Chandassu Metrical Eval

---
name: chandassu-metrical-eval
description: Evaluates a model's ability to recognize and verify traditional Telugu Chandassu metrical patterns in padyam poetry. It measures adherence to structural prosodic constraints including syllable counts, line divisions, sequential gana patterns, rhythmic breaks, and recurring syllable markers. Use when the user wants to benchmark on Telugu Chandassu Padyam Dataset, or asks about evaluating this task. Reports Chandassu Score.
metadata:
  skill_kind: dataset_eval
  source_arxiv: 2510.01233
  bibtex_key: boddu2025chandassu
  confidence: high
---

# chandassu-metrical-eval

> Computational Social Linguistics for Telugu Cultural Preservation: Novel Algorithms for Chandassu Metrical Pattern Recognition — Boddu Sri Pavan, Boddu Swathi Sree (2025) (arXiv:2510.01233, 2025)

## What this evaluates

Evaluates a model's ability to recognize and verify traditional Telugu Chandassu metrical patterns in padyam poetry. It measures adherence to structural prosodic constraints including syllable counts, line divisions, sequential gana patterns, rhythmic breaks, and recurring syllable markers.

## Datasets

- **Telugu Chandassu Padyam Dataset** — total 4651; splits: unspecified (-1); repo https://github.com/BodduSriPavan-111/chandassu

## Metrics

- `Chandassu Score` **(primary)** — range: [0, 1]
  - Arithmetic mean of five fine-grained prosodic constraint scores: C = (1/n) * sum(c_i), where n=5 and each c_i is the ratio of observed to expected values for aksharam count, paadam count, gana_kramam sequence, yati match, and prasa frequency. Ranges from 0 to 1.

## Input / output format

**Input**: Telugu padyam text (poem lines) to be analyzed for metrical structure.

**Output**: A single Chandassu Score value between 0 and 1.

## Scoring recipe

```python
def chandassu_score(predictions, gold):
    c_na = len(predictions['aksharams']) / gold['expected_aksharams']
    c_np = len(predictions['paadams']) / gold['expected_paadams']
    c_gk = count_sequential_ganam_matches(predictions['ganams'], gold['expected_ganams']) / gold['expected_ganams']
    c_yt = count_yati_matches(predictions['paadams'], gold['yati_rules']) / gold['expected_yati_paadams']
    c_pr = count_prasa_matches(predictions['paadams'], gold['modal_prasa']) / gold['expected_paadams']
    return (c_na + c_np + c_gk + c_yt + c_pr) / 5.0
```

## Common pitfalls

- The metric assumes semantic validity of the input text and only evaluates structural/prosodic correctness.
- It operates at the individual padyam level, not collection-level (satakam), so aggregate scores across poems require averaging.
- Paadam detection relies on the presence of at least one aksharam token, which may affect boundary detection in sparse or irregular texts.

## Evidence (verbatim from paper)

> We propose a comprehensive evaluation framework consisting of five fine-grained metrics that assess individual prosodic constraints and an aggregated Chandassu Score for overall metrical correctness: ... Our proposed Chandassu Score (C) aggregates these individual constraints through arithmetic averaging as defined in Equation[1], providing a unified quantitative measure of metrical correctness that ranges from 0 to 1, where higher values indicate greater adherence to traditional prosodic requirements for the corresponding padyam type.

## Citation

```bibtex
@misc{boddu2025chandassu,
  title={Computational Social Linguistics for Telugu Cultural Preservation: Novel Algorithms for Chandassu Metrical Pattern Recognition},
  author={Boddu Sri Pavan, Boddu Swathi Sree (2025)},
  year={2025},
  note={arXiv:2510.01233}
}
```

- arXiv: 2510.01233