Classification of ToBI Patterns into Constituent Types We found it productive to classify our questions according to broad constituents (head-nucleus-tail), see [12], because of the different patterns and meanings that we found to be associated with the different types of final and non-final contours. For prior literature on this strategy, we can point to [13] and [14], which discuss the relationship between ToBI and such broader categories. Ladd [14] (p. 82) discusses the mapping of ToBI final tunes onto British-style nuclei, and also discusses viewing ToBI pre-nuclear pitch accent sequences as constituents corresponding to heads (pp. 210-211). We classified all the pre-nuclear pitch accent sequences in our question data into heads of different shapes. To do this we had to generalize across some instances of ToBI categories, e.g., an initial L+H* or L*+H or H* followed by a !H* or sequence of !H*'s was classified as a falling head. In a few instances, we also identified a pre-head (%H in ToBI notation). 3. Results In this section we present the results of our mapping of the ToBI sequences of high and low pitch accents, phrase accents and boundary tones onto nuclei, tails and heads. Our goal in this categorization was to identify stable meanings associated Speech Prosody 2006 Dresden, Germany May 2-5, 2006 ISCA Archive http://www.isca-speech.org/archive Speech Prosody 2006, Dresden, Germany, May2-5, 2006