English. This paper describes how compounding is treated in the Word Formation Latin derivational lexicon. Through the analysis of some types of Latin compounds, perspectives and limitations of the resource are highlighted; its contribution to theoretical and computational linguistic issues is also outlined.
Word Formation Latin (WFL,
2Among them, notable ones are the lexical network for
Czech DeriNet
The aim of this paper is twofold: on the one hand, it describes how compounding is represented into the WFL derivational lexicon; on the other hand, it aims at highlighting the theoretical and computational contribution of this resource through the analysis of some aspects (i.e. WFRs, input and output lexical categories) of the Latin compounds collected in it. 2
Compared to other Indo-European languages
(e.g. Sanskrit or Greek), compounding in Latin
is generally considered to be not very productive.
According to (Grenier, 1912) and
In the last decades, Latin compounding
(henceforth LC) has received more attention (
The nature of the linking element -i-,3 the relationship between compounding and derivation in Latin, and the classification of Latin compounds, are the main theoretical topics on which attention is focused. However, there are still many questions that so far could not be answered exhaustively due to the scarcity of data collected so far: which were the most productive types of compound in Latin? Through which rules were Latin compounds formed? What PoS did Latin compounds consist of most frequently? What kinds of meaning are expressed by compounding in Latin? WFL allows to fill to answer these questions by providing a large account of quantitative data which can help to better understand the mechanisms of LC. 3
The methodology behind WFL is consistent
with the Item-and-Arrangement model outlined in
During the compilation of WFL, an initial list of
possible compounds has been drawn by taking into
account all possible combinations of V (verb), N
(nouns), A (adjectives), PR (pronouns), and I
(invariables - e.g. adverbs). Some categories have
been filled semi-automatically with the help of
SQL queries. These usually matched a string that
combines a certain lexical element + -i- + another
3A survey of the literature on the nature of this linking
element is in
For each compound, a derivational tree-graph is provided (as in Figure 1). In each graph, nodes are lemmas, and edges are relations showing the kind of WFR involved. Special provisions are made in order to collapse and hide compounding relations according to the user’s choice. This is useful when very productive constituents are displayed in massive multi-tree graphs. The sample collected from the WFL lexical basis consists of 1744 compounds. The fact that all compounds collected from the three dictionaries mentioned above are for the first time categorised and labelled into a language resource allows for a more in-depth overview and for a quantitative analysis on many aspects of LC (e.g. productivity, WFRs, lexical categories involved in compounding). In the following sections, some preliminary considerations on the data currently included in WFL are provided. 3.1
Compound words collected in WFL are created through 59 WFRs. In table 1, the first twenty most productive WFRs are shown.4 The most productive pattern in LC is Noun+Verb: the rule creates both adjectives and nouns, e.g. soporifer ‘soporific’ (sopor+fero) or artifex ‘artisan’(ars+facio). This word formation process is no longer productive in Romance Languages, in which the reverse order (i.e. the Verb+Noun pattern, e.g. Italian portafoglio ‘wallet’ or French porte-parole ‘spokesman’) is the most frequent. In almost all cases, Latin compounds are made up of two constituents. There are only very few (and not productive) cases in which there are three elements, e.g. turpilucricupidus (turpis ’vile’ + lucrum ’gain’ + cupidus ’desirous’; WFR: A+N+N=N) or suovetaurilia (sus ’pig’ + ovis ’sheep’ + taurus ’bull’; WFR: N+N+N=N). The V+V pattern, that in Italian creates nouns (e.g. dormiveglia ‘half-sleep’, lit. ‘to sleep-to stay awake’), in Latin forms mainly new verbs, such as 4N: noun; V: verb; A: adjective; I: invariable form (i.e. adverb, conjunction); PR: pronoun. patefacio ‘to reveal’ (pateo ’to be evident’ + facio ’to do’).
In addiction to other patterns already identified as productive in previous literature (i.e. A+N=A, N+N=N, N+N=A), it is interesting to notice the presence of a significant number of compounds consisting of two invariable forms (e.g. etiamtum, etiam+tum ‘even then, yet’) or two pronouns (e.g. aliquis, alis+quis ‘anyone, someone’) which are generally neglected in studies on Latin wordformation.
As already pointed out by
Table 2 shows the quantitative distribution of the lexical categories (i.e. how many times adjectives are present as the input or as the output PoS) in WFL compounds. More than half of the sample (i.e. 1089 forms, 62.7%) has a verbal second element (e.g. compounds with -facio or a related stem, such as aedifico ‘to build’ or candefacio ‘to whitewash’).
As far as the output of whole compounds are concerned, it is worth noticing that LC creates mostly adjectives (e.g. compounds with -fer as second constituent, such as alifer ‘winged’), followed by nouns and verbs. Conversely, in Romance languages, compounding is exploited to create primarily nouns and less frequently adjectives. In Italian, there are very few cases of verbs obtained through compounding, which are made up of a noun and a verb (e.g. manomettere ‘to tamper 5However, as reported below in section 3.3, in order to interpret correctly the data in Table 2, a distinction should be made between adjectives and adjectival participles, which are categorised here as V. with’); the formation of pronouns and invariable forms through compounding does not seem to be productive anymore.
The main bedrock of WFL methodology lies in its strict relation to the morphological analyser Lemlat and on the PoS categorisation dictated by its lexical basis. As a consequence, the way compound constituents are pigeonholed can sometimes be unconventional. This impacts the representation of compounds in WFL in the following ways: 1. Adjectives that derive or function like participles are not included in the Lemlat lexical basis, because they are seen as part of the verbal paradigm, this means that certain compounds that would be expected to have a A as one of their constituents have a V instead. e.g altivolans (altus + volo) ‘high flying’ can be found among V+V=A compounds rather than among A+V=A. 2. certain type of adverbs ending in -e are considered in Lemlat ablative cases of the adjectival declension, so dulciloquus (dulce + loquor) ‘sweet talking’ is to be found among A+V=A, rather than I+V=A.
Another principle lying behind WFL’s
methodology is that Oxford Latin Dictionary acts like a
sort of manual for solving a number of
theoretical issues. For instance, unlike some traditional
studies on Latin word-formation (i.e.
This paper has provided an overview of how
compounding is represented in WFL, a derivational
lexicon for Latin. This preliminary study, with
its quantitative analysis in the field of LC, shows
the potential for raising new questions and issues
offered by a resource that for the first time
collects all compounds used in Classical Latin. For
instance, representing all compounding rules into
a network, as it has been already successfully done
for the affixal rules listed in WFL,
Peter G.W. Glare. 1982. Oxford Latin Dictionary.
Oxford University Press, Oxford.
Albert Grenier. 1912. Ètude sur la formation et l’emploi des composès nominaux dans le latin archaique. Berger-Levrault, Paris.