Reference MCP Interpret Token

interpret_token

Walk the reading chain for a single token on a tablet.

Arguments

Argument	Type	Required	Description
p_number	string	yes	CDLI P-number
token_id	integer	yes	tokens.id — discover by searching first or browsing the artifact page

To find token_id: call semantic_search with the P-number, or use the artifact ATF endpoint (GET /artifacts/{p_number}/atf) which returns token IDs in the parsed output.

When lemmatized: the complete chain

If the token has an ORACC lemmatization, interpret_token returns the full reading chain with a source at each step:

written_form: "lu₂-gal"  [CDLI ATF, annotation_run 3]
  → normalized: "lugal"  [normalization bridge]
    → lemma: lugal[king]N  [ePSD2, annotation_run 7]
      → sense: "king, ruler; owner"  [ePSD2]

Each step cites its source. The chain is traceable to the original annotation run.

When not lemmatized: ranked hypotheses

If the token has no lemmatization, interpret_token returns 1–3 ranked hypotheses. Each hypothesis is built from:

• Candidate sign readings — OGSL readings for the sign or sign sequence
• Neighbor context — what the surrounding tokens suggest about POS and semantic field
• Genre and period priors — what readings are most common in this genre and time period

Phrasing constraints: hypotheses are always phrased as:

• "resembles the pattern of..."
• "consistent with..."
• "may be a form of..."

They are never phrased as "is X" or "means X." The distinction matters: this is a hypothesis based on evidence, not a confirmed reading.

Response

ToolResponse[ChainPayload]:

{
  "data": {
    "form": "lu₂-gal",
    "p_number": "P227657",
    "token_id": 441,
    "chain": [
      { "step": "written_form", "value": "lu₂-gal", "source": "CDLI ATF" },
      { "step": "normalized", "value": "lugal", "source": "normalization bridge" },
      { "step": "lemma", "value": "lugal[king]N", "source": "ePSD2" },
      { "step": "sense", "value": "king, ruler; owner", "source": "ePSD2" }
    ],
    "hypotheses": null
  },
  "summary": "lu₂-gal → lugal[king]N (Sumerian noun, 'king')",
  "sources": [...],
  "interaction_id": "44"
}

When hypotheses are returned instead of a chain, chain is null and hypotheses is an array of ranked interpretations each with confidence, evidence, and phrasing.

Correcting an interpretation

If the tool's reading is wrong, use submit_correction with the interaction_id from this response, the claim that was incorrect, and what it should say. See submit_correction.