goal: generate a short story of tragic love based on a noun prompt

• code submitted to the Turing Test in Creative Arts hosted by Dartmouth College

• Install: NLTK + WordNet corpus

$ sudo pip install -U nltk

To run :

$ python story.py "dog"

And then the man would say, "The dog recoiled, tail between his legs," which made no sense. Yet she couldn't shake the feeling that they pointed to the box.

Or a noun + adj:

$ python story.py "dirty subway"

A man would say, "“Where’s the subway?” Jerry said." And I would reply, "Lily was in the subway." It was unsettling. Why was I dreaming about a subway? I thought it was because my life seemed so just outside of my reach.

• corpus of 1000 short stories

• generate sentences based on theme and noun prompt using tri-gram Markov Chains

  • 4-grams mimick the corpus text too closely; bi-grams produce jumbled words
  • implementation based on Gist: https://gist.github.com/dellis23/6174914

• Each action (see below) is implemented in action[3].py

• Story is assembled by story.py by the Story class

  • Noun prompt is categorized and parsed for adjectives by Noun class

  • Markov sentences are generated by tools/generated_sentence

• assign gender, age, name for each character
• names based on census names list (randomly assigned)

• create season information and relationship to weather
  • description of setting based on weather
• synonyms are swaped for word variety

EVENTS: actions at places

OPENING: Meeting = Characters + action (“meeting” in this case) + setting -> ACTION 1

• add structural variety + themes
• more nuianced parsing of the input noun (wordnet categories)
• additional strategically placed markov sentences
  • check grammar and word sense