diff --git a/CHANGELOG.md b/CHANGELOG.md
index bcc0ff1..77328d6 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -15,8 +15,23 @@ callers).
 
 ## [Unreleased]
 
+### Added
+
+- **`LanguageRepository` port** in `alfred.domain.shared.ports`. Structural
+  Protocol covering `from_iso`, `from_any`, `all`, `__contains__`, `__len__`
+  — the surface previously coupled to the concrete `LanguageRegistry`.
+  Mirrors the `MediaProber` / `FilesystemScanner` pattern: domain code
+  depends on the Protocol, infrastructure provides the YAML-backed
+  adapter. Tests in `tests/infrastructure/test_language_registry.py`.
+
 ### Internal
 
+- **`SubtitleKnowledgeBase` types `language_registry` against the
+  `LanguageRepository` port** instead of the concrete `LanguageRegistry`
+  class. The default constructor still instantiates the concrete adapter
+  when no repository is injected — behaviour is unchanged for existing
+  callers. Opens the door to in-memory fakes in future tests without
+  loading the full ISO 639 YAML.
 - **Moved `detect_media_type` and `enrich_from_probe` from
   `alfred.application.filesystem` to `alfred.application.release`**.
   They are inspection-pipeline helpers — their natural home is next to
diff --git a/alfred/domain/shared/ports/__init__.py b/alfred/domain/shared/ports/__init__.py
index ee7b70c..4cd0c44 100644
--- a/alfred/domain/shared/ports/__init__.py
+++ b/alfred/domain/shared/ports/__init__.py
@@ -7,11 +7,13 @@ Protocol without going through real I/O.
 """
 
 from .filesystem_scanner import FileEntry, FilesystemScanner
+from .language_repository import LanguageRepository
 from .media_prober import MediaProber, SubtitleStreamInfo
 
 __all__ = [
     "FileEntry",
     "FilesystemScanner",
+    "LanguageRepository",
     "MediaProber",
     "SubtitleStreamInfo",
 ]
diff --git a/alfred/domain/shared/ports/language_repository.py b/alfred/domain/shared/ports/language_repository.py
new file mode 100644
index 0000000..d62c137
--- /dev/null
+++ b/alfred/domain/shared/ports/language_repository.py
@@ -0,0 +1,36 @@
+"""LanguageRepository port — abstracts canonical language lookup.
+
+The adapter (typically loading from ISO 639 YAML knowledge) maps a wide
+range of raw forms (codes, English/native names, aliases) onto the
+canonical :class:`Language` value object. Domain code accepts the port
+via constructor injection; tests can pass a small in-memory fake.
+"""
+
+from __future__ import annotations
+
+from typing import Protocol
+
+from alfred.domain.shared.value_objects import Language
+
+
+class LanguageRepository(Protocol):
+    """Canonical language lookup."""
+
+    def from_iso(self, code: str) -> Language | None:
+        """Look up by canonical ISO 639-2/B code (case-insensitive)."""
+        ...
+
+    def from_any(self, raw: str) -> Language | None:
+        """Look up by any known representation: ISO code, name, alias.
+
+        Case-insensitive. Returns ``None`` when the raw form is unknown.
+        """
+        ...
+
+    def all(self) -> list[Language]:
+        """Return all known languages, in a stable order."""
+        ...
+
+    def __contains__(self, raw: str) -> bool: ...
+
+    def __len__(self) -> int: ...
diff --git a/alfred/infrastructure/knowledge/subtitles/base.py b/alfred/infrastructure/knowledge/subtitles/base.py
index a3d42eb..e5e178b 100644
--- a/alfred/infrastructure/knowledge/subtitles/base.py
+++ b/alfred/infrastructure/knowledge/subtitles/base.py
@@ -2,7 +2,7 @@
 
 import logging
 
-from alfred.infrastructure.knowledge.language_registry import LanguageRegistry
+from alfred.domain.shared.ports import LanguageRepository
 from alfred.domain.subtitles.value_objects import (
     ScanStrategy,
     SubtitleFormat,
@@ -12,6 +12,8 @@ from alfred.domain.subtitles.value_objects import (
     SubtitleType,
     TypeDetectionMethod,
 )
+from alfred.infrastructure.knowledge.language_registry import LanguageRegistry
+
 from .loader import KnowledgeLoader
 
 logger = logging.getLogger(__name__)
@@ -28,10 +30,12 @@ class SubtitleKnowledgeBase:
     def __init__(
         self,
         loader: KnowledgeLoader | None = None,
-        language_registry: LanguageRegistry | None = None,
+        language_registry: LanguageRepository | None = None,
     ):
         self._loader = loader or KnowledgeLoader()
-        self._language_registry = language_registry or LanguageRegistry()
+        self._language_registry: LanguageRepository = (
+            language_registry or LanguageRegistry()
+        )
         self._build()
 
     def _build(self) -> None:  # noqa: PLR0912 — straight-line YAML projection
diff --git a/tests/infrastructure/test_language_registry.py b/tests/infrastructure/test_language_registry.py
new file mode 100644
index 0000000..3e6fd7e
--- /dev/null
+++ b/tests/infrastructure/test_language_registry.py
@@ -0,0 +1,82 @@
+"""Tests for ``LanguageRegistry`` — the YAML-backed adapter for the
+:class:`alfred.domain.shared.ports.LanguageRepository` port.
+
+The port is structural (Protocol), so the assertion that the adapter
+satisfies it is a static one — we exercise the public surface here and
+let mypy / runtime polymorphism do the rest.
+"""
+
+from __future__ import annotations
+
+from alfred.domain.shared.ports import LanguageRepository
+from alfred.domain.shared.value_objects import Language
+from alfred.infrastructure.knowledge.language_registry import LanguageRegistry
+
+
+def _registry() -> LanguageRepository:
+    """Return a fresh registry typed as the port — proves structural fit."""
+    return LanguageRegistry()
+
+
+class TestPortSurface:
+    def test_satisfies_protocol(self):
+        # If LanguageRegistry diverged from LanguageRepository, the annotation
+        # below would already be wrong at type-check time; at runtime, this
+        # just confirms the methods exist.
+        reg: LanguageRepository = LanguageRegistry()
+        assert hasattr(reg, "from_iso")
+        assert hasattr(reg, "from_any")
+        assert hasattr(reg, "all")
+
+    def test_len_reflects_loaded_entries(self):
+        reg = _registry()
+        # The builtin YAML ships dozens of languages — exact count drifts
+        # with knowledge updates, so just sanity-check it's non-empty.
+        assert len(reg) > 0
+
+
+class TestFromIso:
+    def test_known_iso_returns_language(self):
+        reg = _registry()
+        fre = reg.from_iso("fre")
+        assert isinstance(fre, Language)
+        assert fre.iso == "fre"
+
+    def test_case_insensitive(self):
+        reg = _registry()
+        assert reg.from_iso("FRE") == reg.from_iso("fre")
+
+    def test_unknown_iso_returns_none(self):
+        assert _registry().from_iso("zzz") is None
+
+    def test_non_string_returns_none(self):
+        assert _registry().from_iso(None) is None  # type: ignore[arg-type]
+
+
+class TestFromAny:
+    def test_english_name(self):
+        reg = _registry()
+        lang = reg.from_any("French")
+        assert lang is not None
+        assert lang.iso == "fre"
+
+    def test_iso_639_1_alias(self):
+        # "fr" is the 639-1 form, registered as an alias.
+        reg = _registry()
+        lang = reg.from_any("fr")
+        assert lang is not None
+        assert lang.iso == "fre"
+
+    def test_unknown_returns_none(self):
+        assert _registry().from_any("vostfr") is None
+
+    def test_non_string_returns_none(self):
+        assert _registry().from_any(123) is None  # type: ignore[arg-type]
+
+
+class TestMembership:
+    def test_contains_known(self):
+        assert "english" in _registry()
+
+    def test_does_not_contain_unknown(self):
+        assert "klingon" not in _registry()