Repository files navigation

An R-package for analyzing natural language with transformers-basedlarge language models. Thetext package is part of theR LanguageAnalysis Suite, including:

• talk - a package that transforms voicerecordings into text, audio features, or embeddings.
  
  
• text - a package that provides tools formany language tasks such as converting digital text into wordembeddings.
  
  talk andtext offer access to Large LanguageModels from Hugging Face.
  
  
• topics a package with tools forvisualizing language patterns into topics.
  
  
• the L-BAM Library a librarythat provides pre-trained models for different psychologicalassessments such as mental health issues, personality and relatedbehaviours.
  
  

TheR Language Analysis Suite is created through a collaborationbetween psychology and computer science to address research needs andensure state-of-the-art techniques. The suite is continuously tested onUbuntu, Mac OS and Windows using the latest stable R version.

Thetext-package has two main objectives:
* First, to serveR-users as apoint solution for transforming text to state-of-the-artword embeddings that are ready to be used for downstream tasks. Thepackage provides a user-friendly link to language models based ontransformers fromHugging Face.
*Second, to serve as anend-to-end solution that providesstate-of-the-art AI techniques tailored for social and behavioralscientists.
Please reference our tutorial article when using thetext package:The text-package: An R-package for Analyzing andVisualizing Human Language Using Natural Language Processing and DeepLearning.

Recent significant advances in NLP research have resulted in improvedrepresentations of human language (i.e., language models). Theselanguage models have produced big performance gains in tasks related tounderstanding human language. Text are making these SOTA models easilyaccessible through an interface toHuggingFace in Python.

Text provides many of the contemporary state-of-the-art languagemodels that are based on deep learning to model word order and context.Multilingual language models can also represent several languages;multilingual BERT comprises104 different languages.

Table 1. Some of the available language models

SeeHuggingFace for a morecomprehensive list of models.

Text also provides functions to analyse the word embeddings withwell-tested machine learning algorithms and statistics. The focus is toanalyze and visualize text, and their relation to other text ornumerical variables. For example, thetextTrain() function is used toexamine how well the word embeddings from a text can predict a numericor categorical variable. Another example is functions plottingstatistically significant words in the word embedding space.