SGNLP

SGNLP packages a family of models and tools:

• Singapore English understanding: Singlish text normalization, sentiment analysis
• Multilingual code-switching: detecting which languages a passage mixes (English / Chinese / Malay / Tamil mix)
• Local named entities: recognizing Singaporean place names, person names, and organization names
• Paraphrase and summarization: tuned for local Singaporean news and government text

As SEA-LION emerged, SGNLP's role gradually shifted from "flagship product" to "specialty toolkit" — general NLP capabilities ceded ground to LLMs, but specialty scenarios like Singlish still hold standalone value.

The core problem SGNLP solves: off-the-shelf NLP tools perform poorly on Singapore English.

Singlish blends English, Malay, Mandarin, and Tamil and adds distinctive grammar (particles like *lah*, *leh*, *lor*), which leaves out-of-the-box models from spaCy / NLTK / HuggingFace performing badly on Singlish text. SGNLP's pretrained models are fine-tuned on Singlish data and significantly more accurate than generic models.

Relationship with SEA-LION: as an LLM, SEA-LION covers part of SGNLP's surface area, but SGNLP's lightweight models (some under 100 MB) retain an edge in edge deployment and real-time processing scenarios.

SGNLP is an early practical expression of Singapore's "language sovereignty" narrative — even before the LLM era, AISG was already building "language AI tailored for Singapore."

Across the seven transmission levers:

• Lever 3 (Industry Adoption): local customer service, social media analysis, government text processing
• Lever 1 (Foundational Research): Singlish is one of the few "creole Englishes" with genuine academic research value

Take: SGNLP gave SEA-LION a "philosophical predecessor" — the same "build specialty AI for local languages" ethos, simply upgraded from NLP tooling to an LLM.