Tools for Preparing Text for Tokenizers.

piecemaker

Tokenizers break text into pieces that are more usable by machine learning models. While writing wordpiece and morphemepiece, we found that many steps were shared between those packages. This package provides those shared steps.

Installation

You can install the released version of piecemaker from CRAN with:

install.packages("piecemaker")

And the development version from GitHub with:

# install.packages("devtools")
devtools::install_github("macmillancontentscience/piecemaker")

Example

{piecemaker} helps to prepare text for tokenization. For example, it can help you clean out strange encoding, whitespace, and special characters.

library(piecemaker)

piece1 <- " This is a    \n\nfa\xE7ile\n\n    example.\n"
# Specify encoding so this example behaves the same on all systems.
Encoding(piece1) <- "latin1"
example_text <- paste(
  piece1,
  "It has the bell character, \a, and the replacement character,",
  intToUtf8(65533)
)
prepare_text(example_text)
#> [1] "This is a facile example . It has the bell character , , and the replacement character ,"
prepare_text(example_text, squish_whitespace = FALSE)
#> [1] " This is a    facile    example .  It has the bell character ,   ,  and the replacement character ,  "
prepare_text(example_text, remove_control_characters = FALSE)
#> [1] "This is a facile example . It has the bell character , \a , and the replacement character ,"
prepare_text(example_text, remove_replacement_characters = FALSE)
#> [1] "This is a facile example . It has the bell character , , and the replacement character , �"
prepare_text(example_text, remove_diacritics = FALSE)
#> [1] "This is a façile example . It has the bell character , , and the replacement character ,"

Code of Conduct

Please note that the piecemaker project is released with a Contributor Code of Conduct. By contributing to this project, you agree to abide by its terms.

Disclaimer

This is not an officially supported Macmillan Learning product.

Contact information

Questions or comments should be directed to Jonathan Bratt ([email protected]) and Jon Harmon ([email protected]).