50

Dante Ashton wrote:@Kryten, yes, your right, it wouldn't be such a load, but I'm sure I don't need to tell you how much work would need to go in for such potentiality little reward.

Unfortunately not, no. And not necessarily in terms of coding it either...finding a good quality semantically tagged corpus with license terms compatible with F/OSS is probably at least as problematic as the coding problems. I know that the ones I've worked with have _very_ restrictive terms of use attached.

Although...

Dante Ashton wrote:Kiim's natural langauge module is going to have to be easy to replace, as well. For localization.

Maybe, although this is coupled with my earlier ideas...I've seen perl-scripts using nothing more than aspell-dictionaries, some baysian logic, and n-gram computations being able to determine the language in use to a fairly high degree of accuracy. I'm afraid I'm not at liberty to disclose them though. Granted, that was text-only and only using indo-european languages as the source but as long as the speech recognition doesn't tax the computer too much it should be doable to have just one generic module that can adapt to the language in question.

And to be honest: that is, I think, the biggest hurdle to overcome. Speech recognition. I admit, that is not my specialty but from what I've come to understand about it it's really CPU-heavy (not to mention, highly dependent on the speech pattern, accent/dialect, and articulation of the individual).

Dante Ashton wrote:Hmm...well, how about this?

Start off with keywords, not full, natural langauge, just the keywords.
Link into basic programs (without feedback) like K3B, allowing basic control.
Complete keyword system, work on parsing it through 'junk'. (IE: Excess language)
Output through Festival

Bad idea...
We've to build a clean infrastructure first. K3B-hacks won't help us. Create the base-classes and then begin with implementation and there will be a better app...

@Kryten;

Your right, speech recognition is very intensive...but not the kind I hope Kiim to use

Quick tutorial; there are two kinds of speech recognition, open-domain and closed-domain. Open domain learns and can recognize words it's programmers never intended to. Most forms of this use the majority of processing time, and half an hour of training creates around 1.5 gigabytes of data. (The market leader here is Dragon NaturallySpeaking...I really have been hoping for a Linux version, as MS's inbuilt one isnt entirely useless...)

However, Kiim would be using 'keywords' as her only control route, which means we should be able to utilize a closed-domain model; almost 100% accurate and no training required . The downside is this system, by it's very nature, cannot be trained, any expansion requires the insertion of keywords (or a complete rebuild of the model). Another downside is; don't expect to be able to transcribe an email.

I 'think' CMU Sphinx can do both. It depends, really, if you have the memory and processing power to burn (but then again, a computer running Kiim wouldnt really be doing much else...at least, if Kiim were designed in such a way that social networking, IM's, emails, and web browsing could all be done through her)

@The User; Yes...sorry, getting ahead of myself. :P

To help inspire those who like the idea, and to win over those who are thinking about it, I'm currently writing some fiction depicting a day with Kiim. (as I said in the Kiim overview.pdf, I am a writer, researcher and thinker, hopefully my skills, however limited they seem, may be of some use)

Dante Ashton wrote:However, Kiim would be using 'keywords' as her only control route, which means we should be able to utilize a closed-domain model; almost 100% accurate and no training required . The downside is this system, by it's very nature, cannot be trained, any expansion requires the insertion of keywords (or a complete rebuild of the model)

I believe such system as KIIM should be able to use any model of input, no matter text open/closed domain etc. BUT yes, a closed-domain system should be developed first (after text mode) because simply it is easier to do

We should never forget how fast processors is going further and it is possible, that tasks that looks heavy today could be executed in 0.1sec in 3years time

A closed domain system means more manual labor to get right.

I personally think we should look at open-domain from a FOSS and usabilty point of view. Brainshare time, anyone?

(by the way, just wondering if we should open up a freenode channel or something for this. I have to wonder if the devs and admins are getting a little pead off at this thread...unless they're interested in it, that is

@Lukas, your right, we must never forget Moore's Law. I suspect that within a year or so, open-domain speech recognition will fall into acceptable limits.

However, as we are on KDE, we can also look at Threadweaver (Multi-core processing is almost the norm...) as lowering the requirment on machines

I think word should have role(verb, object, etc.) like in rdf. All words should also have priority. We can simple uses K.I.I.M in that way. We can also include some object related function, like the mind can't have color or sun can't have personality. In this way we can simple convert natural language to SQL syntax. We should also allow user to get fedback of working and skip convert step(give K.I.I.M SQL command).

About priority. The idea is quite unfinished, but... -discountable-rati ... 47923.html

Such feature would be very important in cases if you want to replay email to Tom, but you have 4 Toms in your address book. In this scenario right tom would get priority #1 since. O cource we should ask for confirmation, but to confirm is easier than to select from list and then confirm

Ooh, 20 votes!

Hmmm...Lukas, that is a very interesting idea. As is yours, Lachu.

Ok, lets just make sure I understand this; lets say I have a program, which has a value attributed to it by how many times I use it. However, I have begun using a program which does the job far more effectively, and because I havent used the old program, it looses points. Once that loss gets to a certain level, the system may recommend uninstalling it, perhaps?

(Note; I think we should take a stroll through the Brainstorm section and see if there is anything else of use that may not necessarily survive on it's own)

who does uninstalling anything in linux ?

Anyway such rating system should be system-vide. So same rating shoul apply for eg sorting in kickof. Such communication would be useful for everything - if user installs new app and wants to have it as default, he should only add star in menu - no need for deep configuration

Ha! Well, in certain cases, I do. Especially when I've found software to be not quite up to the task I was expecting, or I've found a better one.

Oh; I've just confirmed ConceptNet is still under development.

Does KIIM is going to be manager for user inputed (text or voice) data or for any sort of info like noise level accelerometer (if avaliabe) GPS data etc and could do job without any or minimal user interaction like lower volume in silent area, change IM message, some KDE settings according location (gps or connection name) etc?

I would like second one

Forgive me, I must have too much blood in my caffine streams, but do I understand that correctly? That you'd wish for Kiim to be able to adjust 'herself' in order to fit a particular time/location?

Hmmm...Interesting concept, Kiim would have a 'profile' for preloading apps at Home, and at Work...

Hmmm...the forum's new facelift has gotton me thinking about what an appropiate icon would be for Kiim, any ideas?

Synopsis about the Kiim idea;

To create an interface using natural language processing (either through speech or through typing) to make a computer more usable. To create and expand upon automating the computer to the point where it can more or less handle itself. To create a system that is usable for non-techies.

Also, to extend upon the desktop metaphor, and bring a far more 'human' side to it then the traditional desktop metaphor ever could.

K.I.I.M
Show [installed] programs to converting flv to mp3

And K.I.I.M answers.

K.I.I.M
Is it necessary to install program to converting flv to mp3?

And K.I.I.M answers No, because we have one

K.I.I.M
Which program to converting flv into to mp3 do you suggest

And K.I.I.M answers: You have installed <program name>. Please, using this one; or: <program name> is best rated

K.I.I.M recognize program as keyword with high priority. It next call special DBUS functions or talk with pacagekit.

We can also do map of words for K.I.I.M(some tree type, so Uninstall goes to program with 50 points and to user with 1 point). K.I.I.M would recognize what user mean without program word. Our tree can also learn, but it's very complicated.

K.I.I.M should be only agent for talk. But we prefer to have wide system integration. We should have people profiles and privileges levels(connecting to anacondi or social desktop). For example our friend can asks K.I.I.M where are they. K.I.I.M can answer: According to GPS data I'm(laptop-special-ubuntu/user name) in Chicago.

K.I.I.M should be also used to define privileges for certain users.

In future we can allow K.I.I.M to make remote command if user are recognised as user(prompt password and user name). With way to define privileges level we can control, that we can remotly only check system parameters. It would be realised only in far future(it could be dangerous).