Visual comparison of inventories for a given language

[DanielSWolf]

Many languages have more than one inventory. Some of those inventories are near-identical, some differ greatly.

I'd like a way to get an overview of how these inventories differ; to see which phonemes are consensus across all inventories, which appear in most of them, and which are exclusive to one or two inventories. To this end, I'd love a direct comparison feature on the Phoible website.

To show what I mean, I created a small working prototype of this feature. You select a language and a strictness for grouping phonemes, and it displays a table comparing all inventories for that language.

[bambooforest]

@DanielSWolf -- wow, this is sleek and very useful. Thanks! Since most inventories are singletons, it might be helpful to have an option to drop those or to indicate somehow which languages have multiple inventories.

There will also be some systematic differences between sources, e.g. an inventory from UPSID and an inventory from another source will always differ by tones, since the former doesn't document tone.

[xrotwang]

Do you think there is some sort of standard or default similarity - i.e. something that most people would agree is a useful metric? If so, we could link to the top-five most-similar inventories from each inventory page, as we do in Concepticon - where it shows e.g. that "Swadesh 100" isn't specific enough, and in fact that there are lists "in between" two versions of Swadesh's 100.

[xrotwang]

A similarity metric that works across all inventories would probably depend on phoible/dev#333 , though :)

[camoverride]

@xrotwang - this might be off-topic, but I spent a bit of time playing around with similarity metrics using phoible data: https://github.com/camoverride/notebooks/blob/master/notebooks/Sound_Similarity.ipynb

[xrotwang]

@camoverride thanks for the pointer! Your exploration of similarity metrics is quite interesting - and exactly the kind of analysis that we aim to make easier with CLDF (see cldf/cldf#93).