The 1948 paper
Claude E. Shannon published A Mathematical Theory of Communication in two installments of The Bell System Technical Journal, volume 27: pages 379–423 in July 1948 and 623–656 in October. The original article bears Shannon's name alone. Its introduction acknowledges earlier work by Nyquist and Hartley while extending the analysis to noise and the statistical structure of messages.
The central engineering problem is reproducing a message chosen elsewhere. Shannon brackets semantic questions for this purpose; he does not thereby prove that meaning is unimportant in every study of communication.
References: [1]
Sources, channels and signals
The paper's system diagram distinguishes an information source, transmitter, channel, receiver and destination, with noise entering the transmission process. Messages are not identical to the physical signals that carry them. The receiver's task is to recover the selected message from what arrives.
The probabilistic framework assigns distinct roles to source statistics and channel behavior. Entropy characterizes uncertainty in the source, while capacity describes a transmission limit. Bell Labs' retrospective stresses that these quantities depend on the particular source and channel, rather than being universal constants applicable to every communication arrangement.
Coding and achievable limits
Two different coding tasks help explain the framework's engineering significance. Compression exploits predictable structure to reduce representation costs; error-control coding adds carefully structured redundancy to protect messages. Their apparently opposite treatments of redundancy address different problems, and both depend on a model.
In the standard discrete memoryless formulation, capacity is the maximum input-output mutual information over input distributions. Below capacity, suitable sufficiently long block codes can make error probability arbitrarily small. Above capacity, this form of asymptotically reliable communication is impossible. Repeating every bit many times can reduce errors in some models, but its falling code rate illustrates why reliability alone is not the complete objective.
These are limiting results, not a promise that a specific finite code will attain capacity under any latency or computational constraint. They provide benchmarks for evaluating implementations. Bell Labs describes the subsequent relationship between theory and coding or compression practice as one in which products progressively approached mathematically predicted limits.
Not the same publication as the 1949 book
The University of Illinois Press issued The Mathematical Theory of Communication in 1949, credited to Claude E. Shannon and Warren Weaver. The National Library of Medicine catalog identifies two contributions: Shannon's mathematical paper and Weaver's Recent Contributions to the Mathematical Theory of Communication. Its bibliographic note identifies the journal installments as the source of the reprinted Shannon text.
The book's joint attribution does not make Weaver a coauthor of the 1948 article. Even the English titles differ in their initial article: A for the paper, The for the book. Keeping author, date and document type explicit prevents the technical paper and the later collection from being treated as a single undifferentiated text.
References: [3]