This post is following the recent post by MAXIME GELLY-PERBELLINI with the same topic, but with a different research mater. Primary sources about the French royal admiralty for the medieval period are rather scarce and one looking for the Admiral of France will at first be very disappointed when looking in the French Royal Chancery Registers. Would it be? In fact the query “admiral” or “admiraldum” with all its declination in the HIMANIS search engine will only return one hit in French (JJ112, fol. 51v) and none in Latin, and for that one will have to take down the confidence level to 14 . For having gone through the printed inventories and through the registers manually, we can say that the Admiral of France is indeed present in the registers more than once. What is the reason for that?

Firstly, the word « admiral » was not part of the trained model. The recognizer have to read it by himself without previously knowing the word. Secondly, the word in itself is subject to reading error, for it contains a sequence of multiple similar downstrokes.

So, how to find the Admiral (or the Visdmiral) in the corpus? While looking in the complete list of indexed words I’ve found following word variation: adjural, annual, adminal and Anviril. The same problem occurs with the Visadmiral, as the computer read Visammal, Visauvel, Visanmal or Visaumal. Clearly, as the computer have to read the word without being previously trained with this word in particular, it is the sequence of “m”, “i” and “r” that causes the reading error. Below a quick summary of all the query that merit to find the foresail admiral. For consistence, all queries have been performed with 14 confidence level.

• Admiral @14 > 1 hit
  • Adjural @ 14 > 34 hits
  • Annual @ 14 > 25 hits (majority of false hit)
  • Adminal @ 14 > 32 hits
  • Anviril @14 > 1 hit

The present version of HIMANIS did not use any dictionary for the word recognition, only a 5gram model. In future development, correcting the list of indexed words automatically by injecting linguistic reference will certainly help improve the recognition.

Having access to the list of indexed word certainly help to find the admiral, but it is not a panacea and does not resolve all the problems. With 4.000.000 words the complete list of indexed word is very difficult to handle on a basic computer..

One can also take benefit of the recal module. But for that you already need to have some good hits and explore how the computer read the word. That is how, in one case when the word “admiral” is separate in two parts ( “ad” and “val”), we found one more occurrence..

(1) The office of admiral of France is still today not very well known. See Mathieu Le Hunsec, « L’amiral, cet inconnu », Revue historique des armées [Online], 266 | 2012, Online since 06 September 2012, connection on 23 November 2018. URL : http://journals.openedition.org/rha/7393