File size: 3,403 Bytes
fa24833
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
---
license: cc-by-4.0
datasets:
- DSL-13-SRMAP/Telugu-Dataset
language:
- te
tags:
- sentiment-analysis
- text-classification
- telugu
- telugu-specific
- bert
- rationale-supervision
- explainable-ai
base_model: l3cube-pune/telugu-bert
pipeline_tag: text-classification
metrics:
- accuracy
- f1
- auroc
---

# Telugu-BERT_WR

## Model Description

**Telugu-BERT_WR** is a Telugu sentiment classification model built on **Telugu-BERT (L3Cube-Telugu-BERT)**, a Transformer-based BERT model pretrained **exclusively on Telugu text** by the L3Cube Pune research group.

The base model is pretrained on **Telugu OSCAR**, **Wikipedia**, and **news corpora** using the **Masked Language Modeling (MLM)** objective. Being tailored specifically for Telugu, Telugu-BERT captures **language-specific vocabulary, syntax, semantics, and idiomatic expressions** more effectively than multilingual models such as mBERT and XLM-R.

The suffix **WR** denotes **With Rationale supervision**. This model is fine-tuned using both **sentiment labels and human-annotated rationales**, enabling stronger alignment between predictions and human-identified evidence.

---

## Pretraining Details

- **Pretraining corpora:**
  - Telugu OSCAR
  - Telugu Wikipedia
  - Telugu news data
- **Training objective:**
  - Masked Language Modeling (MLM)
- **Language coverage:** Telugu only

---

## Training Data

- **Fine-tuning dataset:** Telugu-Dataset
- **Task:** Sentiment classification
- **Supervision type:** Label + rationale supervision
- **Rationales:** Token-level human-annotated evidence spans

---

## Rationale Supervision

During fine-tuning, **human-provided rationales** are incorporated alongside sentiment labels. In addition to the standard classification loss, an **auxiliary rationale loss** guides the model to align its attention or explanation scores with annotated rationale tokens.

This supervision improves:

- Interpretability of sentiment predictions
- Alignment between model explanations and human judgment
- Plausibility of generated explanations

---

## Intended Use

This model is intended for:

- Explainable Telugu sentiment classification
- Rationale-supervised learning experiments
- Monolingual Telugu NLP research
- Comparative evaluation against label-only (WOR) baselines

Telugu-BERT_WR is particularly suitable for **pure Telugu text analysis** when sufficient labeled data and human rationales are available.

---

## Performance Characteristics

Rationale supervision enhances **explanation quality and human alignment**, while preserving the strong sentiment classification capability of Telugu-BERT.

### Strengths

- Deep understanding of Telugu vocabulary and syntax
- Superior handling of nuanced and idiomatic sentiment expressions
- Human-aligned explanations through rationale supervision

### Limitations

- Not designed for cross-lingual or multilingual tasks
- Requires annotated rationales, increasing annotation cost
- Performance depends on availability of sufficient Telugu training data

---

## Use in Explainability Evaluation

**Telugu-BERT_WR** is well-suited for evaluation with explanation frameworks such as FERRET, enabling:

- **Faithfulness evaluation:** How well explanations support the model’s predictions
- **Plausibility evaluation:** How closely explanations align with human rationales

---

## References

- Joshi et al. (2022). Telugu-BERT. EMNLP.