Learning PP attachment for filtering prosodic phrasing

Learning PP attachment for filtering prosodic phrasing Olga van Herwijnen and Jacques Terken Technology Management Eindhoven University of Technology P.O. Box 513, NL-5600 MB Eindhoven The Netherlands 10.M.v.Herwiinen,J.M.B.TerkenPtue.n1 Antal van den Bosch and ErwinNIarsi ILK/Comp. Ling. and AT Tilburg University P.O. Box90153,NL-5000LETilburg The Netherlands {A.vdnBosch,E.Marsi}@uvt.nl Abstract We explore learning prepositional- phrase attachment in Dutch, to use it as a filter in prosodic phrasing. From a syntactic treebank of spoken Dutch we extract instances of the attachment of prepositional phrases to either a govern- ing verb or noun. Using cross-validated parameter and feature selection, we train two learning algorithms, TB I and RIPPER, 011 making this distinction, based on unigram and bigram lexical features and a cooccurrence feature de- rived from WWW counts. We optimize the learning on noun attachment, since in a second stage we use the attachment decision for blocking the incorrect placement of phrase boundaries before prepositional phrases attached to the preceding noun. On noun attachment, IB 1 attains an F-score of 82; RIPPER an F-score of 78. When used as a filter for prosodic phrasing, using attachment decisions from IB 1 yields the best im- provement on precision (by six points to 71) on phrase boundary placement. 1 Introduction One of the factors determining the acceptabil- ity of synthetic speech is the appropriate place- ment of phrase boundaries, realized typically and most audibly by pauses (Sanderman, 1996). In- correct prosodic phrasing may impede the listener in the correct understanding of the spoken utter- ance (Sanderman and Collier, 1997). A major factor causing difficulties in appropriate phrase boundary placement is the lack of reliable infor- mation about syntactic structure. Even if there is no one-to-one mapping between syntax and prosody, the placement of prosodic phrase bound- aries is nevertheless dependent on syntactic in- formation (Selkirk, 1984; Bear and Price, 1990; van Herwijnen and Terken, 2001b). To cope with this lack of syntactic information that a speech synthesis developer may face currently, e.g. in the absence of a reliable parser, several strategies have been applied to allocate phrase boundaries. One strategy is to allocate phrase boundaries on the ba- sis of punctuation only. In general, however, this results in too few phrase boundaries (and some in- correct ones, e.g. in enumerations). A clear example of information about syntactic structure being useful for placing phrase bound- aries is the attachment of prepositional phrases (PPs). When a PP is attached to the preceding NP or PP (henceforth referred to as noun attach- ment), such as in the structure ... eats pizza with anchovies, a phrase boundary between pizza and with is usually considered inappropriate. How- ever, when a PP is attached to the verb in the clause (verb attachment), as in the structure ... eats pizza with a fork, an intervening phrase boundary be- tween the PP and its preceding NP or PP (between pizza and with) is optional, and when placed, usu- ally judged appropriate (Marsi et al., 1997). Deciding about noun versus verb attachment of PPs is a known …

Content truncated to 500 words