Constructional Potentiality: Prlscianlc gra1~nar as a dlsamblguatlon 
technique in the automatic recognition of Latin syntax 
In most languages word order plays the major role in determining 
which words form a single phrase or constitute. A tree s£ructure can be 
abstracted automatically from a sentence by linear determination of the 
major syntactic constitutes. However, in certain highly-inflected 
languages, of which Latin is perhaps the most extreme example, constitutes 
may be broken up in the word order of the sentence. Inflectional endings 
form the signals for placing each constituent in its constitute. Theoretic- 
ally the higher degree of syntactic explicitness provided by inflectional 
morphemes should make these languages easier to analyse.mechanlcally. Yet 
it proves that there is a high degree of ambiguity in the morphemes or 
morphophonemie alternations used to mark inflection and that native 
speakers rely heavily on their ability to identify both~e semantic meaning 
and the inflectional habits of a given word. 
Recognizing this state of affairs those institutes involved with 
automatic parsing of Latin, the Laboratoire pour llAnalyse Statistique 
des Langues Anclennes, Liege, and the Centro Automazlone AnalisiLinguistica, 
formerly in Gallarate but now in Pisa, have constructed rather large 
parsing dictionaries which identify the word-class and inflectional habits 
of lexlcal lenmmta. These parsers do not, to my knowledge, produce tree 
structures of analysed sentences or give any deep-structure representatiflQ. 
Anlaternatlve approach to the automatic syntactic analyses of 
Latin is offered by the constructional syntax o~ Latin developed by ?riscian 
from Greek grammatical theory. Each word is a mc.,mber of at least one 
2-member construction. There are two types of constructlonal relationship 
convenientia (agreement), (a) verb agrees with subject in number, (b) adjective 
agrees with noun; regimen (government), (a) verb governs object, (b) pre- 
position governs noun. Several word-classes do not participate in cons- 
tructions, notably connectives, some ordinal numerals, modifiers and 
adverbs, but these can be located in the syntax of the sentence, usually 
by word order, occasionally by the reduction of all possible constituencies 
to one. Prlsclan's llst of constructions must be extended in some directions 
such as to include the genitive (dependent on the nearest noun) and the 
ablative. 
The deep structure of the Latin sentence can be shown to consist of 
a left side containing, as in Chinese, a cluster of presentences, labels 
and topics, and a right side, which proceeds logically towards the speaker's 
comment on the main topic. The items of the deep structure are formed into 
a syntax tree structure (slmilar to the Russlan~-structure) by virtue 
of their membership in constructions legal in Latin syntax. 
The first word of the sentence, if inflected, has a constructional 
potentiality assigned to each possible inflectional segment metched in a 
look-up of inflectional morphemes. Each inflectional morpheme has a code 
of constructional potentiality assigning to it all possible syntactic roles. 
The probability factor is not the same for this potentiality in the first 
position as for this potentiality further to the right in the sentence. 
The parsing program assigns construction ~des to each possible construc- 
tional pair until it arrives at a series of surface-level hmodes which make 
a well-formed~-structure. 
By adding to the dlctionarythe constructlonal potentiality of all 
lexical items which are basic verbs one can assign values and sem--tic features 
to the words belonging to the multl-member constructions of transitive and 
bl-transitlve verbs. -2 
