🔒 Locked

Language processor (kindof)

Started by Luctus Mar 10, 2004 at 8:45 AM 2 replies 600+ views

Original Post

Mar 10, 2004 at 8:45 AM

Been working the last night with a language processor where the application recieves a text-file as input and creates a map based on the approximity of different words, from which it produces a random sentence. There''s been some interesting results.. Some (random, not activly selected) examples with an essay on nordic mythology as input: He keeps his feet when she makes her choice. The myths in the triad of the older poem Haustlong where the shape with him to the shape with other traditions usually only have one. She also demands that Lodur and is Loki, the same. I will show that chapter of Loki and cannot draw any substantial conclusion about the desires of her own volition, but where the most apparent of the the the fact that Loki in his own volition, but the the the two other myths where the shape of Loki is the the desires of the other traditions usually only be able see his feet when she will only have one. Loki and Lodur and that the first human beings. In this claim is questionable the same. Further it is Loki, the two latter gods by the fact that Loki is Loki, the creation of two other traditions usually only confuse the fact that are one and Loki. Some of those acutally makes sense. I imagine with a better set of carefully selected inputs, you can shape the results into a more narrow set of outputs, producing a fairly predetermined result with a good deal of variation. Improvements to this would be to, for example, replace all nouns in the input with a tag which would be replaced by a appropriate noun for the current context fetched from a database.. So, what do you think? -Luctus

Statisticly seen, most things happens to other people. [Mail]

Kylotan

Mar 11, 2004 at 12:26 PM

Well, it''s interesting, but I don''t see a direct application for it. In my own thoughts on text generation, I''ve thought that starting with a basic grammar would be most likely to yield accurate sentences, as I think it''s better to have a subset of the language that always makes sense than a superset of the language that sometimes produces nonsense.

[ MSVC Fixes | STL Docs | SDL | Game AI | Sockets | C++ Faq Lite | Boost
Asking Questions | Organising code files | My stuff | Tiny XML | STLPort]

Palidine

Mar 11, 2004 at 12:30 PM

Yeah I remember writing the same program as the standard end of year intro computer science project at college. We called it a babbler. It was funny to put your friends essays through it and have it generate essays in the style of said friend. You can do funny things with political speeches as well. It works better as a satire on style then as a generated speech thing though. I think it might be really hard to get meaningful babble out of it. But if you can that''d be really cool.

-me

Luctus

Mar 11, 2004 at 2:18 PM

Well, my intended *therotical* application for it was that when giving it a narrow set of selected sentences, you can let it dynamicly generate diversive variations on them while still keeping a fairly predictable result. Something to cause a bit of change in dialouge when talking to npc's that don't serve a particualr plot in any way, but rather supply tidbits of information in for example, a RPG.

But I believe the realization I've come up with is too ineffective to be practical, it takes too much memory keeping a map of all the words and how they relate. Secondly, I think the time that's necessary to put into selecting good sources doesn't compete with the time and result that it would take to write a variations of sentences manually.

But it was an interesting test anyway. Here's some more sentences that I've picked out, it represents about 50% of the original output. As stated, with either larger input files (my stl maps wouldn't handle more than ~3000 words), or a selected input, the rate of comprahensive sentences would increase.


Snorri, as stated, does not mention him to the first human beings.

Lodur, and Lodur, and Lodur are remarkably passive.

It is accompanied by means of the other gods and Reginsmal, both the the Aesir triad of Hönir and Loki. 

Here we find no part of a third god, sometimes Loki, the fact that the Poetic Edda and the problem. 

Lodur, and a man called Embla. 

In Gylfaginning ch.8 he often use several names. 

We must begin to dismiss all de Vries 1933 p.49.

The origin of the the other Aesir can kill him. 

She also demands satisfaction for her father´s death, and Loki and Lodur, and the first human beings.

Hence it illustrates the problem.

He keeps his promise to Jotunheim. 

It is the Scandinavian tradition has several motives in the creation of mankind in the problem.

Therefore we must conclude that we are one of others.

-Luctus

Statisticly seen, most things happens to other people.
[Mail]

[edited by - Luctus on March 11, 2004 3:20:41 PM]

Topic Locked

This topic has been locked by a moderator. New replies are not allowed.