Non mi capita niente.

Nothing happens to me.

/ˈka.pi.ta/ (càpita). But Giorgio says “capìta”.

He probably thinks it’s “the feminine singular of the past participle of capire” (capita - Wiktionary). :wink:


Capisco, mi capita di continuo.


No, in this case “capita” derive from “capitare”

1 Like

The grammar breakdown correctly identifies the verb as capitare, which tells me that the audio and the grammar analysis software are not cooperating.


Interesting, and I guess that could be a good thing? I wonder if this could be applied somehow to at least flag conflicting sentences for a manual check?

I doubt it. My guess is that the TTS is like a black box that only outputs sound.


I was wondering it, because there is the pronunciation field too, which can also be specified when entering custom sentences.

And since we’ve seen sentences where a completely different word is said, compared to what was written in the sentence, we know they’re not generated on the fly. So I thought there was a way of “correcting” the TTS in the background.

Though I guess a simpler way might be to start with just flagging words that are written the same, but can change meaning depending on the pronunciation. But yeah, all that is probably too complicated and requires some sort of source to match this against.

Isn’t it a major flaw? When learning, one would love them to cooperate. This way we would not waste time catching wrong pronunciations.( Like in the futile battle with the machine)

After all that’s been said, and the errors that we have found, I still think the machine is doing a surprisingly good job trying to parse and speak our human languages. I don’t think they understand us yet, and only when they do will we get perfection, but at what price? :thinking:

I prefer to see these imperfections as reminders to stay on our toes and be vigilant of the machine. It also gives me some satisfaction when I’m able to outsmart our robotic overlords. We should still flag all the errors we find so that others can be warned, and maybe a human voice can step in and cover for Giorgio and Carla when nothing else helps.


I had the exact thought about being on our toes. My version was about our ears to be perked. Agree to everything you said.


True, I still first and foremost already greatly appreciate what we have :relaxed:

If it weren’t for the machines (or the incredible Tatoeba volunteers), we wouldn’t even have this amazing resource at our disposal. Constructing all these language pairings through manual labour would not have been even remotely possible, without needing to charge hundreds of dollars/euros/… for a single language pairing course (as is indeed the case with many of the more “traditional” language learning resources).

And another incredible - and to me indispensable - resource, is that we we have this amazing space to discuss with and help each other figure things out!

Plus it’s been great seeing how far I’ve already come these past months (nearly year), to be able to even notice any potential mispronunciations and grammatical errors, and I imagine others might feel the same.

It shows overall the system does work, and really enables us get a feeling for these things. And now with the added grammar feature, this can only improve even more.

I was only curious if language pairings, which don’t have such an active discussion group, might benefit from some sort of flagging for check if grammar and pronunciation didn’t agree with each other (even just by displaying a little cautionary mark next the sentence itself, to keep a learner on their toes, if they don’t have anyone to verify it with), but then at the same time, I don’t think that, apart from perhaps in English and a handful of other languages, the issue of the pronunciation changing the meaning is as pervasive.