2/2 Markdown rendering everywhere: lessons, questions, hints and menus are now displayed with MD formatting, making everything easier to read in the REPL.

The built-in Julia Basics course has been adapted to use these new features.
Check it out in action here: youtu.be/aEboVhipEfg

#JuliaLang

comparison.parameters → tunable parameters used for the building of the "comparison" object.

comparison.effect_sizes → pairwise differences & Cohen’s d

comparison.summary → number of collocates & mean scores

comparison.results → collocates & scores per year (see the table below for the year 1814)

What this call does
1️⃣ Splits the corpus into subcorpora by :year.
2️⃣ Analyzes each subcorpus using LLR to compute collocates of node_word.
3️⃣ Aggregates everything into structured outputs👇

#JuliaLang #CorpusLinguistics #AssociationAnalysis #TextAssociations.jl #NLP #DigitalHumanities

See atantos.github.io/TextAssociat... for the whole data analysis scenario and the whole output.

#JuliaLang #CorpusLinguistics #AssociationAnalysis #TextAssociations.jl #NLP #DigitalHumanities

julia>comparison = compare_subcorpora(
corpus_df,
:year, # Split corpus by year
node_word, # Word to analyze
LLR; # Association (Log-Likelihood Ratio)
windowsize = 5,
minfreq = 1
)
#JuliaLang #CorpusLinguistics #AssociationAnalysis #TextAssociations.jl #NLP #DigitalHumanities

And here is the resulting DataFrame
#JuliaLang #TextAssociations.jl

#JuliaLang

The package is now publicly available as an early release (not yet registered in the Julia package registry), and you can already explore its functionality on GitHub:
🔗 github.com/atantos/Text...

Feedback, ideas, and experiments are welcome. 7/7

Many well-known and lesser-known association measures, together with tools for collocation network analysis, are now coming together under one roof complemented by powerful corpus-based functionality. 6/n

revealing what is salient within a given dataset, discourse, or community.

A new #JuliaLang package, TextAssociations.jl, is taking shape, laying the groundwork for exploring how words connect and co-occur in a transparent, data-driven way. 5/n

under one roof, alongside corpus-based functionality.

While embeddings and RAG systems operate within pretrained semantic spaces, association measures remain essential for interpretable, corpus-based insights. They quantify how often words co-occur beyond what chance would predict, 1/4

A new Julia package called TextAssociations is taking shape, building a foundation for studying how words connect and co-occur in a transparent, data-driven way.
Many familiar and less familiar association measures as well as the building collocation network analysis are finally coming together 3/n

While embeddings and RAG systems operate on pretrained semantic spaces, association measures remain essential for interpretable, corpus-based evidence. They quantify how often words co-occur compared to chance, showing what is salient within a specific dataset or community. 2/n

Thank you for your kind words!