Introduction

Kai Niu, Ping ZhangSpringerBriefs 2025 · DOI · OpenAlex

Abstract: This chapter retrospects the research history of semantic information. By investigating the semantic information of various source modalities and downstream tasks, we find that synonymity is the essence of semantic information and synonymous mapping is a basic relationship between the semantic information and syntactic information. Then we give a systematic framework of semantic information theory and summarize the major contributions of this book.

This document serves as the introductory chapter to the book The Mathematical Theory of Semantic Communication (2025) by K. Niu and P. Zhang. It establishes a new theoretical framework that extends Claude Shannon’s classic information theory into the realm of meaning, proposing a mathematical foundation for "Semantic Communication."

The Bottleneck of Classic Information Theory

Classic Information Theory (CIT), established by Claude Shannon in 1948, revolutionized data transmission by focusing on the technical problem: how accurately can symbols be transmitted? However, CIT explicitly ignores the meaning of those symbols.

The classic Shannon communication system model.

As modern communication technologies approach the theoretical "Shannon limits," engineers face a bottleneck. We can transmit bits near-perfectly, but we aren't necessarily transmitting information efficiently. Previous attempts to incorporate semantics (Level B communication) relied on propositional logic or fuzzy variables, which fail to scale to complex data like high-resolution images or video. This paper seeks to fill that gap by establishing a systematic mathematical theory for semantic communication that handles complex source modalities.

The Essence of Semantic Information: Synonymity

The authors identify synonymity as the core essence of semantic information. They introduce several novel concepts to bridge the gap between syntactic (bit-level) and semantic (meaning-level) information:

  • Synonymous Mapping: This is the cornerstone of the theory. Semantic information is defined through the lens of synonymity—the idea that many different syntactic representations (words, pixels, waveforms) can map to the same single meaning. This is modeled as a one-to-many mapping from a semantic alphabet to a syntactic alphabet.
  • Semantic Entropy ($H_s(\tilde{U})$): A new measure for the uncertainty of semantic information, using the "sebit" (semantic bit) as the fundamental unit.
  • Maximum Likelihood Group (MLG) Decoding: A new decoding rule where the decoder looks for the "synonymous group" that most likely represents the intended meaning, rather than looking for an exact bit-string match.
Synonymous mapping across text, speech, image, and video modalities.

Illustrative Modalities

While theoretical, the framework is illustrated across four primary media modalities:

  1. Text: Mapping synonyms like "happy," "joyful," and "content" to a single semantic concept.
  2. Speech: Recognizing that different waveforms or Mel spectrograms can represent the same spoken sentence.
  3. Image: Utilizing object-oriented segmentation to identify semantic entities (e.g., "bears" or "sky") from raw pixels.
  4. Video: Extracting sequences of frames that map to a high-level semantic description, such as "A jockey guides his horse."

Key Insights and Theoretical Results

The "aha moment" of this research is the realization that while syntactic data is myriad and complex, semantic meaning is often singular. This leads to significant efficiency gains: because semantic entropy is proven to be less than or equal to its classic counterpart ($H_s(\tilde{U}) \leq H(U)$), semantic coding can achieve much higher compression ratios than traditional lossless methods.

Furthermore, the authors derive a specific formula for Gaussian Semantic Capacity: $C_s = \frac{1}{2} \log [S^4 (1 + \frac{P}{\sigma^2})]$, where $S$ represents the average synonymous length. This suggests that semantic channel coding can improve transmission capability because the semantic capacity ($C_s$) is generally greater than or equal to the classic Shannon capacity ($C$).

Limitations and Future Outlook

The authors acknowledge that while they have tackled synonymity (many-to-one mapping), they have largely set aside the problem of ambiguity (one-to-many mapping, where one symbol has multiple meanings), which they consider a secondary challenge for future work. Additionally, the "stable mappings" required to extract semantics often rely on deep learning or human knowledge bases, which may present computational challenges for real-time hardware implementation. This chapter serves as a roadmap for the rigorous proofs and applications detailed in the subsequent chapters of the book.

Summary updated

Abstract

Abstract In this chapter, we first retrospect the research history of semantic information. By investigating the semantic information of various source modalities and downstream tasks, we find that synonymity is the essence of semantic information and synonymous mapping is a basic relationship between the semantic information and syntactic information. Then we give a systematic framework of semantic information theory and summarize the major contributions of this book.
BibTeX
@inproceedings{niu2025introduction,
  author = {Kai Niu and Ping Zhang},
  title = {{Introduction}},
  booktitle = {SpringerBriefs},
  year = {2025},
  doi = {10.1007/978-981-96-5132-0\_1},
}

Loading related content...

Figure