Seasons

I just ignore it...

Here is an idea for a natural language learning bot - step 1: break down the user's sentence into individual words

step 2: store them in order of usage in a database

step 3: for each word in user's sentence search database for occurance of word, using the next occuring word to continue the sentence, making new sentences in correct gramatical order (assuming the user talks with proper grammar) based on learned words

Here is another idea:

step 1: break down the user's sentence into individual words

step 2: store 2 files: a record of the conversation, each word on a seperate line, and each sentence separated, and a database file. The database file will have a record for each word containing the word name, the total amount of times the word has been used, and a value for the total times associated with every word in the database.

step 3: use the following algorithm to sort out overused words and prevent sentences like "it the and it or and for the": V = A/T x 100, or Value of link between two words = Association total divided by the total amount of times the word has been used, times 100 to make an integer instead of a fraction.

step 4: perform algorithm on every two consecutive words in the user's sentence, to filter out overused words and get the main idea of the sentence, and pick the two consecutive words with the highest link value (V)

Step 5: search conversation record for context that includes both words, and use that context as a response.

I think the first idea is roughly how Daisy and Billy work, isn't it? Needs massively more space and power to really succeed, but it'll come in time - we'll have cheap as chips multi-Tb storage, multi-THz processors and multi-Gbit broadband by 2020.

And of course "assuming the user talks with proper grammar" doesn't take account of the various Leeds flies that are bound to end up in the proverbial ointment if it's online and chatting freely. And if it isn't, it's going to take an awful lot of man-hours as a one-man DIY project.

I don't know why Daisy-esque bots can't handle preformatted conversation, but when I feed her Shakespeare, her responses become scene-sized and still incoherent (splendidly florid, admittedly.) It ought to be possible to train NLP bots with cut+paste data, but I haven't found a way yet.

its not such a large DIY project. Just make a "suggest" program that suggests the rest of a sentence based on past consecutive occurances, and adapt it into a bot. using Visual Basic, i succesfully created the suggest proram but when i tried to turn it into a bot, i stumbled on dealing with the database file and coding instructions for special cases such as end of file conditions where there is not consecutive occurances. The result was a bot that learns efficiently like daisy and talks incoherently like billy. Sometimes I got some intelligable talk out of it, but most of the time it was mish mash. that is probably because I did not write the program properly.

And, for some reason it only recognises data learned while the program is running, it can't seem to start at the begining of the database. again, probably a coding error. The second idea on the other hand, will require a colossal amount of data and processing power, since it has to record the entire database squared - for each word it has to make an association record for each other word in the database and has to update everything every time the user talks. That can be avoided by using a spreadsheet type database where instead of making colossal records, it can work like a times table, with the association values between words in the box where the two words meet. that would save a lot of disk space and speed requirement.

all in all, an online version of such a bot would be interesting, to watch how its vocabulary and coherence evolves.

Whens the next Bot-holiday?

They're random. Not even the Professor knows when they'll happen.

Really? How often does it happen, like every month?

Every day has a 1/10 chance of being a holiday, as I recall.

1/10? thats too high of a chance for me!