The Phrases and Pronunciation

My final poetry-generating Python program is based on noun phrases, a markov chain, a recursive function, and a loose poetic structure. The vocabulary was sourced from three public domain source texts:

The Sailor’s Word Book, An Alphabetical Digest Of Nautical Terms, Including Some More Especially Military And Scientific, But Useful To Seamen; As Well As Archaisms Of Early Voyagers, Etc. By The Late Admiral W. H. Smyth (1867)

Emily Post’s Etiquette: In Society, In Business, In Politics And At Home (1922)

Fyodor Dostoevsky’s Notes From Underground (1864)

These texts make for interesting juxtapositions. For one thing, Post describes good manners for society, while Dostoevsky’s narrator is an outcast from society. The Sailor’s Word Book is a dictionary, but its definitions of sailor’s words are the most poetic of my three source texts—especially when taken out of context. The admiral who wrote this book in 1867 refers to ships as female (i.e. “her head”) and officers as male (i.e. “steersman”) which makes everything sound like an analogy. Likewise, Post and Dostoevsky use plenty of gendered pronouns. They also use proper nouns like “Mrs. Worldly” and “Simonov.” These texts refer to people—whether they are objects (such as a ship), individuals or society as a whole. As a result, when I feed these three texts into my program, they often seem to be talking about the same thing.



My Markov object creates a dictionary of probabilities for transitioning from one word to the next, but I added some special features. Its “generate” function takes a starting word as an argument. If the starting word exists in the source text (Sailor’s Words + Dostoevsky), it will generate text based on its dictionary of probable transitions from that word to any other word.

  # Generate a text from the information in self.ngrams
  def generate(self, start_word):
    from random import choice

    # If the start_word is in the text, generate a markov chain
    if start_word in self.all_words:
      current = choice([item for item in self.ngrams if item[0] == start_word])
      output = list(current)

      for i in range(self.max):
        if current in self.ngrams:
          possible_next = self.ngrams[current]
          next = choice(possible_next)
          current = tuple(output[-self.n:])

      output_str = self.concatenate(output)
      return output_str
      return ''

This allowed me to create a generate_question() function that starts a Markov chain with What When Where Why or How. The questions are part of my poetic structure.

def generate_question():
    next_line = underground.generate(random.choice(['What', 'How', 'Why', 'When', 'Where']))
    return next_line

I felt it was important to feed sentences into the Markov chain, rather than lines. So I made an array of Strings and used TextBlob to parse the sentences which I fed into my Markov chain like this:

underground = MarkovGenerator(5, 6)
utext = []
for line in open('../texts/notesfromunderground.txt', 'r'):
    utext.append(line + ' ')
utext_sentences = TextBlob(" ".join(utext).decode('ascii', errors='replace')).sentences
for sentence in utext_sentences:



I used TextBlob to generate a list of noun phrases from Etiquette and Sailor’s Words.

Noun phrases are a very interesting way to parse a text. In fact, I could have stopped there and made a poem out of noun phrases like these:

dead men 's dust
vulgar salutation
long lewd lither loiterer
moral elevation ensues
float of a fishing-line
conical tops of mountains

This is a series of noun phrases from Sailor’s Words:

barrel of small arms.
barrel of the wheel.
barrier of ice. ice
olive oil. _yellow basilicon_
battening the hatches. securing
battens of the hatches.
beam of the anchor. synonymous
land 's end bore
before or abaft the beam.
bend on the tack.
berthing of the head.
beset in ice. surrounded

This is a series of noun phrases from Etiquette:

politeness implies
gentlemanly character
social intercourse
whole detail
person shows respect
time implies
reciprocal respect
gentlemanly character
womanhood combined.

Originally, I created a Noun Phrase List object, but this was time consuming to generate every time, so I just saved them as a text file using the NPLC object I created in my nplc module.

post = NPLC('../texts/sailorswordbook.txt', 4)
npz = open("sailor_nouns.txt", "wb")
for i in post.noun_phrases:
    npz.write(i.encode('utf-8', errors='replace').strip() + '\n')


I thought that these noun phrases could give my poems some rhythmic structure, especially if repeated. I sorted them by length:

# generate three noun phrases
nOne = random.choice(postnouns).strip()
nTwo = random.choice(postnouns).strip()
nThree = random.choice(sailornouns).strip()

print "nOne: " + nOne + ', nTwo: ' + nTwo + ', and Three: ' + nThree

# sort the noun phrases by length
if len(nOne) > len(nTwo):
    nX = nOne
    nOne = nTwo
    nTwo = nX

if len(nTwo) > len(nThree):
    nX = nTwo
    nTwo = nThree
    nThree = nX

I also created a function to prepend an appropriate preposition to a noun phrase

# This function tries to figures out what preposition to put before
# a Noun Phrase based on whether it is plural and based on its starting letter.
def preposition(line):
    first_letter = line[0]
    for word in line.split():
        tb = TextBlob(word)
        for w, t in tb.tags:
            if t == 'NN':
                b = Word(word)
                if word == str(b.singularize()):
                    # print word + " is probably singular like " + b.singularize()
                    if not_a_vowel(first_letter):
                        return random.choice(['The ', 'A ', '']) + line
                        return random.choice(['The ', 'An ', '']) + line
                elif word == str(b.pluralize()):
                    return random.choice(['The ', 'Some ', 'Many ', 'Of ', 'For all of the ', '']) + line
    ## if it gets to this point, we dont know if it is plural, so just figure out if 'a' or 'an'
    if not_a_vowel(first_letter):
        return random.choice(['A ', 'The ', '']) + line
        return random.choice(['An ', 'The ', '']) + line

That uses my not_a_vowel function:

def not_a_vowel(letter):
    for char_to_check in ["a", "e", "i", "o", "u"]:
        if letter == char_to_check:
            return False
    return True

Which I originally wrote in order to approximate the number of syllables in a gibberish word. I did not deal with the complexities of “Y.”


The noun phrases, questions and markov chains help give my poems rhythm and space. I pick up the pace with a recursive function called redo_line:

## This function takes a word an changes it to another word that rhymes or has the same # of syllables.
## Sometimes, it also appends a random nounphrase
## rhyming_word is in a separate file, the rhymebot module.
def changeWord(word):
    new_word = ''
    #   nounize(word)
    if word.upper() in all_words:
        new_word = rhyming_word(word, 1)
        if new_word:
            return new_word.lower()
            new_word = str(random.choice(syl_lookup[syl_bible[word.upper()]]))
            # new_line = new_line + " " + new_word.lower()
        gibSyl = gib_syls(word)
        new_word = str(random.choice(syl_lookup[gibSyl]))
        # new_line = new_line + " " +
    return new_word.lower() + " " + random.choice(postnouns + sailornouns).strip()

## This function iterates recursively through a line, three words at a time.
## It transforms the last word each time by calling the changeWord function.
def redo_line(line):
    words = line.split()
    if len(words) > 0:
        w = words.pop()
        y = changeWord(w)
        if len(words) > 3:  #crucial variable
            new_line = ' '.join(words)
                print ' '.join(new_line.split()[-3:]) + " " +y.lower()
                print ' '.join(new_line.split()[-2:])
        if len(words) == 3:
            new_line = ' '.join(words)
                print ' '.join(new_line.split()[-2:]) + " " +y.lower()
                print ' '.join(new_line.split()[-1:])
        if len(words) == 2:
            new_line = ' '.join(words)
                print ' '.join(new_line.split()[-1:]) + " " +y.lower()
                print ' '.join(new_line.split()[-1:])
        if len(words) == 1:
            print words[0] + ' ' + y
            print underground.generate(words[0])
            print underground.generate(words[0])
            print words[0] + w

I originally developed this function for my midterm project. Here, I use it towards the end of the poem to iterate through the poem’s three noun phrases and the poem’s opening line.

You can view the full code on my github and the two sample poems from my reading are excerpted below:


young people
How is enjoyment in this to be
An elderly guest
phrases and pronunciation
phrases and pronunciation
pronunciation phrases and gillon
and pronunciation phrases remodeled
phrases and pronunciation Townsend's
guest phrases and catamaran
guest phrases liquefied
guest amdahl's
elderly simonette
As far as
  my personal opinion is concerned, to care only for well-being seems to
  me positively ill-bred.
time by his sighed perfect little house
that time by kerins
at that time bly
attention at that duhaime
my attention at cooperate
distracted my attention introduced
Apollon distracted my ritalin
that Apollon distractedly
fact, that Apollon hausfeld
in fact, that 'm chapter xxv
thing, in fact, inhibit
good thing, in aux telegraph pole
a good thing, doron
was a good er luncheon menu=
was a wood
was japonica
It chides
It would be to the interests of humanity and courtesy were it made
It was first devised for
  the service of mortars, and named after the inventor, Gomer, in the late
The elderly guest
The phrases and pronunciation
The phrases and pronunciation



A true purpose
A social world
A senior officer 's signature

A fore-and-aft sail, setting with a boom and gaff, in
  ships, synonymous with the spencers of brigs and schooners, and the
  spanker or driver of ships.

The true purpose
The social world
The senior officer 's signature

In size, as a ship's boat, smaller than
  the barge, and, like it, carvel-built.
true purpose
d��butante wears
The senior officer 's signature

true one--arising from lateral pressure and the effect of sea when
purpose of firing signals,
  as the officer who commands her is particularly ordered to carry no
  cargo, ammunition, or implements of war.

Why are we perverse and ask for something
ask for something fu dining-room service.
and ask for fudging
perverse and ask reyer
we perverse and hunk
are we perverse duguid?
are we britz
are indri
Why moondreamer?
Why did I invite myself to this dinner
Why are you so
A true purpose
social world
A senior officer 's signature

One thought on “The Phrases and Pronunciation

Leave a Reply

Your email address will not be published. Required fields are marked *