Skip to content

the maddening thoughtless people

December 31, 2010

technological progress in the following categories
facilities
equipment
the organizations provided
construction of repression
with the object of salaries and organizational affiliations who seemed deadly

criminals ignored the travel in his head towards the very queer
or had become sane
conformance with tears into losing all
exactly what the decision
to become erroneous
tiny helpless in the darkness it preaches technological progress by producing
uncommon nonsense
filled up in records and staggering feats of historical figures
recognized
and a great curiosity and beautiful garden
the feeble wail his identity and understanding within a bud
anywhere except the reality

and a single thing’s crooked
going to fabrication
falsification
a million people had done the dangers
the moment would result

 
 

December 26, 28 2010, supervised generation on a bigram language model using part A of Method 770ac8b0-5626-426c-bb79-1adf9ad13324 (human accomodating increasing levels of machine randomness). Corpus: Through the IngsocGrant (Orwell 1984, Carroll Wonderland and Looking Glass, NSF GPG). Generator: ePoGeeS.

Basically I tried to use method Method 770ac8b0-5626-426c-bb79-1adf9ad13324 (human accomodating increasing levels of machine randomness) on the Through the IngsocGrant corpus, but I really didn’t like the results, so I just stuck with what I had after the first part of the method. The output of the full method is shown below. I think it was just a bit too incoherent… not sure why; I think the corpus did not lend itself to coherence on fully-automated generation. I wish I knew how to quantify that… I’ve been thinking of empirical evaluations recently. There was this great paper that came out this year, “Automatic Analysis of Rhythmic Poetry with Applications to Generation and Translation“, (E. Greene, T. Bodrumlu, and K. Knight), where they used chained Finite State Transducers to build language models and generate, but they once again raised the issue of evaluation. Unsurprising, given that the paper was at EMNLP, known for requiring strict evaluations. Anyway, I’m thinking of doing random sampling on automated generation to build a mini-corpus, then having the corpora of various generators rated analytically by people on Mechanical Turk – there have got to be some starving English Lit grad students on there who could do the job. (If nothing else, it’d be an interesting use of Mechanical Turk, there was even the “NAACL HLT 2010 Workshop on Creating Speech and Language Data with Amazon’s Mechanical Turk” to give it a bit of respectability.) All right, here was the wack output:

 
 

technological progress in the following categories
facilities
equipment
the organizations provided
construction of repression
with the object of salaries and organizational affiliations who seemed deadly

then the low level because the name is
a physical facility including the reason that
habitual style of the typefaces identified
fragments even conceivable that the interminable
certification required proposal the preceding three
whether it’s full proposal preparation functions

criminals ignored the travel in his head towards the very queer
or had become sane
conformance with tears into losing all
exactly what the decision
to become erroneous
tiny helpless in the darkness it preaches technological progress by producing
uncommon nonsense
filled up in records and staggering feats of historical figures
recognized
and a great curiosity and beautiful garden
the feeble wail his identity and understanding within a bud
anywhere except the reality

nothing just beginning the king interrupted
before the solicitation and solemnly rising but
recommended for participation in addition if
all the essential point the completed the maddening
thoughtless people than the more people
the second stage the same and went

and a single thing’s crooked
going to fabrication
falsification
a million people had done the dangers
the moment would result

 
 

December 26 2010, supervised generation on a bigram language model using Method 770ac8b0-5626-426c-bb79-1adf9ad13324 (human accomodating increasing levels of machine randomness). Corpus: Through the IngsocGrant (Orwell 1984, Carroll Wonderland and Looking Glass, NSF GPG). Generator: ePoGeeS.

Advertisements
No comments yet

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: