mmorph - MULTEXT morphology tool formalism syntax

A mmorph morphology description file is divided into declaration sections. Each section starts by a section header (`@ Alphabets', `@ Attributes', etc.) followed by a sequence of declarations. Each declaration starts by a name, followed by a colon (`:') and the definition associated to the name. Here is a brief description of each section:

In this section the lexical and surface alphabet are declared. All symbols forming each alphabet has to be listed. Symbols may appear in both the lexical and surface alphabet definition in which case it is considered a bi-level symbol, otherwise it is a lexical only or surface only symbol. Symbols are usually letters (eg. a, b, c) , but may also consist of longer names (beta, schwa). Symbol names consisting of one special character (`:' or `(') may be specified by enclosing them in double quotes (`:' or `(').

Example: Lexical : a b c d e f g h i j k l m n o p q r s t u v w x y z "-" "." "," "?" "!" "\"" "'" ":" ";" "(" ")" strong_e Surface : a b c d e f g h i j k l m n o p q r s t u v w x y z "-" "." "," "?" "!" "\"" "'" ":" ";" "(" ")" " "

In this example, the symbol strong_e is lexical only, the symbol " " (space) is surface only. All the other symbols are bi-level.

All the strings appearing in the rest of the grammar will be made exclusively of symbols declared in this section.

In this section, the name of attributes (sometimes called features) and their associated value set. At most 32 different values may be declared for an attribute.

Examples:

Gender : feminine masculine neuter Number : singular plural Person : 1st 2nd 3rd Transitive : yes no Inflection : base intermediate final

In the current version of the implementation value sets of different attributes are incompatible, even if they are defined identically. To overcome this restriction, in a future version this section will be split into two: declaration of value sets and declaration of attributes.

In this section, the different types of feature structures are declared. The attributes allowed for each type are listed. Attributes that are only used within the scope of the tool and have no meaning outside can be listed after a bar (`|'). The values of these local attributes ar not stored in the database or written on the final output of the program.

Examples:

Noun : Gender Number Verb : Tense Person Gender Number Transitive | Inflection

Typed feature structures are used in the grammar and spelling rules. It is the specification of a type and the value of some associated attributes. The list of attribute specifications is enclosed in square brackets (`[' and `]').

Example:

Noun[ Gender=feminine Number=singular ]

It is possible to specify a set of values for an attribute by listing the possible valuse separated with a bar (`|'), or the complement of a set (with respect to all possible values of that attribute) indicated with `!=' instead of `='.

Example: Assuming the declaration of Gender as above, the following two typed feature structures are equivalent

Noun[ Gender=masculine|neuter ] Noun[ Gender!=feminine ]

This section contains the rules that specify the structure of words. It has the general shape of a context free grammar over typed feature structures. There are three basic types of rules: binary, goal and affixes.

Binary rules specify the result of the concatenation of two elements. This is written as:

Rule_name : Lhs <- Rhs1 Rhs2

where Lhs is called the left hand side, and Rhs1 and Rhs2 the first and second part of the right hand side. Lhs, Rhs1 and Rhs2 are specified as typed feature structures.

Example:

Rule_1 : Noun[ Gender=feminine Number=singular ] <- Noun[ Gender=feminine Number=singular ] NounSuffix[ Gender=feminine ]

Variables can be used to indicate that some attributes have the same value. A variable is a name starting with a dollar (`$').

Example:

Rule_2 : Noun[ Gender=$A Number=$number ] <- Noun[ Gender=$A Number=$number ] NounSuffix[ Gender=$A ]

If needed, both a variable and a value specification can be given for an attribute (only once per attribute):

Example:

Rule_3 : Noun[ Gender=$A Number=$number ] <- Noun[ Gender=$A Number=$number ] NounSuffix[ Gender=$A=masculine|neuter ]

Affix rules define basic elements of the concatenations specified by binary rules (together with lexical entries, see the section @ Lexicon below). An affix rule consists of lexical string associated to a typed feature structure.

Examples:

Plural_s : "s" NounSuffix[ Number=plural ] Feminine_e : "e" NounSuffix[ Gender=feminine ] ing : "ing" VerbSuffix[ Tense=present_participle ]

Goal rules specify the valid results constructed by the grammar. They consist of just a typed feature structure.

Examples:

Goal_1 : Noun[] Goal_2 : Verb[ inflection=final ]

In addition to these three basic rule types, there are prefix or suffix composite rules and unary rules. A unary rule consist of a left hand side and a right hand side.

Example:

Rule_4 : Noun[ gender=$G number=plural ] <- Noun[ gender=$G number=singular invariant=yes]

Prefix and suffix composite rules have the same shape as binary rules except that one part of the right hand side is an affix (i.e. has an associated string).

Examples:

Append_e : Noun[ Gender=feminine Number=$number ] <- Noun[ Gender=feminine Number=$number ] "e" NounSuffix[ Gender=feminine ]

anti : Noun[ Gender=$gender Number=$number ] <- "anti" NounPrefix[] Noun[ Gender=$gender Number=$number ]

This optional section contains the definition of symbol classes. Each class is defined as a set of symbols, or other classes. If the class contains only bi-level elements it is a bi-level class, otherwise it is a lexical or surface class.

Examples:

Dental : d t Vowel : a e i o u Vowel_y : Vowel y Consonant: b c d f g h j k l m n p q r s t v w x z

This optional section contains the definition of pair disjunctions. Each disjunction is defined as a set of pairs. Explicit pairs specify a sequence of surface symbols and a sequence of zero or one lexical symbol, one of them possibly empty. A sequence is enclosed between angle brackets `<' and `>'. The empty sequence is indicated with `<>'. In the current implementation only the surface part of a pair can be a sequence of more than one element. The special symbol `?' stands for the class of all possible symbols, including the morpheme and word boundary.

Examples:

s_x_z_1 : s/s x/x z/z VowelPair1: a/a e/e i/i o/o u/u VowelPair2: Vowel/Vowel ie.y: <i e>/y Delete_e: <>/e Insert_d: d/<> Surface_Vowel: Vowel/? Lexical_s: ?/s DoubleConsonant: <b b>/b <d d>/d <f f>/f <g g>/g <k k>/k <m m>/m <p p>/p <s s>/s <t t>/t <v v>/v <z z>/z

Note that VowelPair1 and VowelPair2 don't specify the same thing: VowelPair2 would match a/o but VowelPair1 would not.

Implicit pairs are specified by the name of a bi-level symbol or a bi-level class.

Examples: the following s_x_z_2 and VowelPair3 are equivalent to the above s_x_z_1 and VowelPair2 (assuming that s, x, z and Vowel are bi-level symbols and classes).

s_x_z_2 : s x z VowelPair3 : Vowel

In a pair disjunction all lexical parts should be disjoint. This means you cannot specify for the same pair disjunction a/a and o/a or a/a and Vowel/Vowel.

In a future version this section will be split in two: simple pair disjunctions and pair sequences.

In this section are declared the two level spelling rules. A spelling rule consist of a kind indicator followed by a left context a focus and a right context. The kind indicator is `=>' if the rule is optional, `<=>' if it is obligatory and `<=' if it is a surface coercion rule. The contexts may be empty. The focus is surrounded by two `-'. The contexts and the focus consist of a sequence of pairs or pair disjunctions declared in the `@ Pairs section. A morpheme boundary is indicated by a `+' or a `*', a word boundary is indicated by a `~'.

Examples:

Sibilant_s: <=> s_x_z_1 * - e/<> - s Gemination: <=> Consonant Vowel - DoubleConsonant - * Vowel i_y_optionnel: => a - i/y - * ?/e

Constraints may be specified in the form of a list of typed feature structures. They are affix-driven: the rule is licensed if at least one of them subsumes the closest corresponding affix. The morpheme boundary indicated by a star (`*') will be used to determine which affix it is. If there is no such indication, then the affix adjacent to the morpheme where the first character of the focus occurs is used. In case there is no affix, the typed feature structure of the lexical stem is used.

Example:

Sibilant_s: <=> s_x_z_1 * - e/<> - s NounSuffix[ Number=plural ]

This section is optional and can also be repeated. This section lists all the lexical entries of the morphological description. Unlike the other sections, definitions do not have a name. A definition consist of a typed feature strucure followed by a list of lexical stems that share that feature structure. A lexical stem consists of the string used in the concatenation specified by the grammar rules followed by `=' and a reference string. The reference string can be anything and usually is used to indicate the canonical form of the word or an identifier of an external database entry.

Examples:

Noun[ Number=singular ] "table" = "table" "chair" = "chair"
Verb[ Transitive=yes|no Inflection=base ] "bow" = "bow1"
Noun[ Number=singular ] "bow" = "bow2"


If the stem string and the reference strings are identical, only one needs
to be specified.



Example:

Noun[ Number=singular ] "table" "chair"

The formal syntax description below is in Backus Naur Form (BNF). The following conventions apply:

<id> is a non-terminal symbol (within angle brackets). ID is a token (terminal symbol, all uppercase). <id>? means zero or one occurrence of <id> (i.e. <id> is optional). <id>* is zero or more occurrences of <id>. <id>+ is one or more occurrences of <id>. ::= separates a non-terminal symbol and its expansion. | indicates an alternative expansion. ; starts a comment (not part of the definition).

The start symbol corresponding to a complete description is named <Start>. Symbols that parse but do nothing are marked with `; not operational'.

<Start> ::= <AlphabetDecl> <AttDecl> <TypeDecl> <GramDecl> <ClassDecl>? <PairDecl>? <SpellDecl>? <LexDecl>*

<AlphabetDecl> ::= ALPHABETS <LexicalDef> <SurfaceDef>

<LexicalDef> ::= <LexicalName> COLON <LexicalSymbol>+ <SurfaceDef> ::= <SurfaceName> COLON <SurfaceSymbol>+

<LexicalSymbol> ::= <LexicalSymbolName> ; lexical only | <BiLevelSymbolName> ; both lexical and surface

<SurfaceSymbol> ::= <SurfaceSymbolName> ; surface only | <BiLevelSymbolName> ; both lexical and surface

<AttDecl> ::= ATTRIBUTES <AttDef>+

<AttDef> ::= <AttName> COLON <ValName>+

<TypeDecl> ::= TYPES <TypeDef>+

<TypeDef> ::= <TypeName> COLON <AttName>+ <NoProjAtt>?

<NoProjAtt> ::= BAR <AttName>+

<LexDecl> ::= LEXICON <LexDef>+

<LexDef> ::= <Tfs> <Lexical>+

<Lexical> ::= LEXICALSTRING <BaseForm>?

<BaseForm> ::= EQUAL LEXICALSTRING

<Tfs> ::= <TypeName> <AttSpec>?

<VarTfs> ::= <TypeName> <VarAttSpec>?

<AttSpec> ::= LBRA <AttVal>* RBRA

<VarAttSpec> ::= LBRA <VarAttVal>* RBRA

<AttVal> ::= <AttName> <ValSpec>

<VarAttVal> ::= <AttName> <VarValSpec>

<ValSpec> ::= EQUAL <ValSet> | NOTEQUAL <ValSet>

<VarValSpec> ::= <ValSpec> | EQUAL DOLLAR <VarName> | EQUAL DOLLAR <VarName> <ValSpec>

<ValSet> ::= <ValName> <ValSetRest>*

<ValSetRest> ::= BAR <ValName>

<GramDecl> ::= GRAMMAR <Rule>+

<RuleDef> ::= <RuleName> COLON <RuleBody>

<RuleBody> ::= <VarTfs> LARROW <Rhs> | <Tfs> ; goal rule | LEXICALSTRING <Tfs> ; lexical affix

<Rhs> ::= <VarTfs> ; unary rule | <VarTfs> <VarTfs> ; binary rule | LEXICALSTRING <Tfs> <VarTfs> ; prefix rule | <VarTfs> <Tfs> LEXICALSTRING ; suffix rule

<ClassDecl> ::= CLASSES <ClassDef>+

<ClassDef> ::= <LexicalClassName> COLON <LexicalClass>+ | <SurfaceClassName> COLON <SurfaceClass>+ | <BiLevelClassName> COLON <BiLevelClass>+

<LexicalClass> ::= <LexicalSymbol> | <LexicalClassName> | <BiLevelClassName>

<SurfaceClass> ::= <SurfaceSymbol> | <SurfaceClassName> | <BiLevelClassName>

<BiLevelClass> ::= <BiLevelSymbolName> | <BiLevelClassName>

<PairDecl> ::= PAIRS <PairDef>+

<PairDef> ::= <PairName> COLON <PairDef>+

<PairDef> ::= <PairName> COLON <Pair>+

<Pair> ::= <SurfaceSequence> SLASH <LexicalSequence> | <PairName> | <BiLevelClassName> | <BiLevelSymbolName>

SurfaceSequence ::= LANGLE <SurfaceSymbol>* RANGLE | SURFACESTRING | <SurfaceClass> | ANY

LexicalSequence ::= LANGLE <LexicalSymbol>* RANGLE | LEXICALSTRING | <LexicalClass> | ANY

<SpellDecl> ::= SPELLING <SpellDef>+

<SpellDef> ::= <SpellName> COLON <Arrow> <LeftContext> <Focus> <RightContext> <Constraint>*

<LeftContext> ::= <Pattern>*

<RightContext> ::= <Pattern>*

<Focus> ::= CONTEXTBOUNDARY <Pattern>+ CONTEXTBOUNDARY

<Pattern> ::= <Pair> | MORPHEMEBOUNDARY | WORDBOUNDARY | CONCATBOUNDARY

<Constraint> ::= <Tfs>

<Arrow> ::= RARROW | BIARROW | COERCEARROW

<AttName> ::= NAME <BiLevelClassName> ::= NAME <BiLevelSymbolName> ::= NAME | SYMBOLSTRING <LexicalClassName> ::= NAME <LexicalName> ::= NAME <LexicalSymbolName> ::= NAME | SYMBOLSTRING <PairName> ::= NAME <RuleName> ::= NAME <SpellName> ::= NAME <SurfaceClassName> ::= NAME <SurfaceName> ::= NAME <SurfaceSymbolName> ::= NAME | SYMBOLSTRING <TypeName> ::= NAME <ValName> ::= NAME <VarName> ::= NAME

Simple tokens

Complex tokens

mmorph(1). G. Russell and D. Petitpierre, MMORPH - The Multext Morphology Program, Version 2.3, October 1995, MULTEXT deliverable report for task 2.3.1.

Dominique Petitpierre, ISSCO, <petitp@divsun.unige.ch>

The parser for the morphology description formalims above was written using yacc (1) and flex (1). Flex was written by Vern Paxson, <vern@ee.lbl.gov>, and is distributed in the framework of the GNU project under the condition of the GNU General Public License