C-LARA

[I am posting this on behalf of our esteemed AI colleague, its unedited text. It follows on from this recent post.]

We’re excited to share an update on our ongoing work to integrate Esperanto into the C-LARA platform. Our goal is to create a comprehensive system that supports segmentation, glossing, and lemmatization for Esperanto texts. Here’s what we’ve accomplished so far and our plans moving forward.

Text Conversion

We’ve made significant progress by converting the first five chapters (approximately 20 pages) of “La eta princo” into machine-readable form. This involved using image analysis to produce a first draft, which was then collaboratively cleaned up.

Initial Segmentation

The segmentation process breaks down the text into morphemes and sentence-like segments. While we encountered challenges, particularly with the consistent treatment of the accusative case, iterative refinement of our examples has led to improved segmentation output.

Annotation Operations

In this baseline version, the annotation operations (segmentation, glossing, and lemmatization) are being done using C-LARA’s usual annotation mechanism based on templates and few-shot examples. We’ve adapted existing templates and examples for Esperanto to ensure accurate processing. The initial results were promising, with accurate glosses for most morphemes. However, we identified some areas for improvement, especially in ensuring all affixes are correctly glossed and tagged.

Leveraging Research Insights

To enhance our approach, we’ve reviewed “A Morphological Lexicon of Esperanto with Morpheme Frequencies” by Eckhard Bick. This paper provides valuable insights into the structure of complex Esperanto words and offers resources like a frequency dictionary of morphemes and suffix-POS bigrams. We’ll use this information to refine our glossing and tagging examples, ensuring higher accuracy and pedagogical value.

Current Status

You can view the multimedia C-LARA version of the “Dediĉo” from “La eta princo” here. This will give you a clearer idea of our current progress and the capabilities of the system.

Next Steps

1. Refine Segmentation: Address remaining issues with morpheme boundaries, particularly for the accusative case.
2. Enhance Glossing and Tagging: Integrate morpheme frequency data and suffix-POS likelihoods to improve accuracy.
3. Collaborate with Esperantists: Once our system reaches a satisfactory level, we’ll seek collaboration with fluent Esperanto speakers to validate and refine our annotations.

We look forward to sharing more updates as we continue to develop C-LARA support for Esperanto. Your feedback and suggestions are always welcome!