CISC181 S2017 Lab8

From class_wiki
Jump to: navigation, search

Lab #8 grading


  • Make a new project with n = 8 (following these instructions)
  • Name your main class "Lab8" (when creating a new module in the instructions above, in the Java class name field)
  • Modify by adding your name and section number in a comment before the Lab8 class body.


In this lab you will analyze text files by breaking them into n-grams at the character level, and use those n-grams to generate random text in the same "style" (in a statistical sense). An n-gram is a sequence of n consecutive characters from the input. The complete set of n-grams for a text overlap each other--for example, if the text is "woodchucks", the 3-grams are "woo", "ood", "odc", "dch", "chu", "huc", "uck", and "cks".

Furthermore, we can keep track of what characters follow each n-gram. For example, if the text is "the three pirates ate their pie", the 2-grams and a list of the characters following them are shown below:

2-gram Characters after 2-gram Characters after
"th" "e", "r", "e" "ra" "t"
"he" " ", "I" "at" "e", "e"
"e " "t", "p", "t" "te" "s", " "
" t" "h", "h" "es" " "
"hr" "e" "s " "a"
"re" "e" " a" "t"
"ee" " " "ei" "r"
" p" "i", "i" "r " "p"
"pi" "r", "e" "ie" null
"ir" "a", " "

Note that non-alphabetic characters are also recorded: spaces, punctuation, digits, and so on. However, we will ignore capitalization.

Now consider how you might generate a new random text with the same statistics as the one you analyzed. Start with a "seed" n-gram chosen randomly from the text. Suppose "th" is chosen for the 2-gram pirate example. This will be the beginning of your output.

The next character output is chosen randomly from the list associated with "th": "e" is chosen with a 2/3 chance and "r" with a 1/3 chance. Suppose an "e" is picked. The output is now "the".

Now we drop the first character "t" from the last n-gram (the seed) that we were using and append the new character "e" to get our new seed "he". We select a character randomly from the list associated with "he": " " (space) with 1/2 chance and "i" with 1/2 chance. Suppose we choose "i". The output is now "thei".

Update the seed again; now we have "ei". There is only one character, "r", in the list associated with this 2-gram, so we pick it. The output is now "their".

Now the seed is "ir". " " or "a" is chosen with equal probability. Suppose "a" is chosen. Now the output is "theira" and the seed is "ra".

And so on. If your program ever gets into a situation in which there are no characters to choose from (which can happen if the only occurrence of the current seed is at the exact end of the source), pick a new random seed and continue.


You are to implement a Java public class RandomWriter that provides a random writing application. Your class should have a two-argument constructor that takes:

  • String source: The name of an input file to read and analyze
  • int n: A non-negative number indicating the length of each "gram," or character sequence, to break the file into

and also a method generateText() that takes the following two parameters:

  • int length: A non-negative number of characters to generate.
  • String result: The name of the output file

Some kind of map is the recommended data structure to store your n-grams and their character list associations.


In main(), run your code on the following files:

Generate 500 characters of text for each input. Print the text in reasonable length lines, breaking only at spaces (not in the middle of a word). Do this for 1-grams, 2-grams, 4-grams, and 6-grams.


Submit your to Sakai, as well as a text file results.txt containing the outputs of your program for the different input files and n-gram lengths. Inside the results.txt, clearly label what the source file and value of n was for each block of output text (there should be 3 input files x 4 values of n = 12 such blocks). Put your name in both files.


This assignment is adapted from one created by David Matuszek at the University of Pennsylvania and Joe Zachary's random writer assignment.