What Is Sentence-Length Steganography?

Direct answer

Sentence-length steganography is a linguistic hidden-information technique that uses a measured property of predefined sentences as a secondary channel. In a simple parity scheme, a sentence with an odd number of counted words can represent binary 0 while a sentence with an even number represents binary 1.

How sentence-length parity carries bits

The encoder first defines stable carrier-sentence boundaries and one exact counting rule. Each payload bit determines whether the corresponding sentence must end with an odd or even count under that rule. The decoder measures the same sentences in the same order, converts each parity state back to a bit, and reconstructs the byte stream.

A worked benign example

Suppose the rule counts contiguous ASCII letter runs and maps odd to 0 and even to 1. “Quiet guards crossed the western yard.” contains six counted runs and therefore contributes 1. “Quiet guards crossed the yard.” contains five and contributes 0. Reading several carrier sentences in a fixed order produces a binary sequence.

The word-count rule must be explicit

“Word count” is not a universal machine primitive. Software can disagree about apostrophes, hyphenated compounds, abbreviations, numbers, emoji, Unicode letters, and markup boundaries. A reproducible demonstration therefore publishes an exact rule such as counting matches of [A-Za-z]+ rather than relying on an unspecified tokenizer.

Punctuation, numbers, apostrophes, and hyphens

Under an ASCII-letter-run rule, ordinary punctuation contributes no counted word, digits are ignored, and punctuation merely separates neighboring runs. Apostrophes and hyphens also split runs, so a form such as “we're” contributes two alphabetic runs and a hyphenated form contributes one run for each alphabetic segment. Other schemes can choose different rules, but encoder and decoder must agree.

Sentence boundaries are part of synchronization

The decoder must know exactly where each carrier sentence begins and ends. Abbreviations, quotations, headings, lists, and parser-specific sentence segmentation can otherwise shift boundaries. FugitiveAI uses explicit numbered carrier items so the structural boundary is unambiguous before counting begins.

Capacity and naturalness trade off

A binary odd/even choice has a nominal ceiling of one raw bit per eligible carrier sentence. Real capacity is lower because forcing parity may require adding, removing, or replacing words, and too many such adjustments can make prose repetitive or unnatural. A lower-density channel often produces better writing.

Editing can flip bits without changing the topic

Parity channels are extremely fragile. Adding one counted word flips a sentence bit. Removing one does the same. A copy editor can therefore destroy the payload while preserving the sentence's visible topic and factual meaning. Larger edits can also alter sentence boundaries and desynchronize every later bit.

Paraphrasing, splitting, and merging

A paraphrase may replace one phrase with another of different length, combine two sentences, or split one sentence into several. All of those transformations can change parity or carrier indexing. Sentence-length channels are therefore poor choices when text will pass through uncontrolled rewriting or summarization.

Tokenizers and linguistic words are different concepts

A language-model tokenizer generally produces subword tokens, not the same units that a human calls words. Token count, whitespace-delimited word count, Unicode word segmentation, and regular-expression letter-run count can all disagree on the same visible sentence. A sentence-length puzzle must name which measurement actually carries the signal.

How to solve sentence-length hidden-message puzzles

Identify the exact carrier boundaries, counting rule, parity mapping, and order before measuring anything. Count only the designated units in each carrier sentence, record the mapped bit, and preserve synchronization. If decoded bytes are incoherent, recheck the segmentation rule before guessing a different cipher.

Detection and defensive analysis

When the rule is known, detection is mechanical: measure every carrier sentence and inspect the parity sequence. Without the rule, ordinary prose naturally contains a mixture of odd and even lengths, so parity alone is weak evidence. Defensive analysis should preserve original sentence boundaries and raw text before normalization, rewriting, or summarization removes the structural pattern.