I made this application while taking Spanish lessons to find texts that used verb forms I was learning (present, past, future, subjunctive, etc.). First, I scraped ~10,000 Spanish-language poems from the web that are in the public domain. Then, I built a conjugation dictionary for the 550+ most common verbs, using rules to generate regular verbs, and scraping irregular verbs from the web. From this alone, you can identify which poems contain the most type of verbs you're trying to target. To take it further, I used a k-means clustering algorithm in Python to classify the poems into four difficulty levels based on verb form frequencies (poems with lots of verbs in subjunctive future perfect are more difficult than only present tense, for example).

I first completed the project in December 2024, almost entirely by hand, using LLMs primarily to make an interactive Plotly Dash application. (I have a lot of experience with data visualization, especially ggplot2; that was my first time using LLMs to write in a new domain.) Last week, in one hour with Claude, I completely updated the front-end to Javascript and Plotly.js.

I hope someone can find it useful. It might be the only way to answer questions like, "Which 19th-century Colombian poet used the most future tense verbs?" (Depending on the available corpus, of course.)

If I were starting this from scratch today, I would use LLMs to generate the verb forms instead of creating the dictionary that I did. But I would still have the analysis of the poems be deterministic. I also would use LLMs to assess themes and topics as an additional filter. And, of course, extend to more languages :)