When meaning meets purpose: A survey on semantic and task-oriented communications

  • Do, Quang Tuan; 
  • Do, Tung Son; 
  • Truong, Thanh Phung; 
  • Won, Dongwook; 
  • Dao, Nhu-Ngoc; 
  • ... Cho, Sungrae; 
  • 외 1명
Citations

WEB OF SCIENCE

0
Citations

SCOPUS

0

초록

As wireless systems evolve toward sixth-generation (6G) networks, traditional bit-centric approaches cannot support intelligent applications that require meaning-aware and effectiveness-aware transmission. This survey examines semantic communication (SC) and its specialized subset, task-oriented semantic communication (ToSC), which prioritizes task-specific utility over general semantic completeness. Unlike prior surveys that treat this distinction descriptively, we synthesize the information bottleneck framework with empirical evidence from the surveyed literature to characterize the qualitative regimes under which SC and ToSC diverge or converge, and to distill design heuristics. An architectural taxonomy along functional axes (encoding strategy, transmission mode, task-relevance mechanism, and adaptivity) captures the recurring mechanisms shared across implementations. We further analyze the deep learning model families (autoencoders, convolutional and graph neural networks, Transformers, and large language and multimodal models) that implement semantic extraction, compression, and task reasoning. A structured evaluation framework addresses metric definitions and the conflicts and trade-offs between semantic fidelity, task effectiveness, and communication efficiency. Across ten application domains, we identify domain-specific risk factors, unresolved limitations, and four cross-domain weakness patterns: knowledge dependency as the most pervasive bottleneck, weak-baseline comparisons that inflate reported gains, brittleness in safety-critical deployments, and inconsistent evaluation methodology that prevents meaningful cross-domain comparison. Open challenges are consolidated into a dependency-ordered research roadmap that prioritizes foundational theory, standardized benchmarks, and scalable knowledge base management. This survey positions SC as the overarching paradigm for 6G information-directed communications, with ToSC as its action-driven specialization, and offers practical criteria for deciding when each applies and how to evaluate it.

키워드

Semantic communication; Task-oriented communication; Deep learning; Joint source-channel coding; Large language models; Sixth-generation wireless networks; DIGITAL-TWIN SYNCHRONIZATION; IMAGE TRANSMISSION; OPTIMIZATION; EXTRACTION; INTERNET; SYSTEM
제목
When meaning meets purpose: A survey on semantic and task-oriented communications
저자
Do, Quang Tuan; Do, Tung Son; Truong, Thanh Phung; Won, Dongwook; Dao, Nhu-Ngoc; Noh, Wonjong; Cho, Sungrae
DOI
10.1016/j.engappai.2026.116063
발행일
2026-10
유형
Article
저널명
Engineering Applications of Artificial Intelligence
권
182