LIMP: Linguistically-Informed Multi-Strategy Prompting for Telugu Multi-Turn Dialogue Generation

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Advanced, quick

Summary

LIMP (Linguistically-Informed Multi-Strategy Prompting) is an inference-time, training-free framework designed for generating contextually coherent multi-turn dialogue in Telugu. It injects expert linguistic and cultural knowledge into prompt structures to resolve challenges like morphologically encoded social hierarchy, strict SOV agglutinative syntax, and Natyashastra rasa theory. The framework was empirically evaluated using two strategies, LIMP-RAW and LIMP-COT, on 10,000 instances from the IndicDialogue Telugu corpus with GEMMA-3-1B-IT. LIMP-COT, a six-stage analytical scaffold, achieved approximately 2x higher morphosyntactic surface fidelity (Jaccard = 0.0436) than LIMP-RAW (Jaccard = 0.0211). Conversely, LIMP-RAW, a dense constraint prompt, demonstrated near-ceiling semantic fidelity (BERTSCORE F1 = 0.9709), outperforming LIMP-COT (0.9637) and SARVAM-1 (0.9680). This indicates a semantic-lexical dissociation, where no single configuration excels across all metrics. Notably, LIMP-COT with a 1B parameter model surpassed SARVAM-1 (2B) in lexical metrics, suggesting structured linguistic scaffolding is more impactful than parametric scale for form-faithful generation.

Key takeaway

For NLP Engineers developing dialogue systems for morphologically rich, low-resource languages like Telugu, you should consider implementing linguistically-informed, multi-strategy prompting. Prioritize structured linguistic scaffolding, such as LIMP-COT's six-stage analytical approach, over simply scaling model parameters to achieve higher morphosyntactic fidelity. Be aware that semantic and morphosyntactic performance may dissociate, requiring separate evaluation metrics to ensure comprehensive quality in your generated outputs.

Key insights

Structured linguistic prompting significantly improves morphosyntactic fidelity in low-resource agglutinative languages like Telugu, often surpassing larger models.

Principles

• Semantic and morphosyntactic fidelity are orthogonal in agglutinative languages.
• Linguistic scaffolding can outweigh parametric scale for form-faithful generation.
• Expert knowledge injection can resolve complex linguistic constraints.

Method

LIMP employs inference-time, training-free prompting. It uses either dense constraint prompts (LIMP-RAW) or a six-stage analytical scaffold (LIMP-COT) grounded in rasa theory and Telugu grammar.

In practice

• Use multi-strategy prompting for complex linguistic constraints.
• Prioritize structured linguistic scaffolding over larger models for form fidelity.
• Evaluate semantic and morphosyntactic fidelity separately.

Topics

• Telugu Language Processing
• Multi-Turn Dialogue Generation
• Linguistically-Informed Prompting
• Agglutinative Language NLP
• GEMMA-3-1B-IT
• Natyashastra Rasa Theory

Best for: Research Scientist, AI Scientist, NLP Engineer

Related on AIssential

• Evaluating Cross-lingual Knowledge Consistency in Code-Mixed vis-a-vis Indian Languages using IndicKLAR — Code-mixing inputs significantly improves large language model knowledge recall in Indian languages, nearly matching English performance.
• MORPHOGEN: A Multilingual Benchmark for Evaluating Gender-Aware Morphological Generation — MORPHOGEN evaluates LLM gender-aware morphological generation in French, Arabic, and Hindi via a sentence rewriting task.
• Reference-Driven Multi-Speaker Audio Scene Generation from In-the-Wild Priors — Leveraging in-the-wild pretrained foundation models enables realistic multi-speaker audio scene generation from text and reference voices.
• Can AI Save India’s Dying Languages? | Building a Tulu LLM Without Training Data Ft. Prathamesh — Structured prompting and custom romanization enable LLMs to generate low-resource languages without extensive training data.
• A computational study of how Sanskrit-specific morphophonological phenomena behave under noisy spoken input — The study computationally analyzes Sanskrit morphophonology's resilience to noisy speech.
• Building Singlish2Sinhala: A Machine Learning Approach to Sinhala Transliteration — Informal transliteration systems like Singlish present unique NLP challenges due to spelling inconsistencies.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.