Corpus Dashboard
Overall status of the Tamil learner corpus.
Learner levels errors per 100 words, coloured by group
Error types Tamil taxonomy
Corpus coverage texts per level and task
Recently saved
Upload Learner Material
Build a corpus record through metadata, OCR/HTR, verification and annotation.
1. Script and details
Only the fields marked * are needed to start. Everything else can be added later, even after saving.
Learner script *
About the script
Optional details — useful for analysis, can be added later
2. OCR / Handwriting recognition
The document is processed one page at a time. Completed pages become available for transcription review immediately while later pages continue processing. Two handwriting recognisers (Sarvam and Google Vision) read every line; where they differ, you choose in the next step.
Source preview
Recognition result
Developer fallback
3. Transcription Review
Clear OCR (both engines agree, good confidence) goes straight into the transcript. Review only the flagged words and regions. Correct what the OCR misread; keep the learner's own mistakes. Those are annotated in Stage 4.
Source preview
Verified learner text
4. Error Detection & Annotation
Each underlying error appears once, with every detector's evidence. Types use the same Tamil taxonomy for automatic and manual tags. Reject as Not an error or OCR error so transcription mistakes never count as learner errors.
5. Check and save the record
A last look before saving. You can still go back and change anything, and saving again later updates the same record.
Record saved
Records saved on this computer
Analyze
Search and investigate learner texts directly.
Insights
Patterns, trends, comparisons and statistics derived from the corpus.