What are the use cases for encoder-based transformers?

• A. A model used for image generation, audio synthesis, and video tagging.
• B. Text generation, chatbots, and code generation.
• C. Text classification, semantic search, and named entity recognition.
• D. Text-to-speech conversion and audio transcription.

Answer: (C)

Encoder-based transformers are built to understand language by turning input text into rich, contextual representations. That makes them especially strong for tasks that require interpreting and categorizing what’s in a text, comparing meanings, or labeling parts of a sentence.

For text classification, the encoder outputs a representation of the whole input that a simple classifier can map to a category, letting you determine sentiment, topic, or other labels efficiently. For semantic search, you encode both queries and documents into vectors and compare these vectors to measure semantic similarity, enabling fast and scalable retrieval over large collections. For named entity recognition, the encoder provides token-level representations you can feed into a per-token classifier to identify entities like names, organizations, or dates within the text.

These strengths come from the encoder’s ability to capture context across the entire sequence and produce stable representations that downstream heads can use for discrimination or labeling. In contrast, generation-focused tasks such as text generation, chatbots, or code generation rely on decoders or encoder-decoder architectures to produce new text. Text-to-speech or audio transcription involve audio processing and specialized models beyond the encoder-only setup.