Reading the Unreadable — The Grammar of Decipherment Puzzles, from Fez to Chants of Sennaar

Introduction

The first time I launched Chants of Sennaar, the screen seemed to refuse meaning. Symbols on pillars, notices above doors, speech bubbles — none of it readable. An hour later I had matched several glyphs to gate, ascend, prayer. No one had taught me; I had begun reading a made-up language. That moment — the unreadable becoming readable — is the core pleasure of the decipherment puzzle.

A decipherment puzzle asks the player to crack a fictional script or language using observation and inference alone. The central verb is not push or place but map: matching sign to meaning. Here I line up four games that turned invented languages into play — Fez (2012), Tunic (2022), Heaven's Vault (2019), and Chants of Sennaar (2023). I touched on nearby ground in observation as play; decipherment reads as a special branch of it.

Fez's Zu Script — Burying the Cipher in an Optional Layer

Phil Fish's Fez (2012) is a diorama platformer where rotating a 2D plane reveals 3D depth. But its real depth sits in the unreadable marks etched into the background. The Zu writing system is a simple substitution cipher into English, keyed by the pangram 'The quick brown fox jumps over the lazy dog', alongside its own number system and a tetromino code that mirrors input gestures — all scattered across the world.

What matters is that Fez never forces you to decode. You can reach the ending without reading a single glyph. The script is an optional layer, a second game that only the observant begin. This optionality is one pole of decipherment design — a reward for high observation resolution. Fittingly, the full cipher was cracked by the community's collective effort at launch; the depth did not fit in one field of view.

What I take from Fez is the trade-off: making decipherment optional turns it into an intense discovery for some and an invisible layer for most. Required reading or elective — that single choice forks the decipherment puzzles that follow.

Tunic's Manual — Making the Instruction Booklet Itself the Puzzle

Andrew Shouldice's Tunic (2022) took the other pole. It looks like an isometric Zelda, but the true lead is the in-game instruction manual, collected page by page. The pages ape a real booklet, written in a runic script known as Trunic that you can ignore to the end. Read it, though, and you find the secret passages and the true ending were spelled out all along. As I noted in the Tunic review, the manual is both walkthrough and greatest riddle.

Trunic characters represent sounds — consonant and vowel parts fused into one glyph, joined by a line into words. There is no Rosetta stone. You match English word orders against shapes and slowly recover the phonetic values. What impressed me is a page near the end, around 53 to 54, that shows how the script works: a hint, not a full key, leaving the achievement of reading in the player's hands.

Where Fez scattered its cipher across the world, Tunic condensed it into a single object. The unreadable manual stages the feeling that the game 'wasn't written for you', while maximizing the intimacy of the moment it finally is. Focusing observation on one artifact echoes how Return of the Obra Dinn folded its information into a ship's logbook.

Heaven's Vault — Translating by Probabilistic Grammar

inkle's Heaven's Vault (2019) is a rare case that puts decipherment at the system's center. As the archaeologist Aliya, sailing between moons, you translate inscriptions in an 'Ancient' language. Each inscription is split into words, with a few candidate English meanings offered, and you pick the likeliest from context. Ancient words are built from small atomic parts, so translating one word teaches you parts that recur in others.

The frightening part of this design is that you can proceed while wrong. The game does not immediately confirm the answer; your chosen reading becomes Aliya's belief, feeds the story, and can even unlock sites. Carrying an uncertain translation forward is a different tension from the 'I think I get it' of The Witness. I see this tolerance of misreading as what lifts decipherment from quiz to language learning.

Chants of Sennaar — Five Languages and the Resolution of Looking

Rundisc's Chants of Sennaar (2023) builds on the Babel myth: a tower where five peoples each live in a different language. Climbing and descending, you decode the tongues of Devotees, Warriors, Bards, Alchemists, and Anchorites in turn. At the center is a notebook — you sketch a guess for each glyph, and when a page's guesses cohere with context, the page validates.

Chants is clever because it makes you use words the moment you learn them. Learn door and open, and a scene arrives where those two words slip you past a guard. Reading becomes progress itself. The five languages are not isolated; later you play translator between peoples. As resolution rises, the whole tower reads as one long sentence — a peak of the observation-as-play lineage.

As a designer, what I learn most is that notebook confirmation. Grading by the page rather than the glyph creates a gap between the time spent holding uncertain guesses and the moment they all pay off at once. Where Heaven's Vault let misreading stand, Chants gently reclaims it. Same act of decoding; feedback design changes the feel entirely.

What Makes Decipherment Work — Feedback and the Design of Misreading

Line up the four and the conditions for decipherment-as-play appear. First, the fictional language needs a consistent internal grammar; a random pile of symbols never builds a learning curve where one guess begets the next. Fez's substitution and Chants's vocabulary are both few rules combined. Second, at what grain do you return right and wrong — per word, per page, or as a delayed story consequence? That grain sets how long the player holds uncertainty.

Third, reading must pay off at once. Tunic's secret passages and Chants's conversations spend the fruit of decoding immediately; leave decipherment as mere collection and players tire of filling a dictionary. Most delicate is how you treat misreading: correct it instantly (Chants), weave it into the story and leave it (Heaven's Vault), or never flag it (Fez). That one choice all but sets the tone.

At bottom, a decipherment puzzle turns the learning curve itself into play. A dictionary grows inside the player's head, and that growth is the progress. It shares one edge with the observation-as-play lineage: the resolution at which you read the board.

Sources

Basic facts drew on the following. For Fez, Wikipedia: Fez; for Tunic, Wikipedia: Tunic.

Heaven's Vault's language design is detailed in inkle's own How Inkle developed its own ancient language. For Chants of Sennaar, see Wikipedia: Chants of Sennaar.

Closing

The unreadable becoming readable is one of the purest 'I get it' moments a puzzle can give. Fez scattered decoding into an optional layer, Tunic condensed it into one manual, Heaven's Vault wove misreading into story, and Chants of Sennaar gently reclaimed it with a validating notebook. From the same verb — map — come strikingly different textures.

If I built a decipherment puzzle next, I would decide first how to treat misreading. Rush the answer and the scent of language learning fades; leave it be and both dread and discovery deepen. And I would set up at least one scene that spends a freshly learned word before the dictionary fills — to hand the player the moment reading becomes progress as early as I can.