📄 As it ys seyde to fore'. Some Linguistic Evidence in the Process of Compiling Middle English Medical Recipes

Summary

The paper looks at how a fifteenth-century scribe arranged and connected medical recipes in a vernacular remedy book. Earlier scholarship mostly catalogued the contents and structure of such books, and little was known about how they were written from a linguistic point of view. The corpus is one receptarium in MS Hunter 185 (ff. 17r–62r, Glasgow University Library), copied by a single scribe in Anglicana Formata. Alonso applies the discourse-analysis model of Brown & Yule (1996) and separates two kinds of linking. Extralinguistic links are red running titles and margin notes. Linguistic links are the head-to-toe ordering, thematic subgroups, cohesion devices inside groups (anaphora, substitution, ellipsis) and explicit cross-references between recipes. The scribe uses formulaic title openers to show that a recipe belongs to the same group as the one before it. He also points back to earlier recipes, adding more identifying detail when the target is further away. The paper concludes that the collection was planned by someone who knew the material well, and was not a random assembly of recipes.

Key points

• The data is MS Hunter 185, ff. 17r–62r: mostly English recipes, a few in Latin. Remedies run from head to toe (waters, oils, plasters, salves, drinks, powders and more).
• Rubricated titles could run to four lines and sometimes hid the recipe's purpose. Readers coped by underlining key words in a different ink and adding margin notes. Palaeographic evidence suggests the readers, not the scribe, wrote those notes.
• Group markers such as also, an other, item and for þe same let the scribe write the full title only once per group. The following titles are elliptical.
• Backward cross-references (e.g. bifore saide) carry extra descriptors like for þe fluxe. The further away the target recipe, the longer the reference. In one case a descriptor matches the target's rubric word for word.
• Only one forward reference occurs (þat is jtolde next after). It sends the reader to the next recipe for the second part of a treatment.

Interesting ideas

• I1. Headings marked as "same as above" inherit the topic of the last fully written heading, so a document's heading hierarchy works as a chain of ellipsis.
• I2. The length of a cross-reference tends to grow with the distance to its target, so a reference's level of detail hints at how far back to search.
• I3. Reference descriptors often repeat the target's title exactly, which makes title-string matching a cheap and explainable way to link a reference to its target.
• I4. Forward ("told next after") and backward pointers together split one treatment across several recipes, so a procedure's steps can extend beyond a single text unit.
• I5. A recipe's purpose is sometimes stated only in its body ("this plaster is good for…"), not in its title; readers' underlinings show exactly where titles failed.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

• I1 (world): When a section title reads "Another for the same", Mova could copy the purpose and ailment from the last full title into the new entity's state instead of leaving its goal empty.
• I2 (frames): Coreference could treat bare pointers ("as said before") as favouring nearby antecedents. Descriptive pointers would trigger a wider search, matching their descriptor against earlier titles.
• I3 (world): When resolving references within procedural or instructional texts, Mova could first try an exact or near-exact match against stored section titles, then fall back to slower semantic matching.
• I4 (world): The event-order layer could join steps from linked sections into one procedure, so that questions like "what do you do after the plaster?" can be answered.
• I5 (answer): Purpose-question readers should fall back to body sentences of the form "X is good for Y" when a heading is vague, and say which source they used.