This report presents the work accomplished by all partners of the GENELEX France consortium for the definition of a model describing the morphological aspects of an electronic dictionary.

The issue involved in the development of GENELEX (GENEric LEXicon) is the generality of its format, which covers the following aspects:

• maximum coverage: to take into account, for a given data entry, the maximum amount of non-redundant linguistic information. Information that can be deduced systematically from other linguistic information will not be recorded. In addition, the information to be recorded must not be limited by the needs of specific applications.

maximum portability: to be able to support various types of implementations, the GENELEX model must be a semantic and not a physical model of data. GENELEX pursues exactly the same objectives as the Text Encoding Initiative. The realisation of the GENELEX semantic data model was therefore conceived independently of physical models, permitting our partners to choose different physical models for the implementation of GENELEX dictionaries.

minimum discrimination: this project strives not to divide, but to unite... We have therefore searched to remain as independent as possible of a particular linguistic theory. For this reason, the linguistic aspect has been reduced to accommodate the needs of all users, who may then, with formal rules of translation and/or deduction, find the information of interest in a GENELEX dictionary.

All of our partners have distinguished and separated from one another linguistic objects (words), the semantic model, and the physical model of data. We have favored the strategy that seeks for the greatest isomorphy -- or the smallest distance -- between linguistic objects and objects of the semantic model, without concern for the physical model. We call our process the principle of linguistic realism.

In the description of the morphological layer, we have included both recommendations and warnings so that potential users of the GENELEX format may make the best use of it. The GENELEX format provides quality assurance, especially in the transcription of an editorial dictionary to an electronic version or when merging several electronic dictionaries.

This document addresses all potential users of the GENELEX format: lexicographers, linguists, computer scientists and decision-makers. We have presented the issues at stake and justified our choice. Certain points may therefore appear elementary and perhaps even useless. We apologise in advance and ask for your patience and understanding as our document was conceived for general usage.

Certain conventions have been adopted in this document which should aid the reader's comprehension.

Capitalised words indicate that the concept belongs to the GENELEX model; however, it is often a concept known outside this specific domain. For example, we may refer to a noun in the general sense, or we may refer to a Noun as a potential value for the attribute, Grammatical Category. In some cases the capitalised concepts may be followed by an acronym used in future references: Morphological Unit (Um).

This document presents the GENELEX model as instantiated for the French language, therefore the examples are necessarily in French. These are often accompanied by an annotated translation in English. These annotations refer to those concepts, such as gender and number, that may be obvious to a French speaker because of morphological clues but which may escape the English reader who is unfamiliar with French.

In general, the French text appears in italics, and its English translation in reduced normal fonts. Certain morphological aspects may be highlighted in bold.

The examples will be presented as follows: for a Morphological Unit, the lemmatised form is in bold separated from the Inflected Forms by a colon. The Inflected Forms appear in italics and are set off by curly brackets. The Grammatical Category appears in the right-hand margin.

Each translated example is followed by its English translation which may include either or both the semantic and literal translation. The translation appears in a smaller, normal font and will be annotated to reflect gender and number of the nouns, adjectives, and conjugations.

The key words of the GENELEX model are expressed in French, such as the names of relations, entities, fields and their values. The following is list of the GENELEX vocabulary as it appears in the original model, or simply in the text, and its English equivalent.

A Morphological Unit (Um) is a regrouping of words based on morphological properties.

An uninflected Morphological Unit is defined by associating a written form (with its variants) with a Grammatical Category.

An inflected Morphological Unit is defined by an equivalent class of Inflected Forms and by a Grammatical Category (refer to the paragraph on Inflected Forms).

This Morphological Unit is identified by its graphic and/or phonemic lemma. The lemmatised form is the singular form if there is a variation in number, the masculine form if there is a variation in gender, and the infinitive for all verbs.

Unless, of course, if the specified form does not exist. In effect, certain Nouns or adjectives are defective in the singular or the masculine form.

Based on equivalent classes of Inflected Forms, we distinguish in GENELEX:

i. the Inflected Forms of lemmatised forms

ii. the homographic, Inflected Forms of a lemmatised form of another equivalent class

iii. the homographs with or without an equivalent class of Inflected Forms

iv. certain homographs of different categories

v. certain homographs of an identical Category but with different meanings

Two Morphological Units may thus be distinguished although they are homographs or homophones -- in other words, homonyms. On the other hand, we consider a word to be polysemic if it has several meanings and is represented by a single Morphological Unit.

This definition is necessary but insufficient.

Indeed it is circular in that the notion of equivalent classes of Inflected Forms is based on the notion of lemmas. Numerous homographs are processed in the manner described above, but two cases in particular are problematic:

i. an equivalent class of Inflected Forms properly included in another

ii. two equivalent classes of Inflected Forms that are differentiated only by their Grammatical Category.

We must therefore establish formal criteria for splitting Morphological Units.

We have adopted various criteria for splitting Morphological Units, the most important being the criterion of Grammatical Category because of its generality. All other criteria have methods of application dependent on the specific Category.

All of the criteria which will be presented should be considered as recommendations. The lexicographer is free to choose other criteria as long as they are applied to all of the data.

However, all criteria to be presented are sufficient criteria and insure consistency in data encoding.

If two distinct syntactic categories can be associated with a lemmatised form, then we may consider that we have at hand two distinct lemmas.

It should be noted, however, that it is difficult to distinguish certain categories, such as:

á Noun/Adj: (see section III.3)

á Past Participles/Adj, Present Participles/Adj: the past and present participles may be used as adjectives. The opposite is not true.

á phenomena with several derivations, that is, past or present participles used as adjectives, then as Nouns.

These difficulties result from the often imprecise delimitation between nature and function and from the arbitrariness of the nature of these cases. When do we record as adjectives in the dictionary the present and past participles used as such? The same goes for nominalized adjectives.

In GENELEX the coding of grammatical categories is assigned entirely to lexicographers, who have complete control over the status of the categories. Ultimately, they are the ones responsible for the final decision.

All grammatical subcategories concerning Nouns, Determiners and Adjectives are also criteria for splitting lemmas.

á distinction between proper noun/common noun

á distinction between strong and weak personal pronouns

á distinction between indefinite and qualifying adjectives

Gender is a criterion for splitting, except when the variation in gender:

á reflects uniquely the sex of the named animal/person.

á is a free variation on the arbitrary gender, the two genders having the same meaning.

Number, as a system of determination, sometimes strongly alters the meaning of Nouns.

á generic, non-specific, and specific determination.

á determination of discreteness, mass, and abstraction.

Here, we are dealing with the morphology of Nouns independently of the operations that they may undergo. If the difference in meaning is tied only to the determination caused by the number, we do not consider number as a criterion for splitting. In other words, it is a matter of polysemy and not homonymy.

For all other cases, one must refer to meaning as a criterion for splitting. However, we do have reservations about this criterion as well, and prefer to consider that number is not globally an operating criterion for splitting.

However, with full knowledge of the facts, the lexicographer may choose a different option.

Meaning may be a criterion for splitting if no relations (etymological, derivational, rhetorical) can be established between two meanings of a word. Unfortunately, judgement may vary as to the existence or non-existence of such a relation. For this reason, we do not possess formal criteria to separate homonymy and polysemy. Here again, the lexicographer has carte blanche in his choice.

We wish to recall however that the only homogeneous processing envisaged is total polysemy (all meanings which do not present any difference in terms of their variation of form and their Grammatical Category are grouped under the same Um) and that a homogeneous processing of dictionaries improves the performance of their tools for fusion.

It is agreed that there are no inflectional cases in modern French. However, the weak personal pronouns and the relative and interrogative pronouns all have a grammatical function.

It would be difficult however to attribute to all of these pronouns one of the cases from among the following: nominative, accusative, dative, genitive, and locative.

We prefer rather to refer to the function of the pronouns as a criteria for splitting (see also V.2.2 Inflection of Determiners and Pronouns.). Since function does not explicitly appear at the morphological level; it is not represented in the Ums.

We have considered function as a criterion for splitting (refer also to Pronominal Inflections).

Etymology and derivation are not considered as operating criteria for splitting.

A lemma may have several concurrent etymons, a chain of etymons, or several etymons (composition or lexical derivation).

Examples taken from the Petit Robert:

We see also that two distinct lemmas may have identical etymons.

When two different etymologies may be associated with a single word, it is sometimes difficult to decide whether we have at hand concurrent etymologies or different lemmas.

We therefore believe that etymology is not a sufficient criterion for splitting lemmas.

A lemma may be the root for several Derivatives of the same Category, whether or not it is polysemic or homonymic. Consequently, derivation is not a sufficient criterion for splitting lemmas.

In GENELEX, we prefer to have three verb entries (coller, griller, raffiner). Derivation is not sufficient to discern between polysemy and homonymy, the eternal dilemma of lexicographers. (For example, in the Dictionnaire Hachette du français, one finds for these given examples handled as follows: coller - one meaning; griller - multiple meanings; raffiner - two homonyms.)

Autonomous Morphological Units are units which stand alone in a language, as opposed to those which appear only in other Morphological Units, in idioms, in hyphenated words or in fossilised expressions. They all possess a Grammatical Category, one or more syntactic behaviors and possibly one or more semantic meanings.

Autonomous units may be Simple or compound, and they may or may not have derivation relations with other units.

Non-autonomous Morphological Units may be either Simple or Affix. They are employed only in Derivatives, phrases, compound words or set expressions.

Non-autonomous Affix Um : they appear only in derivations ( see Section 3).

Non-autonomous Simple Um : they are found only in compound Morphological Units (that includes compound words, locutions, and fixed expressions). GENELEX has chosen to mark their non-autonomous status. A Category may possibly be attributed to them on an etymological basis or in an arbitrary manner.

Note: The term lemma is often mistakenly used in place of Morphological Unit, although not all Morphological Units are lemmas. (for example: non-autonomous Morphological Units).

A Simple Morphological Unit (Um_S) is associated with a written form (or several in the case of variants) composed of a series of alphabetical characters [a-z, A-Z, ç, é, è, ‘, ê, ‡, à, ‰, Š, ¾, oe, ", •, ô, š, —, ù, û, Ÿ, ƒ, é, è, æ, ç, Ë, å, €, ë, ì, ï, …, î, ô, ó, †] , separators (hyphens, apostrophe, period),and possibly punctuation indicating hyphenation.

A Simple Morphological Unit may have a grammatical category as well as a subcategory. Whether the Um_S is autonomous or not is also indicated (See Section 1.4.).

An Affix Morphological Unit (Um_Aff) may be one of the following types: prefix, infix, or suffix, or of neither of these classes in the case in which it derives its status from the context of its derivation or its composition.

Derivation traces back to a deep structure of lexical morphology. One finds in the lexicon basic lemmas and lexical morphemes of derivation. Morphological derivation is very productive in French: it is estimated that some 35,000 lexical morphemes are considered results of derivation in the dictionaries.

conceive the processes of lexical formation by derivation as linguistic objects and describe all morphological derivational rules while taking into consideration, for example, the editorial practices of traditional dictionaries.

consider that derivation is essentially a syntactic phenomenon and that it is useless to describe systematically morphological information that can only be redundant in the first place. In this case, we appeal to a derivational process in morphology only for those words whose syntax is not well known. (such as cerise/cerisaie: cherry tree/cherry tree orchard).

Morphological derivation is generally dependent on the length of character chains.

The same step of derivation may combine prefixes and suffixes.

The trees of derivation are n-ary and not necessarily binary. Indeed, we remind the reader that at each step, the derivation must apply to autonomous bases.

where a base word, a prefix and a suffix are combined in a single step.

The derivation tree may in certain cases by unary, indicating either an improper derivation or a partially represented derivation. The lexicographer is free to denote a relationship between two Simple Morphological Units without specifying the mechanisms of prefixing or suffixing brought into play.

The derivation components, which have been recorded, are identified by a rank, indicating their order of appearance in the Derivative. In the case of partially represented derivations, no indication of rank is noted. This lack of rank distinguishes them from improper derivations whose rank is indicated by the ordinal number 1.

The system of inflection and the Category of the derived form are determined by the affix and the base word.

Prefixes rarely cause a change in class. The Grammatical Category and the system of inflection therefore remain the same as those of the base word.

There are however several exceptions:

Suffixes may or may not entail a change in class. It is the suffix that determines the necessity of such a change.

Other suffixes cause a change in Category. They constrain the Category of the base and derived forms.

These suffixes however do not always allow one to deduce unequivocally the Category of the base and of the Derivative. Certain suffixes may be applied to a group of categories, as far as both selection and production are concerned.

In the case in which the suffix produces a noun, the suffix may also determine the gender of the derivative.

However, in all cases, the affix is the determinant as its value permits us to predict systematically the categories of the base and the Derivative. An affix must therefore carry the following information: the Category selected, the resulting Category, and the Gender of the resulting nouns. Based on the prescribed information, we may provide help for decisions and consistency checks for the derived Morphological Units.

The same may be said about inflectional system associated with a derivative: it is often deducible from the suffix when the latter possesses an Inflected Form. This is not, however, systematic and therefore we have decided not to make a constraint of it.

The derivation relation refers to Ums (Simple or Affix) and indicates for each, its linear order of appearance and its status in the derivation.

The derivation is a relation between different Ums and is applied by default to all of their graphic and phonemic Morphological Units (Umgs and Umps).

In certain cases however, we may specify that for a given derivation of a Um and its derivation components, the recorded derivation should apply only to certain ones of its possible graphic or phonemic variants, and to certain ones of its graphic or phonemic radicals. The radicals and the variants are referenced by their number.

For each affix, we specify which of its combinatory variants (refer to II.2.3 Radicals and their Combinatory Variants) plays a role in the process of derivation. For this reason we number the radical of the affix.

The prefix in, for example, has various surface forms according to the base word to which it is attached. Here we have a case of regressive assimilation.

The suffix tion also has various forms depending on the base word with which it is associated.

For each base word, we specify which of its radicals is active during the derivation process.

In most cases, the Umg (graphic Morphological Unit) of the base serves to form the Derivative. The radical is equivalent therefore to the Umg .

In other cases, we observe a change in the base during the process of derivation.

Reduplication of consonants (associated with denasalization)

The Derivative takes a form that is not identical to that of the base.

In order to describe a derivation process, one must first identify the base in the modern language from which the word was derived.

During the derivation process, phonological and contextual variations on the base word may be observed. Retrograde alterations are characterised by an alteration on the base word by the right context.

These variations may be more or less significant, especially concerning verbs from the 3rd group.

In addition these variations may have origins that are more or less ancient, which leads us to etymology. As concerns the etymological criteria, lexicography clearly distinguishes two cases:

In these pairs, the base and the Derivative do not have the same etymon. These adjectival derivations are not represented by a morphological derivation but rather by either a syntactic derivation (relationship between two syntactic units) or by a semantic derivation (relationship between two semantic units).

The definition of a compound is a highly controversial question. We do not claim to have settled these issues.

GENELEX provides a framework that allows the realisation of certain complex expressions from the morphological level. Phrases (afin de - in order to), compounds (garde-malade - home nurse), certain synapses [Benveniste 1974] (fils à papa - daddy's boy) or set expressions (se tourner les pouces - twiddle one's thumbs) may therefore be recorded as Compound Morphological Units (Um_C) by the lexicographer who so desires.

The reader should be forewarned about the terminology concerning complex expressions used in GENELEX. The term, "Compound Morphological Unit" does not cover the traditional notion of "compound word". Formally, it deals only with those complex expressions that a lexicographer has chosen to encode in the morphological layer of the GENELEX model. This choice is conceptually based on a certain number of linguistic criteria that in effect define the very notion of a Um_C.

You will find, however, for purposes of illustration, that we have established criteria for definition in order to validate our model. The users of GENELEX may decide either to utilise or amend the model.

By Morphological Compound, we mean the complex expressions that satisfy at least one of the following criteria:

vi. non-existence of a semantic composition

The Nouns referring to compound numbers are not registered, because they define an infinite set.

None of theses criteria are necessary. Only (ii) is sufficient.

A Compound Um is defined by its components which, by definition, must already exist. A component may be a Simple, autonomous Um, a Simple, non-autonomous Um, an Affix Um, a contracted Um, or a Compound Um. The components of a given compound may be of different types.

A Compound Morphological Unit is distinguished from other Ums by the fact that it has no graphic or phonemic Morphological Units. However, it has a Grammatical Category, and perhaps sub-categories.

From a single Compound Morphological Unit branch N Composition relations , N being the number of Components.

Each Composition relation associates with the Compound:

and consists of (from attributes on the relation itself) the following information:

The graphic (and phonetic) form of a Compound Um is deducible from the form of its components carrying information of the punctuation marks (presence or absence of hyphens, spaces, or apostrophes) between the components. Consequently, it is not necessary to note the Umgs of Compound Ums in the model.

A Compound Um is therefore only indirectly identified by a series of alphabetic characters: [a-z, A-Z, ç, é, è, ‘, ê, ‡, à, ‰, Š, ¾, oe, , •, ô, š, —, ù, û, Ÿ, ƒ, é, è, æ, ç, Ë, å, €, ë, ì, ï, …, î, ô, ó, †] as well as space marks, hyphens, apostrophes and underlines.

One may wish to consider certain units as Compound units despite the fact that they do not have a punctuation mark indicating hyphenation. The absence of such a graphic separator is indicated in the model by an attribute, "joined", on the composition relation.

If a component has graphic variations, then the composition inherits these by default, that is if no restrictions are indicated on the composition relation.

Compounds, as other Morphological Units, are inflected. That is, the graphic and phonemic Inflected Form may be deduced from its system of inflection. (Refer to V.3 System of Inflection for Compounds).

Non-autonomous Compound Ums do not exist.

The number of components is equal to at least two.

A Compound Um may have the same Um as a component more than one time.

The components may be Ums of various types: simple words (autonomous or not), Derivatives, affixes, or compound words.

Short or abbreviated forms are represented in the same way as Morphological Units (Simple or Compound), that is, as a separate entry related to the corresponding developed unit. The relation has attributes in order to indicate if the short form is an abbreviation, a set of initials, or an acronym.

The different types of abbreviations may be distinguished by the usage attributes (frequency, dating, ...) . For example, certain abbreviations are viewed as simple graphic shortcuts (fam. for familier) whereas others have become a part of the language and replace either entirely or partially the original expanded form (métro for métropolitain).

A Um may have several short forms.

Inversely, a Um may in certain cases be the short form of several expanded Ums.

The lexicographer will decide, of course, the importance of these factorisations.

We have decided to make two distinct Morphological Units for the short form and for the expanded form rather than two variations of the same unit for the following reasons:

the short form may sometimes have a different syntactic behaviour than that of the expanded form: one may refer to a petite P.M.E. (semantically, a small business) but never to a petite Petite et Moyenne Enterprise (literally, a small Small and Medium (sized) Enterprise; P.M.E. is a classification of businesses ).

The Contracted ("agglutinated") Morphological Unit (Um_Agg) is used to record written contractions of two units.

The Um_Agg does not possess a Grammatical Category. It is represented in the model as having a composition relation with the elements from which it is derived:

Unlike the case of the Compound Unit, the spelling of the Um_Agg is not obtained by concatenating the spelling of the components. Therefore, the Um_Agg has its own Umg and Ump.

The lexicographer has the choice of entering the contracted paradigms or not.

The Um_Agg is also characterised by the attribute indicating whether it is mandatory or optional. This feature, as well as the facility for recording paradigms, would be particularly adapted to those languages that have a higher frequency of contraction than in French. For example, in Italian, contraction of the preposition con with the definite article is optional.

There are various types of information to be processed in the morphological layer: graphical, phonemic, morphological features, grammatical categories, and etymological.

So as the GENELEX morphological layer may correspond to the linguistic notion of morphology, we have not limited morphology to the graphic form as have many electronic dictionaries for application. This is validated in our model by two fundamental points: the distinction between homographs (refer to Section I.1 Definition of Morphological Units) and the grouping of graphic variants (refer to Section 2.2 below).

Morphology is not limited to the written form because we must also consider the phonetic form.

A Morphological Unit is an abstraction of graphic and phonemic Morphological Units. This abstract entity serves as a pivot between the morphological, syntactic and semantic layers of our model.

Morphological units are identified by their written or graphic form, that is, by a series of alphabetic characters with or without accent marks, in upper or lower case letters: [a-z, A-Z, ç, é, è, ‘, ê, ‡, à, ‰, Š, ¾, oe, ", •, ô, š, —, ù, û, Ÿ, ƒ, é, è, æ, ç, Ë, å, €, ë, ì, ï, …, î, ô, ó, †] as well as space marks, hyphens, apostrophes and underlines. The underline is used strictly as a syllabication mark.

As for those characters which carry an accent mark, we conform to the Text Encoding Initiative, which has chosen the ISO-8879 norm. This norm codes the accented characters in the form of publicly declared SGML entities, or "character-entities." This coding is formally non-ambiguous as the nomenclature of character-entities is part of the ISO norm and that these entities are demarcated between opening and closing marks (&) and (;).

This notation permits us to convert systematically character-entities into accented characters dependent on the operating system or word processing program without running the risk of errors by simple substitution of a chain of characters.

The SGML coding must not only serve to exchange and transfer data but must also remain completely transparent to the user of a GENELEX electronic dictionary. To conserve it as an editing format would make the document illegible and render it a source of errors.

The graphic form of a Compound is computed from the graphic forms of its components.

When, for a single Morphological Unit, various lemmatised forms are concurrent (whatever the context), we have at hand a case of graphic variants. [M. Mathieu-Colas, 1987]

As variants, these forms are logically attached to the same lemma rather than split into different lemmas.

For those words which may or may not include a graphic separator, the lexicographer is free to represent this by either a Compound Morphological Unit with variations on the separator or by a Simple unit and its variants.

Dating, frequency of usage and style may serve to distinguish the variants.

However, such differentiations are not always possible, creating a lexicographic problem of recognition for an independent entry.

The notion of "entry" stems from the linear order of traditional, printed dictionaries. As GENELEX is an electronic dictionary, it is not necessary to conserve this notion. All variants may be placed on the same level if we do not assign an entry header. In this case, only the information attached (regionalism, dating, frequency and style) will vary.

The graphic form(s) of a Compound is(are) deduced from the components rather than recorded directly under a single(or more) Compound Umg(s).

The graphic variations on the Compound have two origins:

variation on the separator preceding a component (the list of separators preceding the component is indicated in the relation R_Compose).

graphic variation of the component

We allow however, the possibility of not systematically applying the graphic variation of the component to the Compound.

The derivation of all graphic forms of a Compound in full consists of listing all combinatorials of the sources of variation.

In certain cases, the extensive list produced may be very long.

We will take "radical" in its traditional sense of a "bare" (base) morpheme (form 0) pertinent for a Umg. Radicals are used in the process of inflection and derivation.

In certain cases, the Umg itself constitutes this form 0 and is noted under Radg0.

Note: by definition Radg0 always has the same value as the corresponding Umg. Referring to Radg0 consequently amounts to referring to the Umg. It is therefore not necessary to implement Radg0.

In other cases, the "bare" morpheme is distinct from the Umg. It is noted under Radg1, Radg2, ..., RadgN

Note that the radical is where combinatory variants (as opposed to free variants at the level of the Umg) are expressed.

All variants of a Um that are not free but combinatory are recorded under Radg and not under Umg.

combinatory variants of Simple Morphological Units

combinatory variants of affixes

impact of combinatory variants on Compound Morphological Units

Combinatory variants may have a label in the attribute "contexte_var."

A phonemic transcription identifies the pronunciation of a Morphological Unit and its variations that are regular, such as the pronunciation or non-pronunciation of the silent e. The phonetic transcriptions that correspond to these variants, essential for applications, are derivable from the phonemic transcription to which all pertinent lexical information is attached.

A phonemic transcription comprises of a linear list of phonemes corresponding to an alphabet. For French, we give the following example:

[p, t, k, b, d, g, f, s, S, v, z, Z, l, r, m, n, N, J, w, h, i, e, a, y, {, u, o, @, ', +, *, :]

In this alphabet, most of the signs are taken from the International Phonetic Alphabet (IPA). Others have a special interpretation.

The mark of syllabication (underline) may be introduced in the phonemic transcription.

When, for a single Morphological Unit, there are concurrent pronunciations (whatever the context) with no regular phonetic variations, we attribute to them all of the necessary distinct phonemic transcriptions.

As variants, these forms are attached to the same Morphological Unit. Dating, frequency and style may distinguish the phonemic variants.

but such is not always the case:

As for graphic Morphological Units, we do not consider the notion of entry header.

Refer to Free Written Variants of a Compound.

We treat the term "radical" in its traditional sense of a "bare" morpheme (form 0) pertinent for a Ump. The radicals are used in the process of inflection and derivation.

In certain cases, the Ump itself constitutes the form 0 and is therefore noted under Radp0.

In other cases, the "bare" morpheme is distinct from the Ump. Here we note it under Radp1, Radp2,..., RadpN.

Note: By definition Radp0 always has the same value as the corresponding Ump. Referring to Radp0 consequently amounts to referring to the Ump. It is therefore not necessary to implement Radp0.

Note also that the radical is where combinatory variants (as opposed to free variants at the level of the Ump) are expressed. For this reason the combinatory variants of affixes are recorded under Radp and not Ump.

For a given Morphological Unit we observe a symmetric axis between the graphic representation of a lemma and its phonemic form. We find on both sides of this axis the graphic and phonemic variations as well as the graphic and phonemic systems of inflection. The combinations of Morphological Features are shared by these two systems.

The consortium has pondered for a long time on whether GENELEX should include all the traditional parts of speech (nouns, verbs, articles, adjectives, pronouns, prepositions, adverbs, conjunctions, interjections and particles).

The principal debates have concerned the Category "determinative," the grouping of the categories "Nouns" and "adjectives," and the status of Past Participles.

Several syntactic theories do not agree on the definition (category, function, or both) and range of determiners, although they all appeal to the notion. All agree on the traditional concepts -- articles, adjectives, indefinites, cardinal numbers, possessives, demonstratives and interrogatives. On the other hand, some include whereas others exclude the forms with de (adverbs, or SN of quantity).

simple forms : adverbs of quantity

complex forms: SN of quantity (N or Pro Head).

We have chosen to introduce Determiner ("Déterminant") in the list of Categories. The traditional possessive adjectives have played a decisive role in our choice. Indeed, the difference in the written forms of notre / nôtre is not a free variation (as we have defined it) for the realization of the adjective. Here, it is the case of two distinct lemmas distinguished only by Category. (refer to Section I.1.2 on Criteria for Splitting Lemmas).

We have adopted the most common point of view:

i. We define the following Categories: Noun, Adjective, Determiner, Adverb, Verb, Preposition, Pronoun, Interjection, Conjunction and Particle.

ii. As with Adverbs, Determiners may be considered as a category or as a function. This function, however, does not appear in the morphological layer of the dictionary.

iii. The Determiners cover the traditional categories of articles, interrogative and demonstrative adjectives, indefinite adjectives, possessives and cardinal numbers.

We exclude the forms introducing de: Adverbs, Pronouns and Nouns, as these analyses of quantitative and partitive expressions are numerous and varied. Note that the behavior of Adverbs, Nouns and Pronouns which may appear in these expressions are at any rate described in the syntactic layer. This information is therefore not lost. Those who wish to consider Adverbs, Nouns and Pronouns of quantity as Determiners may consequently reinsert the pertinent information in a dictionary for application and verify that the Ums of these categories have the following syntactic properties:

1- SN[SPEC[#] ... N] or SN[...#... de SN]

2- P[il y en a #]

Complex Determiners may be processed in GENELEX as all complex forms. In other words, they may be recorded at various levels of the model: at the morphological level (la moitié, la plupart, un peu), at the syntactic level (deux douzaines de kilos de pommes de terre), and at the semantic level (un ab"me de plaisirs). As such, the Dét Category behaves like Advs, Preps and Conjs, which may also have simple or complex surface realizations.

We could well have grouped Nouns and adjectives in the same Category (Noun-adj), as several morphological and syntactic reasons support this treatment [Noailly 85]. The two categories carry the marks of gender and number. Adjectives may also be nominalized. Nouns may appear in the same positions as adjectives.

In fact, this question revives the ancient debate on the distinction between nature and function. Only nature (absolute property) should be recorded in the lexicon [Milner 90]. Therefore, in the case of Nouns and adjectives, the fact that certain functions are defining criteria for a category does not exclude another category from satisfying the requirements. Although a Noun may fulfill the function of a nominal modifier (epithet) or an object of a copula (attribute), it nevertheless remains fundamentally a Noun .

On the other hand, not all adjectives may be nominalized.

in the case of adjectives having a pronominal equivalent (exact or approximate)

in the case of ellipses, as opposed to nominalizations.

We distinguish Nouns from Adjectives in the last of the Grammatical Categories.

The Category recorded in morphology, however, could be identical to (in the majority of cases) or distinct from the Category recorded in syntax. In terms of integrity, this difference is authorized only if there is a gap between the Grammatical Category and the functional Category (verb / adverb, for example, is unlikely).

We could well have processed Past Participles systematically as deverbative adjectives or as Inflected Forms of verbs.

In fact all Past Participles cannot be employed systematically as adjectives. This is the case with participles formed from intransitive verbs (in the sense of Dubois).

We are therefore forced to place Past Participles in the Inflected Forms of verbs and indeed to introduce the Morphological Feature of gender for verbs although it may not necessarily be pertinent for any of the other Inflected Forms.

Some of the grammatical categories that we have included in the GENELEX model may be sub-typed; thus we may differentiate, for example, proper nouns from common nouns, co-ordinating conjunctions from subordinating conjunctions, etc. We have decided to take these differences into account by attaching a Grammatical Sub-Category to the Category that it specifies.

We propose the following values for the attribute Grammatical Sub-Category: Proper, Common, Possessive, Demonstrative, Partitive, Definite, Indefinite, Cardinal, Ordinal, Exclamative, Qualifying, Interrogative, Relative, Coordination, Subordination, Weak Personal, Strong Personal, Impersonal.

Of course, all combinations of Category/Sub-Category are not legitimate. Here below is a list of the authorised Sub-Categories according to Grammatical Category.

Certain units do not carry any Morphological Features: conjunction, interjection, etc.

The model for the French language possesses the following Morphological Features: Gender, number, number-possessor, person, tense, and mood. It is evident that each language defines its own system of Morphological Features.

Morphological features are difficult to define except by consulting an extensive list of their values and of the categories to which they are applicable.

Morphological features can be combined following different patterns that are applicable for a given type of Um:

i. gender, number .

ii. person, gender, number .

iii. person, gender, number, number-possessor .

iv. mood, tense, person, gender, number.

The various grammatical categories of nouns, adjectives, etc, are partially defined by their morphological characteristics. The general characteristics are as follows:

adjectives, determiners, nouns and pronouns carry the mark of gender and number.

verbs carry the mark of mood, tense, person, gender and number.

adverbs (except tout), prepositions, conjunctions and interjections have no Morphological Features.

The combination of Morphological Features and Category are therefore related. There is not however, a complete overlap between the Category and the authorized combinations of Morphological Features. To illustrate, the adverb tout carries the mark of gender and number in very specific environments (before a feminine word that begins with a consonant).This particular case may be handled by either an adverb that may be inflected or an adverb having two contextual variants tout and toute.

It is nevertheless possible to propose to the lexicographer, default combinations of Morphological Features according to the Grammatical Category. We must nonetheless authorize these combinations in order to guarantee the independence of combinations of Morphological Features from their Category.

Noun, Adjective, Determiner, Pronoun :

Verb:

These values depend on the Category to which gender is applied.

In French, all nouns have a specific, arbitrary gender.

Gender is not necessarily determined by the sex of the denoted person/animal.

grammatical inversion of gender/sex

a single grammatical gender for both sexes (epicene nouns)

When there is a match between gender and sex (masculine-male, feminine-female), we have a case of natural gender. However, we must also note that natural gender itself covers two cases:

i. no morphological variation on the same Morphological Unit (i.e., two completely separate words) (refer to the definition of a Morphological Unit below). In this case, nothing prevents the choice of the gender from being completely arbitrary.

ii. morphological variation on a single Morphological Unit. In this case only gender is determined by the sex of the denoted person/animal.

"The separation of inanimate nouns into feminine and masculine , often interpreted as natural, stems in fact from the metaphorical male/female distinction which influences our perception. In such a manner, certain objects or concepts are conceived, under the influence of an arbitrary grammatical gender, as in essence male or female. In fact, gender often carries a symbolic value. Ex: the opposition between la lune (the moon), symbol of femininity, and le soleil (the sun), symbol of masculinity." [Marina Yaguello 1981].

The morphological variations in gender of some inanimate nouns may fluctuate arbitrarily.

We distinguish two types of variations in the gender of nouns. One is a function of the sex of the denoted (animate nouns). The other is a free variation of an arbitrary gender (inanimate nouns).

In the model, we may:

decide not to take into account such a distinction in morphology. We may compare variation in morphological gender to the denotative constraint "animate"/"inanimate" in semantics. It is for the user to draw the necessary conclusions. In this case, we do not distinguish in morphology

on the other hand, we may decide to take into account the said distinction in morphology. We consider in this case that the variants of gender of inanimate nouns may be processed as orthographic or phonetic variants by associating two Umgs with a Morphological Unit: one for the masculine, and the other for the feminine form. The same goes for two Umps.

Adjectives receive their gender and number from the noun that they modify.

Epicene adjectives do not undergo any spelling change according to gender.

This is also true for cardinal numbers.

It is also the case for "invariable" adjectives (which carry nonetheless Morphological Features), either because they are words of foreign origin or that are of "expressive formation." (cf. Grevisse)

Certain adjectives have only a single gender (cf. Grevisse):

adjectives with limited usage.

adjective that apply only to animate nouns of a single sex.

Determiners receive their gender from the noun that they modify.

For a given written form, certain determiners have only a single value in gender (mainte, certaine, quelle, ma, le, etc), others have several (les, etc).

(Refer also to the Section III.4 Status of Past Participles.)

The rules of agreement for Past Participles determine the authorized Inflected Form among the four possible Inflected Forms of a participle. This is valid for both gender and number.

Auxiliary verbs (être or avoir) and the syntactic Sub-Category (direct or indirect transitives, essential pronominals, direct or indirect reciprocal pronominals, impersonals) of a verb are all criteria in these rules.

If a lexicographer wanted to use this information to code Past Participles in morphology, he must take into account that all information, and in general all information on the formation of complex tenses, is attached to the syntactic layer of the model. In effect, the existence of complex tenses and their auxiliary(ies) is considered as a syntactic phenomenon.

The choice is left up to the lexicographer. He must, however, adhere to the constraint of coherence of the model (correspondence of a group of inflected past participial forms to the syntactic Sub-Category) and refrain from reworking the model.

These constraints of coherence are based on:

i. variable Past Participles

ii. invariable Past Participles

Note: We are aware that the categories of direct transitives, indirect transitives, and intransitives are very schematic and do not alone suffice to take into account all usage.

The defectives are nevertheless registered.

iii. no Past Participles

When a pronoun refers to a noun (anaphorically or deictically), it takes the gender of the noun.

When a pronoun refers to an utterance (anaphoric usage), it is masculine. There are those who would consider it neuter in spite of the fact that the masculine form is the neutralizing form in French. Recall that reference is not necessarily registered in the Morphological Feature of a word (pronoun on morphologically singular, but semantically plural). GENELEX offers the neuter gender (cf. demonstratives and quoi) to the lexicographer to use as he sees fit.

When a pronoun refers to an entity being uttered (deictic usage), it is masculine, feminine or neuter (for demonstratives).

For a given form, certain pronouns have only a single value in gender (il, le, lequel, laquelle, chacun, cela, etc); others may have several (je, les, lui, qui, que, etc).

In French, the possessives receive their gender from the word that they modify. The gender of the possessor plays no role, reducing the situation into the following general case:

possessive determiners have the gender of the noun that they modify.

possessive adjectives have the gender of the noun to which they refer.

possessive pronouns have either the gender of the noun, of the utterance to which they refer (textual anaphora), or of the entity identified deictically.

For a given written form, certain possessive may have two values in gender (votre, notre, nôtre, vôtre).

The majority of nouns vary according to number. Some, however, are used only in the singular (le sud -the South), and others, only in the plural (fiançailles - engagement).

The sense of a noun may vary with number, as the type of determination is often linked to number, which often plays the role of an operator (refer to Section I.1.2.3 Number as a Criterion for Splitting). We must therefore be vigilant in the description of possible forms by anticipating all associated inflections and all cases in which this operator is applicable.

Some nouns are identical in the singular and the plural.

Adjectives (except cardinal numbers) receive their number from the noun that they modify.

Certain Adjectives with limited usage are used only in the plural (cf. Grevisse).

Others have the same spelling in the masculine singular and the masculine plural.

This is also the case of invariable Adjectives (refer to Section IV.3.2 Gender of Adjectives).

Cardinal numbers except un (one) are always plural

Determiners carry the number of the noun that they specify (except in the cases of compound determination that we do not consider as determiners in morphology).

Numeral determiners are always plural except for un. The same is true for the indefinite determiners plusieurs, divers, différents. The determiners aucun and chaque are always singular.

It is associated with person.

Refer to Gender of Past Participles.

Some past participles do not present spelling changes between the masculine singular and the masculine plural.

Pronouns take the number of the noun to which they refer.

When a pronoun refers to an utterance (anaphoric usage), it is singular.

When a pronoun refers to one or several entities being identified deictically, it is either singular or plural.

For a given form, certain pronouns have only a single value in number (autrui, rien, chacun, ils, elles, les, ceux, auxquels, lesquels, les miens, etc), while others may have both values (qui, que, dont, etc).

Some pronouns are always singular: autrui, rien, chacun.

In French, the possessives carry the number of the possessor and that which is possessed.

Possessive determiners therefore carry the number of the noun that they specify as well as the number of the possessor.

The possessive adjectives therefore carry the number of the noun that they modify as well as the number of the possessor.

The possessive pronouns have the same number of the noun that they represent (textual anaphora), or the number of the entity referred to in a statement (deictic use) as well as the number of the possessor.

Lacking a consensus on another model, we have taken mood in the traditional sense. We are aware, however, that the distinction between mood/tense is at times arbitrary, and that it is possible to group together all the pertinent values while prohibiting certain combinations. However, this distinction, besides its consensual, if not traditional, aspect, is actually very useful for constraining morphological behaviour from syntax. In this way, we may distinguish indicative noun clauses from subjunctive noun clauses.

We represent arbitrarily the imperative by mood = imperative, tense = present, the past and present participles respectively by mood = participle, tense = past and by mood = participle, tense = present.

Only simple tenses are noted. Complex tenses formed regularly from simple tenses have been excluded.

The Past Participle is considered as a verbal form, and is not handled systematically as an Adjective. In the morphological model, it is the residual trace of complex tenses which are handled in the syntactic part of the model.

Tense is considered as a syntactic property and is represented in the syntactic layer of the model.

Personal pronouns, as the term implies, always express person.

Relative and interrogative pronouns have forms for virtually all persons possible (except lequel, auquel, duquel, quel, which are only in the third person).

Theories diverge on whether an infinitive, when it has a subject, carries or not the mark of person. GENELEX accepts both points of view, as it does not require that person be applied to all moods (cf. participial mood).

Some verbs are expressed only in the third person:

impersonal verbs (pleuvoir, falloir, s'ensuivre, etc)

certain defective verbs (seoir, échoir, etc).

By "Graphic Inflected Form" we mean a couplet of graphic form-combination of Morphological Features produced by a system of inflection.

Notice that some Inflected Forms are identical to the graphic Morphological Unit. They are nonetheless Inflected Forms. Our definition of Inflected Forms is therefore not a naive definition of "that which is conjugated," "that which is put in the feminine," "that which is pluralized," because the infinitive of Verbs, the singular, masculine form of Adjectives and the singular of Nouns with a single gender are considered as Inflected Forms as well.

Only Verbs, Adjectives, adjectival Determiners and Nouns belong in a system of inflection.

Although relative and interrogative pronouns (except all forms of quel) possess Morphological Features, they are not produced by a system of inflection.

As for prepositions (except contracted forms), adverbs (except tout in certain conditions), interjections and conjunctions, they do not have Morphological Features and are considered "invariable."

By inflection we mean the description of an equivalent class of Inflected Forms. Inflection is therefore not conceived as an action (conjugation, feminization, pluralization), but as a state of existence. An equivalent class of Inflected Forms is viewed in a static and not dynamic manner.

This definition presents two advantages:

1. Information on the graphic form-combination couplet of inflectional features has been completely flattened. Indeed, flattening information can only be advantageous for a generic model.

2. We represent in the same simple model phenomena that are very different and at times very complex.

arbitrary variation

regular variations

irregular variations

specific variations

absence of variation

defectives

We call the method which permits us to construct an equivalent class of Inflected Forms for a Morphological Unit a system of inflection ("Mode de Flexion").

The Inflected Forms derived from a Umg are more or less numerous according to the Category, the maximum being 51.

As Inflected Forms may be derived from a Umg and since there is a great number of common systems of derivation, we do not represent Inflected Forms in the dictionary. Here we adopt the method of traditional dictionaries.

However, we find it necessary to know the system of inflection of a Morphological Unit as it is a piece of information in itself (regular, marginal, and irregular formation) and because it permits us to generate the Inflected Forms.

We wish to remind the reader that in GENELEX, Morphological Units are essentially defined by a Grammatical Category and Sub-Category, and a paradigm of inflection. On this basis, we can group {je, tu, il, elle, nous, vous, ils, elles}{I, you, he; she; we; you; they(m), they(f)} or {ma, ta, sa, mon, ton, son, mes, tes, ses, notre, votre, leur, nos, vos, leurs} {my, your, his/her (possed fs, possor s), my, your, his/her (possed ms , possor s), my, your, his/her (possed m/fp, possor s), our, your, their (possed m/fs, possor p), our, your, their (possed m/fp, possor p)} ,under the same Morphological Unit and consider them as various Inflected Forms of the same entry -- as only the combination of Morphological Features distinguishes them. Notice also that the names "subject personal pronoun," "strong personal pronoun," and "impersonal pronoun" cover precisely the paradigm of inflection.

These entries are identified, as other entries, by their canonical form (il, son). Remember that an electronic dictionary, as opposed to a traditional dictionary, may be consulted by Inflected Forms.

Finally, those who wish to note compounds in their dictionary will see the advantage, indeed necessity, of constructing such groupings.

There are two valid methods for the derivation of Inflected Forms:

addition of an affix to a radical

removal or addition of characters for a graphic or phonemic Morphological Unit.

This method of derivation is particularly useful for describing verbal inflections because:

the radical is always different from the Umg.

at times we must select a radical among several choices.

The radicals are identified by a number (n-th rad). For the derivation of an Inflected Form, we select (by its number) the pertinent radical and add a termination corresponding to the indicated combination of mood-tense-person-gender-number.

It is possible that the radical is identical to the graphic and phonemic Morphological Unit. In this case the radical must carry the number 0.

Inflected forms are derived from a Umg by removing or adding a series of characters. (Recall that the label for a Umg is the same as its radical 0.) In an equivalent class of Inflected Forms, at least one form corresponds to the Umg. For this form the derivation by removal/addition is not necessary.

We authorized a "joker" in the removal and addition of characters. This measure permits us to group together conjugations, notably those with alternations of accent marks and those with reduplication of consonants.

The value of a joker is found by comparing the written lemma and the chain to be removed. A joker, noted as $, may represent one or several characters.

Remove: é$er Add: è$e

Certain paradigms do not adapt well to this method of derivation:

Inflectional variants occur when, for a given Umg, the same combination of inflectional features may have different surface forms.

The inflectional variants may be described in terms of:

i. Variation in the radical. In certain cases, it is possible to use concurrently two different radicals for the same Inflected Form.

ii. Variation in the suffix

iii. Combination of two variations

The variation in inflection may influence either the entire paradigm of inflection or only certain Inflected Forms.

i. verbs that have two regular conjugations

conjugation with oi and oy, and with e and é

conjugation with ou• and conjugation with oi, oy and o

ii. verbs that have variants in certain forms only

There are various "labels" that indicate the source of variation, encoded under "contexte_var".

iii. Nouns that have two plurals, a French and a foreign plural

iv. Verbs that have variants in certain forms only, variants dictated by the type of sentence.

Note: We distinguish between variation in spelling and variation in inflection. Concurrent lemmatized forms are variants of written lemmas. All other forms are inflectional variants.

As for all other Morphological Units, Inflected Forms of Compounds are not given in full but are derived. The Inflected Forms of a Compound are derived from the Inflected Forms of its Components.

Keep in mind that the Inflected Forms of the Components are not stored in full and that each Component has a system of inflection which permits us, based on a Combination of Morphological Features, to derive the desired Inflected Form. To select an Inflected Form of a Component, it suffices to specify the corresponding Combination of Morphological Features.

An Mfc (System of Inflection of Compounds) enters into the relation of Composition (R_Compose) between a Compound and a Component. The Mfc allows us to select, for each CombTM of a Compound, the Inflected Form(s) of the pertinent Components.

Using this system we will elaborate various cases.

Morphological compounds may consist of any type of simple words and be themselves of any Category.

The traditional spelling formed the plural based on the second component: ramasse-miettes, whereas the recent spelling reform proposes the singular form: ramasse-miette.

The attribute contexte_var is used to record this distinction:

Notice that we have also described the process of derivation for the compound épingle à cheveux.

A lemma may have concurrent etymons, a chain of etymons, or several etymons (composition or lexical derivation).

Examples taken from the Petit Robert.

(Refer also to the Section I.1.3 Conclusion on the Criteria for Splitting)

The GENELEX model of morphological data was conceived to represent the variations in form and the grammatical properties of words. In this chapter, we shall present a brief description of the terminology that we have adopted.

The forms of words are grouped into Morphological Units (Um). A Morphological Unit corresponds either to an invariable word: adverb, preposition, etc, or to a paradigm of forms related by inflectional operations: conjugation, change of gender or number, etc. For example, all simples forms derived from conjugation of a verb comes from the same Um. The notion of Um corresponds often to that of "lemmas" in traditional dictionaries, in which only one of the forms (the infinitive) is at the head of an entry.

This grouping has, as an indispensable complement, a set of information on the possible inflectional variations for each Um, represented in the following manner: Each Um is linked by its Umg and Ump and their system of inflection to one or more combinations of Morphological Features (CombTM) in which the pertinent features are recorded: gender, number, tense, etc. For example, the Um of the adjective net is related to four combinations generated by the combinatory variations of gender and number.

Um and CombTM are defined without any reference to the spelling or pronunciation of the words. They concern uniquely the combinatorial aspect of Morphological Feature s, which is often, but not always, reflected in spelling and phonetic variations. For example, the four CombTMs of the adjective net have a total of four different spellings, but no difference of pronunciation in modern French (except the liaison in the plural). Two parallel systems are therefore put into place to represent respectively the spelling and phonetic variations of words. Let us begin our discussion with the spelling variations.

In the absence of free spelling variations, a single Um corresponding to a simple word is related to a one and only one orthographic (graphic) Morphological Unit (Umg) which indicates the spelling of the word, or more precisely, the spelling of its canonical form: the infinitive for a verb, the masculine singular for an Adjective, etc. The distinction between Um and Umg serves to represent free variations.

In the case of nouns, Adjectives and verbs, there is a problem of representing the inflections associated with these categories. For these words, each Umg is related to a system of orthographic inflection (Mfg) which characterizes the manner in which the Inflected Forms are derived from the canonical form. For example, each verbal Umg possesses a conjugation that it shares generally with a number of other verbs. This conjugation is represented in the form of an Mfg.

Given an Mfg and a CombTM, we may associate with them an elementary rule of derivation for the Inflected Forms (Cff). For example, if we consider the conjugation of laisser, represented by an Mfg and the CombTM of the present participle, these two entities are related to a Cff which specifies: remove -er, add, -ant.

Each Umg and Mfg has their equivalent phonemic Morphological Unit: a Ump and a Mfp which correspond to the same information and derivation but which act on the phonemic transcription rather than the graphic form.

The separation of Morphological Units was our first challenge. In the view of normalization, we proposed the following rules.

Because Grammatical Category is considered as a basic grammatical information, we have established it as an attribute of a Um. Homographs of distinct categories give rise to distinct Ums -- such is the case for the adjective ferme (Luc a été ferme sur ce point - Luc was firm on this point) and for the noun ferme (Cette maison est une ferme - This house is on a farm). This rule is more difficult to apply in the case of numerous noun-adjective homographs in which the noun and the adjective have the same CombTM and the same forms -- fragile (Luc est fragile. - Luc est un fragile. (Luc is frail.)). In these type of examples, the precise nature of the Grammatical Category is dependent on syntax; moreover, if the morphological description is limited to a Um characterized as an adjective, it is sufficient to generate all CombTMs and all forms. The lexicographers is therefore allowed to ignore the existence of a noun such as fragile.

The notion of Um is founded upon that of inflection. The delimitation of Morphological Units presupposes a distinction between inflection and derivation.

The passage from masculine to feminine is considered a result of inflection, except when it results in a change of meaning, in which case several Ums are involved. For example, the masculine and feminine forms of a participle or of an adjective such as net, and the forms of a normal [+human] noun such as adepte are grouped into their respective and unique Um. On the other hand, the masculine and feminine forms of manche (La po‘le a un manche. The skillet has a handle (m)- La chemise a une manche trop longue. The shirt has a sleeve (f) that is too long.) are separated into two distinct Ums. Such is also the case for the masculine noun moteur (motor) and the feminine noun motrice (motive) which designate distinct concrete objects. The masculine and feminine forms of the word interface have the same meaning, and therefore are grouped under a single Um. The processing of grammatical gender here goes further than simple morphology. However, it does not go as far as representing sex or distinguishing between arbitrary and natural gender.

The passage between the singular and the plural is always considered as a derivation from the inflectional system, for reasons of simplicity. For example, the noun obligatorily plural échecs is attached to the Um. of the noun échec.

The distinction between inflection and derivation poses problems that lexicographers know only too well. These problems manifest themselves when we search to delimit Ums. Besides the verb agacer, which possesses a past participle agacé in the Um, should we always include an adjective agacé? We do not propose any particular rules in addition to those followed by lexicographers.

The splitting of Ums is a manner to differentiate certain meanings, such as masculine manche (handle) and feminine manche (sleeve), when the two meanings present differences in their variations of form or in grammatical properties. It is not possible to represent all distinctions of usage by distinguishing Ums. For example, the various uses of the feminine manche (La chemise a une manche trop longue. The shirt has sleeves that are too long- La partie a eu trois manches. The game was played in three rounds.) are grouped under a single Um.

Two distinct paradigms define two distinct Ums. For example, the verb ressortir has two different conjugations: (Luc ressort de la cuisine. Luc is coming out from the kitchen (again)- Cette affaire ressortit à la compétence de ce tribunal. This matter is under the jurisdiction of the court.) This verb is therefore represented under two Ums. Two paradigms may be differentiated uniquely by the existence of certain CombTMs that are present in a paradigm but absent in another. To illustrate, the human noun comique (Luc est un comique - Luc is a clown), variable in gender and in number, does not have the same paradigm as the invariable noun comique (La situation est d'un comique irrésistible - The situation is irresistibly funny.), which has only one CombTM -- the masculine singular. Nevertheless, when distinct uses have minor differences in terms of the existence of certain Inflected Forms, they are grouped into a single Um. For example, certain uses of the verb sentir have a variable past participle (le chien sent les sacs- the dog smells the bag), and others an invariable past participle (Les sacs sentent la sardine - the bags smell like sardines). This minor difference, however, does not justify splitting the word into two Ums and finds its place in syntax. Assessment of the situation is left to the lexicographer.

A Um corresponding to a simple word or to an element of a word is generally related to one and only one Umg, which serves to describe the spelling of the word and of its Inflected Forms. This information has been placed into a distinct entity of the Um because of the existence of spelling differences. In the case of free spelling variations with or without change in pronunciation, as in clé which is written as clef as well, a specific Umg is reserved for each of the variants and a unique Um serves to describe the information independent of the spelling, thus permitting us to express the equivalence of the different spellings. If the word admits an inflection, we distinguish several Umgs only if the canonical form (the infinitive for verbs, the masculine singular for adjectives, etc) has several spellings, such as clé and clef, or several genders, such as interface. When there are free spelling variations in certain Inflected Forms but not in the canonical form, we represent the word under one Um, with the differences registered in the inflection. Asseoir, for example, has only one infinitive but two Inflected Forms for "he sits" - (assoit, assied). (refer to Section 6.1 below).

The spelling of compounds is processed with the help of a relation of composition (see 2.4 Relations between Ums, in this section.). A Um which corresponds to a compound is therefore not directly linked to any Umg.

A Um corresponding to a simple word or to an element of a word is generally linked to one and only one phonemic Morphological Unit (Ump) , which serves to describe the pronunciation of the word and its Inflected Forms. This information has been placed into a distinct entity in the Um because of the existence of variation in pronunciation. In the case of free variations in pronunciation with or without change in spelling, as in razzia, pronounced either [razja] or [radzja], a specific Ump is reserved for each of the variants and a unique Um serves to describe the information independent of the pronunciation, thus permitting us to express the equivalence of the different pronunciations. If a word admits an inflection, we distinguish several Umps only when the canonical form (the infinitive for a verb, the masculine singular for an adjective, etc) has several pronunciations, as razzia, or several genders, as interface. When there are free phonemic variations in certain Inflected Forms but not in the canonical form, we represent the word under one Ump and the variations are registered in the inflection. The noun lobby, for example, has two pronunciations in the plural -- lobbies : [lobi, lobiz].

Regular phonetic variations are not taken into consideration at the level of Ums and Umps : a silent e which could either be pronounced or not, as in retirer, or a u which could either be syllabic or non-syllabic, as in tuer, does not justify the coding of several Umps. These variations generate, however, several phonetic transcriptions for the same word: in the International Phonetic Alphabet, [tye] and [ty+e]. Given their regularity, they correspond to a single phonemic transcription, for example /ty+e/ for tuer, whereas the variants of razzia, which do not stem from any regular variation in French, justify two distinct phonemic transcriptions -- /razia/ and /radzia/. The lexicographer may also classify the variations, following their regularity, at either a phonemic or a phonetic level. Only phonemic variations are taken into consideration in the Ump. As for regular variations, categories at the phonetic level and all lexical information that concerns them are encoded in the phonemic representations, and an algorithm is associated to the phonemic system to generate the phonetic variants. The pronunciations of compounds is processed with the help of a relation of composition (see 2.4 below.) A Um which corresponds to a compound is therefore not linked directly to any Ump.

Ums do not represent simple words solely, but also elements of a word: prefixes, etc, and compounds, as in sur le vif. Indeed, there are relations between Ums.:

either derivational, that is, interior to a simple word, as for example condamner and condamnation.

or compositional, that is, exterior to simple words, such as vif and sur le vif.

These relations permit us to encode the decomposition of a Um into a series of Ums. They indicate the order of the element in the series, for example 3 for the relation between vif and sur le vif and 1 for the relation between électro- and électrochoc. They also indicate the presence and the value of the characters of separation.

written separators : spaces in sur le vif, hyphens in sur-le-champ, an apostrophe in aujourd'hui, and an absence of separators in électrochoc.

phonemic separators: introduction of an element of liaison as in mot à mot.

The combinations of Morphological Features serve to characterize all Inflected Forms of a Um and to indicate the Morphological Features of these forms. Even when a Um has only one Inflected Form, as fiançailles, the attributes of the unique CombTM indicate the gender and number. For words of a fixed gender, such as lit, inspection of the CombTMs permits us to deduce its gender and the fact that the gender is fixed. There are four CombTMs for nouns and adjectives, 51 for verbs (only simple tenses are represented: the formation of complex tenses is represented in the syntactic layer), and several others for determiners and pronouns. There is also a special "empty" CombTM for those invariable words for which Morphological Features are not relevant (prepositions, adverbs, ...) - all the attributes of this CombTM have a value sans (without).

A Umg gives the spelling of a simple word or of an element of a word. In the case of free spelling variation, we have one Umg per variant. Each Umg may therefore carry information on the style (level of language) to which the variant belongs, its frequency of use, the date of its first attestation in the language, or geographical location of its use.

Umgs all possess a system of inflection. The Umgs of words which carry Morphological Features (nouns, adjective, verbs, and certain grammatical words) are linked to an Mfg which specifies the existence, the number and spelling of the Inflected Forms, even if there is only one such form. The Umgs of words without Morphological Features are in relation with an empty Mfg.

The Mfg concerns words which carry Morphological Features (nouns, adjectives, verbs and certain grammatical words). It specifies the existence, number, and spelling of the Inflected Forms, even if there is only one such form. Two simple words have the same Mfg if and only if they have the same number of Inflected Forms with the same combinations of Morphological Features, and if the spellings of these forms are deducible in the same manner for the two words. It is, for example, the case for the verbs parvenir and prévenir. Only simple tenses are registered in the attributes of Mfgs. The formation of complex tenses (être parvenu, avoir prévenu) and of the pronominal form (se prévenir) are processed in the syntactic layer.

Two Mfgs may differ uniquely by the existence of certain CombTMs, present in one but absent in the other. For example, the human noun comique (Luc est un comique), variable in gender and number, does not have the same Mfg as the invariable comique (La situation est d'un comique irrésistible), which only has a single CombTM -- masculine, singular. However, when inflectional paradigms have only minor differences in certain Inflected Forms, only one Mfg is attributed. For example, in accordance with tradition, the conjugation of the verb b‰tir, with a variable past participle, is not distinguished from that of agir, whose past participle is invariable. The Mfg for b‰tir is attributed to agir and the syntactic description indicates if the past participle is variable or not.

The "empty" Mfg is a special one, having only one "empty" CombTM and only one Cff (elementary calculation rule for an inflected form) with a null Removal and a null Addition.

A Cff is an elementary rule of computation for a graphic Inflected Form. Associated with an Mfg, the Cff is used to calculate the graphic Inflected Form. There are two types of rules:

i. Addition of a termination to a radical

ii. computation of a radical by Removal of a chain of characters from the radical and Addition of a suffix

In the most irregular conjugations, the chain to be removed and to be added may extend to the entire form, for example étaient as computed from être.

With an Mfg and a CombTM may be associated a Cff which permits us to find the Inflected Form of this CombTM for those words which inflect according to the Mfg. For example, if we consider the conjugation of laisser, represented by an Mfg, and the CombTM of the present participle, these two entities are linked to a Cff which specifies in substance: remove -er, add -ant. If free variants exist for certain Inflected Forms, several Cffs are associated with the same Mfg and the same CombTM. For example, the Mfg of asseoir and the CombTM of the present, third person singular are linked to two Cffs:

remove -eoir, add -ied,

remove -eoir, add -oit.

A Ump gives the phonemic transcription of a simple word or of an element of a word. If there are free phonemic variations, we have one Ump per variant. Each Ump may therefore carry information on the style (level of language) of the variant, its frequency in usage and the date of its first attestation in the language. Remember that the lexicographer may classify the variations in pronunciation, following their regularity, at either a phonemic or phonetic level. Only phonemic variations are taken into account in the Umps. For the phonetic level, refer to the section above.

In the case of free variants; the correspondence between the different Umgs and Umps of the same word is not obvious. It is always explicitly indicated by a direct relationship between these entities.

The Mfp concerns words which carry Morphological Features (nouns, adjectives, verbs and certain grammatical words). It specifies the existence, number, and a phonemic transcription of the Inflected Forms, even if there is only one such form. Two simple words have the same Mfp if and only if they have the same number of Inflected Forms with the same combinations of Morphological Features, and if the transcriptions of these forms are deducible in the same manner for the two words. It is for example the case for the verbs croire and pourvoir. As it is for Mfgs, only the simple tenses enter into consideration for the attributes of Mfps: the formation of complex tenses and of pronominal forms is processed in the syntactic layer. The definition of Mfps is completely parallel to that of Mfgs: two Mfps may differ uniquely by the existence of certain CombTM, present in one but absent in another. However, when morphological paradigms have only minor differences in the existence of certain Inflected Forms, the same Mfp is attributed to them (see Section 5 on Mfgs above). We see that the Mfgs are in complex relation with the Mfps, because two words may have the same Mfg and yet have distinct Mfps (amer and léger) or vice versa (croire and pourvoir).

A Cff (see Section 6 above.) is an elementary rule of derivation of a phonemic Inflected Form. There are two types of rules:

i. addition of a termination to a radical

ii. computation of a radical by Removal of a chain of characters from the radical and Addition of a suffix

In the most irregular conjugations, the chain to be removed and that to be added may extend to the entire form, for example sera from être. This computation is carried out by phonemic and not phonetic transcriptions. The regular phonetic variations that appear during inflection does not complicate the computation. For example, in the conjugation of céder, the phonetic alternation between [é] and [è] (in céder [sede] and cède [sed]) does not affect the phonemic transcriptions: céder /sede/, cède /sed/.

With an Mfp and a CombTM may be associated a Cff which permits us to find the Inflected Form of the CombTM for those words which are inflected according to the Mfp. For example, if we consider the phonemic inflection of courir, represented by an Mfp, and the CombTM of the third person singular of the future, these two entities are linked to a Cff which specifies in substance: remove /ir/, add /ra/. If there are free variants in certain Inflected Forms, several Cffs may be associated with the same Mfp and the same CombTM. For example, le Mfp of asseoir and the CombTM of the present third person singular are related to two Cffs:

remove /uar/, add /ie/.

remove /r/, add the empty termination.

The relation is written/ is pronounced expresses the link between a Cff used to calculate a graphic form and a Cff used to calculate a phonemic form. In the majority of cases, this relation is deducible from the CombTM, as a function of the Umg, the Ump, the Mfg and the Mfp.

This relation is, however, necessary when there are variants in the graphic and phonemic systems. We must be able to link each graphic variant to the correct phonemic variant and vice versa.

Ex: imppre20p [asseyez] /aseje/ [assoyez] /asuaye/

An Mfc enters into the relation of composition (R_Compose) between a compound and a component. The Mfc permits us to select for each CombTM of the compound the Inflected Form(s) of the pertinent components.

Um_C points to R_Compose which:

points to a component .

gives the place of this component in the compound (linear order)

indicates the possible separators before the component (sepg/p)

points to the Mfc of the compound, given the component.

The Mfc points to couplets of CombTM (Comb_Comb) which allows us to associate with one CombTM of the component the CombTM(s) of the pertinent compounds.

Each of these couplets:

points to one CombTM of the compound

points to one to N CombTMs of the component (for the selection of the pertinent Inflected Form(s)).

specifies, as needed, the contextual variant (contexte_var) to be selected among the Inflected Forms of the compound.

The conceptual model of GENELEX has been expressed to a great extent in terms of an entity-attribute-relation model (Merise).

Many constraints of integrity are expressed in this EAR model: type of objects, type of relations, cardinality of relations, etc. However, as the model was not conceived to express rules - doing so gives rise to extreme complications - certain constraints had to be expressed in the accompanying document (restriction on the combination of values). It follows that the morphological model of GENELEX is a combination of EAR formalism and comments in natural language (NL).

An SGML DTD (Definition of Type of Document) is a physical model of grammar which describes the marking of data.

During the transition from the conceptual model (EAR + NL)to the GENELEX DTD, the EAR models had to be translated in a very systematic way. We have attempted to express formally all of the integrity constraints, expressed in natural language, in the DTD.

Certain rules for translating the EAR formalism to SGML have been applied:

A constraint file was developed to faciliate cross-referencing. The constraints that are expressed therein appear as comments and are, as such, ignored by the SGML parser; they express, in an intuitive syntax, the reference types.