Kenneth Benoit

Please include all of your output, for which we recommend using `reprex::reprex()`. It looks like you have not loaded **quanteda** yet.

It looks like your version of R is out of date. Try installing R 4.1.3, updating your packages, and then re-installing **quanteda**. ```r update.packages(ask = FALSE) install.packages("quanteda") ```

Interesting example! Our processing of globs takes place in `utils::glob2rx()`, and it it not perfect. Rather than splitting punctuation and then trying to reconnect them by compounding, I suggest tokenizing...

On the second item, we don't have `language` implemented at the moment via `dfm()`, so it's better to stem the tokens first, and then create the dfm, for non-default language...

Thanks @ojwb that's really informative and useful. @koheiw @jtatria time for a replacement of SnowballC, or just a German stemmer package?

Good use case, I agree this is tricky in the absence of a function or operator for this. On naming, it's a syntactic deviation to use `dictionary_subset` since all other...

What about overloading the `[` operator with a `levels` argument? Like in `drop` in `[.data.frame`?

Sounds good. It's a pretty specific use case, but I can see how it would be useful for named elements. Would it work for numeric indexes, and negative indexes too?

You should be able to supply `df` -- or if that does not work, `as.character(df)` -- as the argument for `texts` in `findThroughts`.

Yes it is, although as @koheiw and I are planning a more general discussion of fcm manipulation options for v1.5 I'm going to add this as an issue there just...