Thanks :)
The first draft I played around with had a lot more details on the language issue, roughly along the following lines:
— The plants can detect events in the real world through sound + the ability to detect uneven light distributions on their stem, which effectively lets them see.
— Correlating events occurring in the real world to their chirps allowed the research team to construct a primitive dictionary (e.g., containing words like sun, rain, etc) as well as understand the underlying grammar of the language.
— Remaining words (for example, dumb vs smart) are worked out from context (i.e., a plant whose poem is called dumb is more hesistant to offer another one).
— All this is done in 9 months (as opposed to 50 years) with the aid of some futuristic machine learning software, which the foundation has in the late 90s.
Ultimately, I ended up not using all this material — it added some verisimilitude but bogged down the entry in detail.