D. Gibbon, U Bielefeld
Top - Description - Functional specs - Design specs - Assignments
Citation: publication, online tool
Demos based on language atlases of Côte d'Ivoire
| |
Example: virtual maps of Kru languages These virtual maps are based on differences between the consonant inventories of Kru languages of Côte d'Ivoire (see Kru demo link for further information). The animation cycles through different numbers of the main pairwise distinguishing features/properties (consonants, not phonetic features), from distance 0 (0 differences) to the maximum distance 19 (19 differences). The main point to note is the gradual emergence, first of the Eastern and Western Kru groups, and then the emergence of the entire Kru family. More detailed phonetic feature criteria, or other criteria may lead to differences in details.
|
Example: similarity dendrogram for Kru consonant inventories
The languages are grouped into clusters on the basis of their similarity in terms of shared consonants. The clusters are grouped into larger clusters, also on the basis of consonants shared between clusters. The similarity relations expressed by the dendrogram correspond closely to previous linguistic classifications in some cases, but not in others, a result which is to be expected since only consonants were used, while earlier classifications are also based on lexical, grammatical comparison and on the comparison of phonological constraints.
|
Top - Description - Functional specs - Design specs - Assignments
Description
DistGraph is a parametrised online tool (select Interface for data input and parameter settings) for displaying distances between entities and for discovering clusters of similar entities such as related languages, based on differences (= distances) between selected sets of properties / features. The tool is parametrised to be able to specify the maximum range of distances to be displayed. The larger the specified distance, the more distance relations are included. With smaller distances, the graph tends to split into subgraphs showing less related clusters of units.
The empirical reasons which determine these distances are complex. The entities may indeed differ exactly as calulated, but different methods may have been applied to analysing the entities, or there may simply be errors in the data. Therefore careful plausibility checks with reference to other analyses are required.
Top - Description - Functional specs - Design specs - Assignments
Functional specification: 'learning by doing'
The basic methodologial principle for training in language and speech resource creation for speech and text is 'learning by doing'. In this context, learning by doing means small student projects with specific tasks in which students learn to develop and process speech and text resources autonomously. Learning can also be in the area of software tool development: the software tools provided are models which be used for the creation of further tools by students with a computational background.
|
Distance graph generation workflow. |
Top - Description - Functional specs - Design specs - Assignments
Design specification
The distances are not represented by edge length in the graphs (in general this would result in too many dimensions to render on a 2-dimensional screen), but are shown as edge labels and also colour coded. (Edge distance rendering can be approximated by dimensionality reduction, but software libraries for this are currently not available on this server.)
|
Browser compatibility: The graphs are rendered correctly by the Firefox, Chrome and Dolphin browsers. After a parameter setting change, some Microsoft browsers (e.g. in Windows 8, WindowsPhone 8) do not update parameters correctly and the output needs to be reloaded in order to do so.
Top - Description - Functional specs - Design specs - Assignments
Suggested assignments
The idea for this online tool occurred to me during a lecture by Gerhard Jäger, Tübingen, on modelling language distances in relation to geography and migration, and later in discussions on this theme with Stavros Skopeteas, Bielefeld, and Firmin Ahoua, ILA, Abidjan. Grateful acknowledgements to François Kipre Ble, ILA, Abidjan, for loaning a copy of the Kru language atlas used for the demo data, and to colleagues at ILA for donating the two-volume Kwa language atlas. Many thanks to Jue Yu for suggestions on improving the user interface, and to Jolanta Bachan for overall assessment.