Hybrid attention-based multi-class classification of Ethiopian legal texts using deep learning
Amlakie Aschale Alemu, Malefia Demilie Melese, Daniel Arega Mengesha, Misganaw Aguate Widneh · 2026
Read Hybrid attention-based multi-class classification of Ethiopian legal texts using deep learning by Amlakie Aschale Alemu, Malefia Demilie Melese, Daniel Arega Mengesha, Misganaw Aguate Widneh (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/plos-pone-0348805.txt.
Description
Amlakie Aschale Alemu and colleagues taught deep learning models to sort Ethiopian legal documents into categories automatically, a pressing need as the country's digital legal records multiply. Legal language is dense and context-heavy, so the team layered attention mechanisms, which let a model focus on the words that matter most. These sat on several network designs that read text both as patterns and as sequences.
The strongest design, combining a pattern-reading network with a two-direction sequence reader plus self-attention, classified documents with accuracy above ninety-nine percent under two different testing schemes. The authors report consistent gains from adding attention across all the designs tried. They present the system as a step toward searchable, well-organised digital law for a legal system whose documents are rapidly moving online.
Catalog record
| CREATOR | Amlakie Aschale Alemu, Malefia Demilie Melese, Daniel Arega Mengesha, Misganaw Aguate Widneh |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 47697 |
| LICENSE | CC_BY |
| RIGHTS | CC BY |
| SOURCE | PLOS |