Dutch Map Task Corpus, 1999

Abstract copyright UK Data Service and data collection copyright owner.This project dealt with the phonetic details of intonation in Dutch and English. It focused on the alignment of intonational targets (e.g. local peaks and valleys) with the vowels and consonants in speech. Limited past research had suggested that this is systematic, but the factors that affect it are not well understood. The depositor's earlier research suggested that in many cases intonation targets are anchored to specific sounds (e.g. the beginning of the vowel following a stressed syllable). This kind of precision was rather unexpected, because investigators have concentrated on more variable effects (e.g. the closer a target is to the end of a word, the earlier it is aligned). The main goal of this project was to determine how general this anchoring is, what kind of landmarks (consonants, vowels, word ends, etc.) can serve as anchors, and how much the alignment of anchored targets can be affected by more variable factors. One practical motivation for this research was to provide the basic knowledge for improvements to synthetic speech. Most of the empirical research of the proposed project consisted of experiments in both English and Dutch, in which carefully selected sentences were read aloud and detailed acoustical measurements made of the speech. The depositor also studied short (5-10 minute) dialogues spoken under somewhat controlled conditions these are the Map Task dialogues deposited in this dataset. English and Dutch were chosen because their sound structures are similar enough that conclusions can be generalised from one language to the other, yet different enough that certain kinds of experimental controls can be used in one language which would be impossible in the other. Also, both languages support important speech technology industries.Main Topics:This corpus of natural Dutch conversation was collected as part of a project primarily concerned with the phonology and phonetics of intonation. The Map Task procedure for collecting spontaneous speech was used. The Map Task is a widely used tool in the study of dialogue, because it allows researchers to study conversations which are completely spontaneous and yet remarkably predictable and consistent. The task works as follows: the two participants to the conversation each have a map showing a variety of pictured landmarks with names like shepherd's hut or Green Mountain. The maps may differ slightly in detail; crucially, one map (the instruction giver's map) has a route marked on it; and the other (the instruction follower's map) does not. Neither speaker can see the other's map, and in some versions of the task (but not this one) the speakers cannot see each other. The task is for the instruction giver to explain to the instruction follower where the route passes, referring to the various landmarks along the way, accurately enough that the instruction follower can reproduce the route on his or her own map. The basic reference on the Map Task is Anderson et al, (1991), The HCRC Map Task Corpus, Language and Speech 34, 351-366. Further information on the Map Task is available at: <a href=http://www.hcrc.ed.ac.uk/dialogue/maptask.html> http://www.hcrc.ed.ac.uk/dialogue/maptask.html </a> The point of using the Map Task was to obtain natural productions of certain intonation patterns (e.g. various kinds of question intonation) which are difficult to obtain in reading experiments without explicitly instructing the speakers how to speak (and sometimes not even then). The most important manipulation of the maps was to select landmark names that manifested the phonological structures that the depositor was interested in, and that contained consonant types which would permit easy analysis of pitch patterns. However, the basic conversational task was unaffected by these manipulations, and conversations in the corpus are entirely comparable to those recorded in various languages elsewhere. So far as the depositor is aware, no other Map Task corpus exists in Dutch. The conversations were recorded at the phonetics laboratory of the University of Nijmegan on 5 February 1999 (day 1) and 8 February 1999 (day 2). In both cases a complete quad (4 speakers, 8 conversations) was recorded. The speakers were all students at the university. The maps were based on maps from the original HCRC Map Task. The distribution of the landmarks and the route on the giver's map were identical to the originals, but the actual names of the landmarks were in Dutch and in most cases the pictures had to be adapted as well