Learning to predict pitch accents and prosodic boundaries in Dutch

Abstract

We train a decision tree inducer (CART) and a memory-based classifier (MBL) on predicting prosodic pitch accents and breaks in Dutch text, on the basis of shallow, easy-to-compute features. We train the algorithms on both tasks individually and on the two tasks simultaneously. The parameters of both algorithms and the selection of features are optimized per task with iterative deepening, an efficient wrapper procedure that uses progressive sampling of training data. Results show a consistent significant advantage of MBL over CART, and also indicate that task combination can be done at the cost of little generalization score loss. Tests on cross-validated data and on held-out data yield F-scores of MBL on accent placement of 84 and 87, respectively, and on breaks of 88 and 91, respectively. Accent placement is shown to outperform an informed baseline rule; reliably predicting breaks other than those already indicated by intra-sentential punctuation, however, appears to be more challenging.

Exploring the relationship between intonation and the lexicon: Evidence for lexicalised storage of intonation

“…These studies are firstly situated in the domain of psycholinguistics, where several experiments demonstrate that the familiarity or frequency of prosodic parameters influence speech processing, perception and production (Braun et al, 2006;Braun and Johnson, 2011;Mandel et al, 1994;. A second research area which provides evidence for lexicalised storage of intonation is the area of machine learning, where various studies showed that word identity helps in predicting pitch accent location (Brenier et al, 2006;Nenkova et al, 2007;Pan and Hirschberg, 2000;Pan and McKeown, 1999), and where instance-based learning of prosody outperforms other types of learning (Marsi et al, 2003).…”

Frequency matters

“…Up to now, virtually no exemplar-theoretic research has examined pitch accent prosody (but see Marsi et al (2003) for memory-based prediction of pitch accents and prosodic boundaries, and Walsh et al (2008)(discussed below)) and to the authors' knowledge this paper represents the first attempt to examine the relationship between pitch accent prosody and information status from an exemplar-theoretic perspective. Given the considerable weight of evidence for the influence of frequency of occurrence effects in a variety of other linguistic domains it seems reasonable to explore such effects on pitch accent and information sta-tus realisations.…”

An analysis of prosodic boundary detection in German and Austrian German read speech

“…Several prosodic annotation systems have been previously proposed. Some of them combine acoustic, lexical and syntactic features (e.g., [5] for American English, [6] for French), others use lexical and syntactic information alone (e.g., [7] for Dutch). Since our aim is to build a tool that can be incorporated into the annotation process of a not-yet annotated database and also into an automatic speech recognition (ASR) system, the requirement for the system is to exclusively use acoustic information.…”