https://snap.berkeley.edu/project?user=toontalk&project=question%20answer is a project that answers questions about Snap! using the Snap! Wikipedia article. You can replace the passage with anything else you want to ask questions about. It does a good job but is not infallible - see the answer to the question "Who is Brian Harvey?"
I asked "Who made Snap?" and it said "Scratch"
... and so we learn why access to big data is important in machine learning. :~)
Try "Who created Snap! ?".
I feel like using google search API locked onto or in favor of the Snap! wiki website might be a better way of doing this. Google has a good way of looking for what youre trying to say.
comparison:
search "who made the eiffel tower" vs "why was the eiffel tower made"
or "what does entropy mean" vs "how is entropy used"
You mean the Duck Duck Go search API, right?
Whatever honestly. A lot of these search engines employ various AI techniques to be able to guide the user in the right direction even if the same keywords appear in different orders or in a different meaning. Usually, they give you a top result with some text containing the most useful or pertinant information from the article or source. That means you don't even have to actually open up the wiki page, as it should be able to highlight your answer already!
Thats why when you google some history stuff and see that an article has some text that seems very useful, you click on it to go read the rest/full information.
The issue I guess with this though is that sometimes the first result is not the best and it might link the wrong kind of article because it didnt really understand the question. I guess there may be some more well adapted search engines, or perhaps the wiki itself could use a bit of machine learning to get a
related topics section.
Sounds like an excellent project! Thanks for volunteering.
Maybe I should have given some motivational context for this question answering block. Users can use it inside their projects such as
- Adding speech input and output for questions and answers (note for this a single sentence answer is typically all that is required)
- Contacting Wikipedia for an extract to use as the source of answers using a URL such as
Note also unlike Google search - the question answering block works without a network connection (assuming the text passage is part of the project).
The overall point is to provide building blocks for Snap! users to make their own Alexa/Siri/Cortana/... Both to understand better how these things work but maybe to make a version with their own special twist.
whenever I ask a question on mobile, it says "sorry, I don't know the answer". Is this just on mobile, or not?
EDIT: I just realized that what I'm typing isn't getting typed into the ask box
duck duck go was made for more private web searching. so almost any web browser will work basically the same.
Android or iOS? Version, browser, browser version? We recently fixed a bug like this on Android.
well, technically it was on my Braillenote touch plus, but that runs on android, so it's android, but I found out how to fix it, I just close the virtual keyboard and tap on the ask box again and it works.
Thanks, that's a helpful report. If Morphic 2 doesn't just automatically fix it we'll look again.
it works now without closing the virtual keyboard
Oh, okay, never mind. :~)
Another point is that the passage used to answer questions can be private. E.g. your notes from a meeting or a doctor's notes on a patient. No way to use Google or Duck Duck Go.
Was Snap inspired by Scratch? I also have a suggestion, with the 3.0 version of Scratch, maybe there should be updates to Snap.
Yes indeed. Snap started as a modification of Scratch 1.4 titled BYOB.