Google stuck in NLP uncanny valley

Google Maps (China) tries automated toponym translation and fails.

Putting aside the question of how useful street-level translation really is for tourists rather than, say, conquerors (if you can’t read local language maps, how will you read the street signs?), this is actually 80% of the way there for pure transliteration items (a bit stilted). The main problems, in red circles, come from compound toponyms with attached literal meanings (X Pond, X Hill, X Course, Old X, etc.) or with target-language conventions already (Massachusetts X, Oxford X, etc.). Those require translation rather than transliteration, and there the Google bot enters uncanny valley with its vain attempts, which bothers people a lot.

All in all, there were only a few true full-blown errors. One was “Boston College”, which got translated as the phrase used for “Boston University”. Another was “Fort Independence”, which was just wrong to be carried with no syntactical change into Chinese. “General Edward Lawrence Logan” was parsed very badly, resulting in a transliteration of “General” as if some kind of Hispanic first name.

Transcription: How Chinese Wikipedia fell into disarray

The evolution of the Chinese language Wikipedia follows a tortuous path. I suppose I’ve been around since the beginning, but really only to watch from the sidelines. In the beginning it was mostly mainland users who dominated in numbers, but since a year or two ago, with the on-again-off-again filtering of mainland Chinese users, the site has shifted towards more users from Taiwan, Hong Kong, and elsewhere.

In recent months, some changes were made to the site with interesting implications. These changes are fairly unique to the Chinese language site but there is something to be learned from them.
(Read the article)