Database of Cross-Linguistic Colexifications

• Modularization (make the code be capable of taking new data as input and automatically producing a project with little tweaking and most things being done with help of templats and config files)
• Website (modify current website structure to be useful on all devices, using bootstrap and other advanced frameworks)
• Port from PHP to JS (replace the current browse functions to use JS and CSV/JSON-input instead of former PHP/SQLITE-code)
• Contribution-Template (create some kind of template, probably email-based, for those who want to contribute new data)

• carry out a language comparison analysis based on the colexification data
  • note that we first have to determine how to measure similarities among languages (Hamming, or something else?)
• enrich and combine our data
  • currently, we have a lot of language data, but some resources are in separate parts (geo-locations are in json, basic informations are in csv)
• switch or add classification of Glottolog
  • currently, we use ethnologue classification, later on, we should switch to Glottolog. For this, we have to export the data somehow, and add it to our resources on the different langauges
• User-Interface
  • simple idea, probably also simple to realize it: add a user-interface where users can add polysemies or colexifications from languages they speak or know
  • this should be independent from the main data at first, but it could be used as a separate source later, in case we get enough information from users
• Display families instead of languages
  • following B. Wälchli's idea, we could use langauge families instead of languages in the geovisualization and highlight them in dependence of colexifications which are found in the data
  • for the geographical representation of language families, the geographical midpoint could be used (alternatively, one language could be used as a representative per family)

• code for computation of the whole network (preferably as json-like format in output)
  • has been added now, but we still need to include the communities for later inter-reference between the simple browse and the JS-browse
• the computed full network
  • added as gml and json
• code for computation of community clustering with help of igraph
  • added, see "communities.py"
• a csv-list containing node and community for all communities in the network
  • has been added now and will be written in the folder output/ as communities.csv (in the format defined for js code)
• a labelling for all communities (preferably the centroid or how that's called for each community)
  • currently, I simply take the node with the highest degree, it is given in the last part of the filenames in communities/-folder
• a csv-list with semantic fields of IDS and nodes belonging to the fields

Use the link.py script to create a network from all wordlists files in the folder "wordlists". The data here is no real network, but rather a collection of all links that is stored in a format suitable for accessing it on the website. The script also automatically generates an updated sqlite3 file that is accessed on the website using the php code in the website-folder.

All essential files for the website that can dynamically be changed are in the folder website/files. In order to update the website, update the files there manually and then run the "makehomepage.py"-script that will use the templates and the content in the files and create a new version of the website. This version is accessible via the folder clics.de/.

In order to retrieve a GML-output file, just use gml.py without any arguments. The script reads in the graph contained in "links.sqlite3.txt" and converts it to gml-format. Additionally, the script adds a tagging of IDS concepts: All items present in the Swadesh-100 list will be tagged accordingly.

In addition to pure gml-export, the script automatically converts the gml-graph to json. This is useful for applications that make use of JavaScript or the like.

run "get_data.py" in order to retrieve some basic statistics regarding the data in our sample.

Use the script "communities.py" to carry out such an analysis. I recommend to run the script with option "infomap", since this is the fastest analysis. It yields some balanced amount of community clusters. But we need to look at the results if we label the communities in order to see what this actually means.

A workflow-shell script that runs the code that has been done so far piece after piece is available now. For convenience, the current steps are repeated here:

# [1] get all links in the data in wordlists/
python links.py wordlists/

# [2] get basic statistics
python stats.py

# [3] get gml graph of the data
python gml.py

# [4] get communities using infomap
python communities.py infomap

Be careful about step 1, where a folder needs to be passed as argument (in our case "wordlists"), and step 4, where argument "infomap" defines the method that is used for the calculation. Infomap is rather fast, the alternative, the original algorithm by Girvan and Newman (2002) is very slow, and will take a long time.