Author: Taylor B. Arnold
License: LGPL-2
The cleanNLP package is designed to make it as painless as possible
to turn raw text into feature-rich data frames.
A minimal working example of using cleanNLP consists of loading the
package, setting up the NLP backend, initializing the backend, and running
the function cnlp_annotate. The output is given as a list of data frame
objects. Here is an example using the udpipe backend:
library(cleanNLP)
cnlp_init_udpipe()
annotation <- cnlp_annotate(input = c(
"Here is the first text. It is short.",
"Here's the second. It is short too!",
"The third text is the shortest."
))
annotation
$token
# A tibble: 27 x 11
doc_id sid tid token lemma space_after upos xpos feats tid_source
* <int> <int> <chr> <chr> <chr> <chr> <chr> <chr> <chr> <chr>
1 1 1 1 Here here "\\s" ADV RB Pron… 0
2 1 1 2 is be "\\s" AUX VBZ Mood… 1
3 1 1 3 the the "\\s" DET DT Defi… 5
4 1 1 4 first first "\\s" ADJ JJ Degr… 5
5 1 1 5 text text No NOUN NN Numb… 1
6 1 1 6 . . "\\s" PUNCT . NA 1
7 1 2 1 It it "\\s" PRON PRP Case… 3
8 1 2 2 is be "\\s" AUX VBZ Mood… 3
9 1 2 3 short short No ADJ JJ Degr… 0
10 1 2 4 . . "\\n" PUNCT . NA 3
# … with 17 more rows, and 1 more variable: relation <chr>
$document
doc_id
1 1
2 2
3 3
The token output table breaks the text into tokens, provides lemmatized
forms of the words, part of speech tags, and dependency relationships. Two
short case-studies are linked to from the repository to show sample usage of
the library:
Please see the notes below, and the official package documentation on CRAN, for more options to control the way that text is parsed.
You can download the package from within R directly from CRAN:
install.packages("cleanNLP")
After installation, you should be able to use the udpipe backend (as used
the minimal example and case-studies above; model files will be installed
automatically) or the stringi backend without any additional setup. For most
users, we find that these out-of-the-box solutions are a good starting point.
In order to use the two Python backends, you must install the associated
cleannlp python module. We recommend and support the Python 3.7 version of
Anaconda Python.
After obtaining Python, install the module by running pip in a terminal:
pip install cleannlp
Once installed, running the respective backend initialization functions will provide further instructions for download the required models.
There have been numerous changes to the package in the newly released version 3.0.0. These changes, while requiring some changes to existing code, have been carefully designed to make the package easier to both install and use. If you are running into any issues with the package, first make sure you are using updated materials (mostly available from links within this repository).
The cleanNLP package is designed to allow users to make use of various NLP annotation algorithms without having to worry (too much) about the output format, which is standardizes at best as possible. There are four backends currently available, each with their own pros and cons. They are:
- stringi: a fast parser that only requires the stringi package, but produces only tokenized text
- udpipe: a parser with no external dependencies that produces tokens, lemmas, part of speech tags, and dependency relationships. The recommended starting point given its balance between ease of use and functionality. It also supports the widest range of natural languages.
- spacy: based on the Python library, a more feature complete parser that included named entity recognition and word embeddings. It does require a working Python installation and some other set-up. Recommended for users who are familiar with Python or plan to make heavy use of the package.
- corenlp: another Python library (formally Java) that is an official port of the Java library of the same name.
The second two backends (spacy and corenlp) require some additional setup, namely installing Python and the associated Python library, as documented above. To select the desired backend, simply initialize the model prior to running the annotation.
cnlp_init_stringi(locale="en_GB")
cnlp_init_udpipe(model_name="english")
cnlp_init_spacy(model_name="en")
cnlp_init_corenlp(lang="en")
The code above explicitly sets the default/English model. You can use a different model/language when starting the model. For udpipe the models will be downloaded automatically. For spacy and coreNLP the following helper functions are available:
cnlp_download_spacy(model_name="en")
cnlp_download_corenlp(lang="en")
Simply change the model name or language code to download alternative models.
If you make use of the toolkit in your work, please cite the following paper.
@article{,
title = "A Tidy Data Model for Natural Language Processing Using cleanNLP",
author = "Arnold, Taylor B",
journal = "R Journal",
volume = "9",
number = "2",
year = "2017"
}
Please, however, note that the library has evolved since the paper was published. For specific help with the package's API please check the updated documents linked to from this site.
Please note that this project is released with a Contributor Code of Conduct. By participating in this project you agree to abide by its terms.