As the FAQ of the library is stating:
Can langdetect handle short texts?
This library requires that a detection text has some length, almost 10-20 words over.
It may return a wrong language for very short text with 1-10 words.
You are trying it on one-word or two-word texts, this is not the use case this library is build for, so you're gonna have wrong results.
For single words without context, you can try to match them with dictionaries of the languages you are targetting.