Welcome to my epic interactive adjective ending practice chatbot.
Yes, you read that right. I made a chatbot that practices adjective endings with you. As much as you want, as patiently as you want and as fun as you want.
Well… okay, that last one is not true. You can’t really dial up the fun.
But I think it’s more fun to use this bot than to do a bunch of fill in the blanks.
If you want to know how it works, check out my description below.
But now, enjoy practicing adjective endings.
How the adjective ending exercise chatbot works
The bot will create multiple choice questions for you, one at a time, where you have to pick the right ending for an adjective that you see in context.
What’s special about this is that you get to select the gender you want to practice.
Some of you might know this from my exercises about German cases.
I think it’s VERY useful to take the noun gender out of the equation because that’s just an additional “uncertainty” that might throw you off.
The exercise generator lets you pick one gender and stick with it. You can always switch, as well, and, if you want the real deal, you can also pick the “mixed” option.
But yeah, the bot gives you an example, offers 4 (or 5) answers and you pick the (hopefully) correct one, either by clicking or typing the number. Do NOT type the adjective and do not type the ending.
In fact, this brings me to an IMPORTANT NOTE:
!! YOU CAN NOT REALLY CHAT WITH THE BOT !!
You can try of course, but it won’t take you anywhere meaningful. That is simply not the purpose of this, and I pretty much eliminated that option. If you want to chat, you can do that with ChatGPT directly.
All right.
Now, this really is a prototype and it’s using LLM technology, so chances are very high that you’ll see some minor or major glitches.
Here’s a little overview of what might happen:
- little variance in examples (for instance using the same noun three times in a row)
- 5 options instead of four with the correct one being double
- a German example that is in fact grammatically wrong
- random freezing
- faulty or nonsensical grammar explanations after you gave a wrong answer
As of now, the bot is set to give you a quick grammar breakdown of the example if you don’t answer correctly.
I’m not sure if that is a good idea, because I have seen some flawed analyses already.
Like, it’ll say “Accusative” when it was Dative and stuff like that.
But I’ll leave it in for now, because I want to get a gauge how often that happens.
Audio Speed
Dark Mode
Welp pretty good tool to geta hang of endings .
Happy you like it!
Finally a perpetual drill tool to drive understanding through repetition and correction in real time – no waiting for result or explanation. Awesome, really!
Enjoy it!
Just FYI I had to downscale the subscription there because it was fairly expensive and barely anyone was using the tool, so now I’m on a more limited plan. So if it stops working at some point, that’s the reason. I’m out of credits then and it’ll refresh next month. No need to hold back though, so far only 1% of credits has been used, so 99% to go.
Prob hard to get accurate often enough but it would be pretty cool if you got it to explain using your method rather than traditional grammar methods.
Oh wow, I don’t think I can do that without either custom training or using some good old if-then programming logic :).
For a normal LLM, this could never work, as it won’t actually understand what I want it to do.
I don’t think it’s actually that hard to do. You are using some sort of llm model api like OpenAI’s api, right? You can prob literally just feed in your article/s content on the subject as part of the prompt and get back decent results.
Of course you need to mod the prompt to follow your articles content as guidance
Damn now I kind of want to scrape all your articles and feed it into a model to make an ai “mini Emmanuel” that I can bother with German questions at will :)
I was actually thinking bout doing that and putting it up here :)
Doooiiiit :)
Prompt won’t be enough for that. The model has a lot of “weight-information” about the classical adjective ending scheme and you can’t really overwrite that fully with a prompt consistently over time. The chat mode also can’t reliably reason and analyse endings, so you’d need some dedicated program to run in the back too.
It was quite the pain in the butt to get this exercise to work and only one third of what’s happening behind the scenes is actually LLM AI. The rest is good old classical coding logic with decision trees and if/else-stuff.
If you ask questions about a topic yes, that can work (never guaranteed though), but as soon as you have a scheme like that it’ll become difficult.
You can try it out and prompt the model to explain your adjective ending mistakes in three tiers:
– You forgot -e
– You didn’t add -en, even though the article is weird
– You didn’t mark the case/gender information on the adjective
I don’t think it’ll even get 70%. The rest will drift toward classical adjective ending explanation.
You have to keep in mind that it’s always just predicting the next word, even with a big cloud of context influencing it. The classical analysis is part of that influence cloud and it’s a matter of probability that it gets to influence the predicted word and such the whole sentence can veer off easily.
Yeah true… I have built ai agents for some very different things but I guess the language reasoning stuff is quite different.
It said “not quite” but then explained the answer I had given
Question 1 :
7:38 pm
___ Musik beruhigt.
7:38 pm
Beautiful music calms.
7:38 pm
Options:
schöner
schöne
schönes
schönen
7:38 pm
7:38 pm
2
Incorrect, not quite.
7:39 pm
Correct: Schöne Musik beruhigt.
7:39 pm
In the noun phrase “Schöne Musik”:
“Schöne” (adjective): nominative case, feminine, singular (modifies “Musik”).
“Musik” (noun): nominative case, feminine, singular.
The AI will do that from time to time, unfortunately. You just have to ignore these kinds of glitches, there’s no way to avoid that 100% of the time, unfortunately.
Mein lieber Haustier ist sehr verspielt. why is it lieber and not liebes?!
Should be “liebes”, you’re right. It’s the AI glitching.
Ich finde es sehr hilfreich!
Freut mich :)!
This is REALLY HELPFUL. I’ve been struggling with adjective endings since my college days and have given up on them as I continue with German as a hobby. Recently I came across an old (1980’s) German Language text at the library which lays out adjective ending in a way that I can finally understand. Your adjective endings quiz generator has been a perfect way to drive home what had previously been only loose concepts.
I don’t have any other feedback at the moment other than I do find it very helpful to be able to sort by gender.
Thank you Emanuel, I’ve been a happy subscriber for some time and appreciate your innovations!
Dave K.
Happy to hear that it’s helpful and that it wasn’t too glitchy :)!
Your examples were more varied and instructive. The Bot glitched less on the second round.
Thanks Emmanuel. Despite encountering a few glitches (i.e. where the tool disagreed then agreed with a small % of my answers) I have found the adjective declension tool easy to access, very helpful with its logic, and effective in stimulating lazy brain cells into German thought.
Awesome, happy to hear that :)!
Moin Moin Emanuel,
ich habe ein bisschen morgendliche Übung gemacht und bin auf diesen Satz gestoßen:
“No, that’s incorrect.
10:11 am
Correct: Ich gehe mit der schmalen Straße entlang.”
das “mit” kam mir komisch vor und habe ich deshalb nachgefragt (chatGTP hehe) er sagt auch, das sei nicht korrekt. Was ist deine Meinung dazu?
It’s wrong!
If you come across something like this, don’t question yourself too much. Chances are very high, that it’s simply nonsense.
Ein schöner Tag …. – schöner wasn’t an option ?
Apart from this, it’s brilliant!
These glitches will happen from time to time. I don’t make these examples, the bit makes them, so I can’t fix anything :).
Hi Emanuel, thanks for the quiz. It was useful and fun to do. I enjoyed it very much.
I have one question though. The bot thinks the following is correct:
Ich höre gerne klassischer Musik.
Shouldn’t it be ‘klassische’?
Yes, definitely.
Thanks for your reply. Schönen Tag noch.
I liked it alot – I found it quite helpful. Didn’t have any buggy problems
Great to hear. Maybe my modifications really worked :)
Loved it… until it stopped working after 10 questions.
What happened then, exactly? Could you restart?
Question 4 was Ich gebe dem ……. Kind ein Geschenk. The choices were all ‘kinden’ variants which was puzzling as ‘klein’ was intended, judging by the explanation. But I liked the exercise.
I like it because I need all the help I can get and drills help.
The AI is still a bit buggy, often none of the four options are correct… Der Apfel ist rotem/e/en/er.
Also seemed to get in a rut in the feminine giving only nominative case examples for quite a while.
Not perfect but thanks for the effort and I am using it.
The cases are selected at random by a proper random generator, so that has nothing to do with ChatGPT. Probably just hit a “feminine” streak.
How often does that happen that none of the options fit, though?
The “show grammar” bit is brilliant
Be careful with that though. There is a bit too much nonsense in it for my taste.
However, if you like this kind of breakdown and you use Chrome browser, you might like my Grammar Analyzer Chrome extension. That one does a better job than the grammar explainer here.
It let’s you mark a sentence on a website you’re reading and will break it down, translate it and paraphrase it.
https://chromewebstore.google.com/detail/ai-grammar-analyzer-by-yo/dbekogoohigpdgepedgkbhekpmknhdfd?hl=de&authuser=0
I think I’ve run into a couple of glitches.
1) While I was doing a list of supposedly all feminine nouns, a neuter one (das Kind) crept in, which threw me off a bit.
2) I chose “frisches” as the ending for “I give the dog fresh water”. Because it seems to me that (to)the dog is Dative, but water is Accusative, because it’s the object of the noun to give. Dog is the indirect object & therefore Dative. Is that not right?
As in: Who gives (I = Nom.)? What do I give (water = Acc.)? To what/whom do I give water (to the dog = Dat.)?
Hey hey, I reviewed the chat and I think the bot got confused with “child”. The noun it was supposed to use was “Musik”.
I don’t think I can fully prevent these glitches unfortunately.
You have several questions where one must choose ‘oft’ vs alternatives. These are useful, but ‘oft’ is an adverb not an adjective, so I am puzzled why they are included. Regardless, thanks for developing this as it has taught me my instincts aren’t as good as I thought. Anne
I hate to say this, but I found the generator unusable. I attempted two sets. In that time there were multiple occasions where I was given the answer to a completely different question that I was asked, that a set of answers would appear without a question, and when I clicked “show grammar” after a correct answer, it said that it was not equipped to engage in chat and moved me forward to some apparently random place. Some of this is expected, of course, but more than half the time GIGO is too often.
This is really weird because 80% of interactions are scripted and don’t involve AI at all.
Sounds like an API problem. Is it possible that you’re using an iPhone?
No. Thinkpad. Using the Opera browser, though/