15 unsupervised bigram verses
Thou keep’st me. Love
a devil, happy title do.
Is lust in me, love him as
I see his palate urge; then
like enough am now stand you
and all in love engrafted, hath my flame.
Of thy parts of battering days. Dear religious
love control. Three winters cold valley-fountain of love
why should example where wasteful time decease: when
rocks impregnable are dead and I ensconce me not
be with this be griev’d at first in effect
dear love are you and die: despite of beauty
and this line some instinct the brain inhearse, I hold
out. Sweet ornament, to stay! Never intermix’d? I.
January 30, 2011, unsupervised generation using random sampling of a bigrams language model built on Shakespeare’s Sonnets. generator: ePoGeeS.
I generated these a while back, and every once in a while I’ve been looking them over and just thinking about them for a while. I think I need to post them and move on though.
I just thought I’d generate a bunch of verses and see what I could see. I random-sampled 15 verses from a bigram word model of Da Sonnets;
- the first 5 verses were 2 lines with 5 accented vowels each (looking up vowels fairly naively via cmudict).
- the next 5 verses were 4 lines of 6 accented vowels.
- the next 5 verses were 8 lines of 8 accened vowels.
and I list them here without removing any. (i.e. no editorial function.)
I should start thinking about discourse coherence and textual analysis. Bigrams models have some serious limitations. How to work around them? Learn a discourse structure?
Write the one. Dear friend
and me in our brains beguil’d.
Converted are dead; knowing
thy hours are not.
Self-will’d, rank thoughts, or
vanished out, when their.
Gross body’s end. Dear
to thrust, no painting.
To please him that sweet beauty
show it is thy blind do
will, love. Dear heart’s part to
thee hence as I assure ye.
He in love you made more can
thy own desert a swallow’d bait, and tenure
of thy large privilege; forgoing simple
truth. Dear friend; my abuses reckon.
Up that she wrought thee, wilt
thou canst move, to be near
love, such strength by a dream. Love
for thou lov’st those lips which.
Yet none but shoot not mine
own bright, badges of them told
thou art forced to thee how
careful housewife runs to mow: despite.
Praise cannot provoke him aid, and burn the beauty still
telling what a mightier way; I am gone, therefore
love you with true in thy deep oaths breach
do none, be thou lead thee, thy deep
trenches in jollity, ever that which flies in me. Three
look, which I derive, for well. Dear friend
inheritors of thine ten times refigur’d thee how
have I to this with weary night a story.
Of life, for I have of me! What
the swart-complexion’d night doth stand in all these I am
mortgag’d to greet it thy continual haste. Love
there is thine when tyrants crests and for then
that this thou some in this, to catch her though
my praise; yet not forbidden usury, that I
live in the hand defac’d so ill well thou shouldst
owe. Love I perceive that love, if they.
Looked but do I most loving thee: despite
of sight, when thou steal his beauty of
you is such a twofold truth needs must
ransom all to one foolish heart from what
could see not show my love of eyes
are you are sweetest odours made better for
beauty’s pattern of that beauty o’er-snowed and ruin’d love
and he shall see, though more worthy of your.
Countenance fill’d up all but I not sinful
loving mourners seem love thee frown on me
oppress’d, and foison of shame, of dian’s this I
in me to a death my will not my true
being shall in over-plus; yet him as men ride
that other to steal thy babe chase thee
how do not thy worth do anything, kind
then although my side his rank before was.
Sometimes I don’t really care how good the poems are. I think the joy is just in taking algorithms and applying them to text generation.
I should start reading up on information retrieval and question answering. The query will be a seed text or a semantic frame; the answer will be a poem based on a given corpus.