edaiofficial's picture
initial commits
78aa4ee

ENGLISH TO SWAHILI

Author: Sackey Freshia

DATA

SAWA corpus - A parallel English-Kiswahili corpus (Guy De Pauw, Peter Waiganjo Wagacha, Gilles-Maurice de Schryver) Link to corpus details - https://www.aclweb.org/anthology/W09-0702/

Corpus details: English words: 542.1k

Swahili words: 442.9k

English Sentences - 33.6k

Swahili sentences - 33.6k

Model

Default Masakhane transformer model (Using JoeyNMT) Link to model:

https://drive.google.com/open?id=1QRltbbrPCR1r5_8bDrB5KzZr3KjAxiyT

Analysis

Example 1

Source:     span
Reference:  shibiri
Hypothesis: kipingo

Example 2


Source:     administer
Reference:  amuru
Hypothesis: msimamizi

Example 3

Source:     Then as to Samood, they were destroyed by an excessively severe punishment.
Reference:  Basi Thamudi waliangamizwa kwa balaa kubwa mno.
Hypothesis: Basi walipo angamizwa kwa Haki waliangamizwa kwa adhabu ya Moto ulio dhaahiri.

Example 4

Source:     You do not believe but say,
Reference:  Hamjaamini, lakini semeni:
Hypothesis: Hauamini ila husema:

RESULTS

  • BLEU train: 15.60
  • BLEU dev: 15.00
  • BLEU test: 3.60