Provide AI with a Glossary: Standardizing Terminology in Conversation

Giving an AI a terminology table at the start of a conversation is a practical way to stabilize the nouns that will appear in every subsequent sentence. The purpose is to prevent the same word from carrying two different meanings within one dialogue. When an AI is told explicitly which characters and labels belong to the discussion, later references are less likely to drift into paraphrase or loose usage. The terminology table does not need to be long, but it does need to be fixed.

对话开始时先给术语表

The terminology table should be provided at the very beginning of the conversation, before any analysis, case reading, or interpretive question is introduced. Its role is not to educate the model in a broad sense, but to anchor the vocabulary of that particular session. If the table is given later, the earlier part of the dialogue may already have established inconsistent usage, and the model may carry those inconsistencies forward even after a correction is made.

A workable opening can be as simple as a short block that lists the relevant terms and their fixed written forms. The table does not need to explain the concepts at length. Its main function is to make the written form of each term explicit so that later output can be checked against it word by word. This is especially important in Chinese-metaphysics material, where the same concept may be written in more than one way, and where a single character can belong to more than one category depending on context.

术语表要包含哪几类

A minimal terminology table for Chinese-metaphysics discussion should include at least three categories. The first is the writing of the ten Heavenly Stems and the twelve Earthly Branches: 甲，乙，丙，丁，戊，己，庚，辛，壬，癸 for the stems, and 子，丑，寅，卯，辰，巳，午，未，申，酉，戌，亥 for the branches. These are the basic calendrical and symbolic units, and their written forms are not interchangeable.

The second category is the names of the ten Ten Gods. The terminology table should list them in the exact form that will be used throughout the conversation. The names include 正官，七杀，正印，偏印，正财，偏财，食神，伤官，比肩, and 劫财. It is not enough to mention that Ten Gods will be discussed; each name should be written out so that the model has a fixed form to reuse.

The third category is the relational terms that will actually be used in the discussion, such as 合，冲，刑，害, and 破. These characters denote specific types of interaction between branches or stems. Because they are short and common, they are easy for a model to blur or substitute. Listing them in the table reminds the model that these are technical terms in the current conversation, not ordinary verbs or general descriptions.

为什么固定写法比解释更重要

Fixing the written form matters more than giving an explanation. Explanations can be rewritten by the model without warning. A definition may be compressed, expanded, paraphrased, or reordered in a way that subtly changes its scope. But a written form is either present or absent, identical or altered. It can be compared character by character.

When the table says that the term to use is 七杀, the reader can check later output directly against that form. If the output contains 偏官, the mismatch is visible even if the surrounding explanation is fluent. An explanation, by contrast, can drift without being easy to detect. The model may say the right word in one sentence and a near-equivalent in the next, while still sounding coherent. Fixed characters make the check mechanical rather than interpretive.

This is why a terminology table should emphasize exact written strings over definitions. Definitions are useful, but they are secondary. The primary commitment is to a spelling convention for the session. Once that convention is stated, the model is less likely to introduce a variant, and the user is better able to spot one if it appears.

术语漂移的两个迹象

Terminology drift can occur even after a table is provided, but the table makes the drift easier to identify. The first sign is synonymous substitution. An example is writing 七杀 as 偏官 and then continuing without noting that the two are the same Ten God. In some traditions and texts, 七杀 and 偏官 are treated as equivalent, but the terminology table should decide which form will be used. If the table fixed 七杀, then the appearance of 偏官 is a drift unless the text explicitly says that the two are being treated as the same term.

The second sign is scope sliding. An example is taking the term 财 and expanding it to refer generally to everything related to money. In a technical discussion, 财 usually refers to 正财 or 偏财 as a Ten God, or to the broader category of wealth-star-related meanings in a specific framework. But if the model begins to use 财 loosely for any financial topic, the boundary of the term has shifted. The table helps here because it can specify the intended scope: whether 财 is to be read strictly as a Ten God category or in a wider everyday sense.

Both signs are easier to notice when the fixed form has been stated. A synonym stands out because it is not the agreed character string. A scope slide stands out because the term begins to appear in contexts that do not match the agreed category.

一份可以复用的术语表

A terminology table can be saved and reused. Once a table has been tested in one session, the same block can be pasted into a different tool or a new conversation without modification. This is useful when moving between chatbots, prompt interfaces, or local models. The table does not need to be rewritten for each environment because its function is independent of the platform: it fixes the written vocabulary of the session.

The table should be kept as plain text or in a simple format that copies cleanly. Any formatting that breaks during paste should be avoided. The goal is that the pasted version is identical to the original, so that the same fixed forms are present in every session.

This site’s terminology pages give a definition for each term and can serve directly as the source for a terminology table. The user can copy the relevant term entries from those pages into the opening block. The definitions themselves may be included, but the essential part is the fixed written form of each term. If a term page lists more than one accepted name, the user should choose one for the table and state that choice explicitly.

After the terminology is fixed, the work of checking output changes. Instead of judging whether the meaning of a sentence is correct in a broad interpretive sense, the user can compare the written forms against the table. If the table says 正财 and the output says 正财, the check passes for that term. If the output says 财星 or 财 without explanation, the user can flag it. This shifts the verification task from semantic judgment to literal comparison, which greatly reduces the cost of review.

The table also reduces the need to reargue definitions in the middle of a conversation. When a term appears, the user can refer back to the opening block rather than repeating what the term means. This keeps the dialogue focused on the actual case or question rather than on vocabulary disputes. The fixed table is a reference point for the entire session, and its value increases as the conversation grows longer.