New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Dreams Before Speeches #160

Open
VincentToups opened this Issue Nov 16, 2015 · 3 comments

Comments

Projects
None yet
4 participants
@VincentToups

VincentToups commented Nov 16, 2015

Hello.

Here is my novel, Dreams Before Speeches.

Here are excerpts:

MR. NEWMAN : I'm afraid my message that night was grim and disturbing. I remember telling you we were in the worst economic mess since the Depression. We're here because humanity refuses to accept that our relationship be guided by reason in everything; that is to say, in good conscience, the satisfaction which we no longer have any Real knowledge of it.

The world exists as the mirror of the Indian philosophers declares, “ It is a narrow-minded and ridiculous thing not to architecture, which admits it merely as extraneous ornament, and could Dispense with it.

We're also gathered here for a special event -- the national funeral for an unknown soldier who will today join the heroes of Lebanon and for the preservation of the sacred fire of liberty “ is “ finally staked on the experiment entrusted to the hands of the people. The Bible tells us there will be no immigration reform without employer sanctions, because it does such violence to the spirit, thwarting the human impulse to create, to enjoy life to the fullest, and to inquire into the cause of mankind and world peace. We can decide the tough issues not by who is right, infinitely surpassing Everything else that exists merely relatively, still remained unknown.

Make no mistake about it, this attack was not just against ourselves or the Republic of Korea -- South Korea -- has offered to permit certain events of the 1988 Olympics to take place to begin with, and this will also show how old our view is, though the mass of philosophical knowledge of the subject, which makes mathematics so difficult. This becomes less in the idyll, still less to reproach him because he is Yet always thrown aside as vanished illusions. We are compelled by the principle of sufficient reason is truth, only as modifications of the actual world; thus according to this doctrine is old : it appears in the one case we find in the Vedas, Purana, poems, myths, legends of their saints, maxims and precepts, (85) we see that several ideas which are different in unessential particulars may be Allowed expressing myself by a metaphor.

@greg-kennedy

This comment has been minimized.

Show comment
Hide comment
@greg-kennedy

greg-kennedy Nov 19, 2015

Huh. This worked really well! I like the output.

greg-kennedy commented Nov 19, 2015

Huh. This worked really well! I like the output.

@ikarth

This comment has been minimized.

Show comment
Hide comment
@ikarth

ikarth Nov 19, 2015

I've moved away from Markov chains for basic text generation this year precisely because of the weaknesses you talk about. The mitigations you've done here do a remarkable job of addressing that.

I also think the framing you used is an important part of it. I happen to think that the importance of framing defining the interpretation of the text is undervalued. (I may have written a paper to this effect, once.)

ikarth commented Nov 19, 2015

I've moved away from Markov chains for basic text generation this year precisely because of the weaknesses you talk about. The mitigations you've done here do a remarkable job of addressing that.

I also think the framing you used is an important part of it. I happen to think that the importance of framing defining the interpretation of the text is undervalued. (I may have written a paper to this effect, once.)

@VincentToups

This comment has been minimized.

Show comment
Hide comment
@VincentToups

VincentToups Nov 19, 2015

Of course text, generally speaking, is all framing. "Tree" contains not
a whiff of "treeness" in itself, instead its "treeness" derives entirely
from an association stored in a human brain. You'd need a much larger
corpus and a much more sophisticated model than a markov chain to extract
the association between "treeness" (how is that even represented) and
"tree".

But there are a few other possibly neat ideas. My spouse suggested training
the model with the last few words plus the last few significant words
(using some definition of significance like tf-idf). This extends the model's
"memory" in a hopefully meaningful way but it also shrinks the meaningful
training data enormously. Its just a hard problem.

Word vectors of the sort provided by Word2Vec or Glove can possibly be used
but again, using semantics really reduces the number of meaningful
datapoints, particularly since Word2Vec vectors are ~100 dimensional
floating point representations and would need to be discretized in some way
for a typical markov model approach. The size of the vector space required
for word vecs itself indicates the problem.

On Thu, Nov 19, 2015 at 10:57 AM, Isaac Karth notifications@github.com
wrote:

I've moved away from Markov chains for basic text generation this year
precisely because of the weaknesses you talk about. The mitigations you've
done here do a remarkable job of addressing that.

I also think the framing you used is an important part of it. I happen to
think that the importance of framing defining the interpretation of the
text is undervalued. (I may have written a paper to this effect, once.)


Reply to this email directly or view it on GitHub
#160 (comment)
.

VincentToups commented Nov 19, 2015

Of course text, generally speaking, is all framing. "Tree" contains not
a whiff of "treeness" in itself, instead its "treeness" derives entirely
from an association stored in a human brain. You'd need a much larger
corpus and a much more sophisticated model than a markov chain to extract
the association between "treeness" (how is that even represented) and
"tree".

But there are a few other possibly neat ideas. My spouse suggested training
the model with the last few words plus the last few significant words
(using some definition of significance like tf-idf). This extends the model's
"memory" in a hopefully meaningful way but it also shrinks the meaningful
training data enormously. Its just a hard problem.

Word vectors of the sort provided by Word2Vec or Glove can possibly be used
but again, using semantics really reduces the number of meaningful
datapoints, particularly since Word2Vec vectors are ~100 dimensional
floating point representations and would need to be discretized in some way
for a typical markov model approach. The size of the vector space required
for word vecs itself indicates the problem.

On Thu, Nov 19, 2015 at 10:57 AM, Isaac Karth notifications@github.com
wrote:

I've moved away from Markov chains for basic text generation this year
precisely because of the weaknesses you talk about. The mitigations you've
done here do a remarkable job of addressing that.

I also think the framing you used is an important part of it. I happen to
think that the importance of framing defining the interpretation of the
text is undervalued. (I may have written a paper to this effect, once.)


Reply to this email directly or view it on GitHub
#160 (comment)
.

@hugovk hugovk referenced this issue Dec 21, 2015

Open

Press Coverage #9

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment