Documentation for the Grammar Matrix Customization Lexicon Library

Introduction

This document explains how to fill out the Lexicon page of the Grammar Matrix Customization questionnaire and presents background information on the person library of the Grammar Matrix Customization System (Bender et al., 2002; Bender and Flickinger, 2005; Bender et al., 2010). General instructions on using the questionnaire can be found here.

The Lexicon page interacts with virtually every library in the customization system. On the Lexicon page, the user can define various lexical types. The features and values available to use in the definition of these lexical types depend on the answers to other parts of the questionnaire.

Citing the Lexicon Library

Most of the work for the Lexicon Library and its implementations has been done as part of Drellishak 2009. The full reference and .bib entry can be found here. The adjectives and copulas were added as part of Trimble 2014 (forthcoming).

Table of Contents

Options

General

The Lexicon page provides means to define grammatical properties of nouns, transitive and intransitive verbs, adjectives, auxiliaries and copulas, determiners and case-marking adpositions. The page bears a red question mark when unfilled. This means that it is possible to create a grammar with an empty lexicon, but generally assumed that it makes sense to define at least a noun and two verb types (typically intransitive and transitive) so that the grammar will be able to parse something. The question mark disappears as soon as at least one noun type and two verb types are given.

Nouns, verbs, adjectives, and copula types that partially share grammatical properties can be defined as a hierarchy, where a general supertype defines properties shared by more than one class of nouns or verbs. The noun or verb classes that exhibit these shared properties will inherit them from the supertype (see the specific documentation on noun and verb classes for details).

Nouns

The following information can be provided for each noun type:

A more detailed description of each of these properties follows below.

Type name

Defines the name of the class of nouns. It should be noted that:

Supertypes

This dropbox allows you to select supertypes of the noun-class. The dropbox will contain all other noun types you have defined, the order is not relevant. You can select several supertypes. This is useful if you want to cross-classify properties such as grammatical gender and countability for classes of nouns.

A noun inherits the feature values of its supertype. For instance:

In the resulting grammar, all nouns belonging to the class of noun_type2 will be 3rd person.

Next to Noun Types, you find a(n experimental) link saying visualize noun hierarchy. It will display the supertype relations that are defined at that moment as a hierarchy.

Cycles of inheritance are not allowed (and they do not make sense anyway). So if noun_type2 is a supertype of noun_type1, noun_type1 may not be a supertype of noun_type2. Note that the supertype-subtype relation is transitive, so if a := b indicates a supertype relation where b is supertype of a, the following is forbidden as well:

a := b.
b := c.
c := a.

In the supertype hierarchy above, c is a supertype of a through b. So a cannot be a supertype of c.

CAREFUL!!!!!!

Validation does not check whether constraints on supertypes and subtypes are compatible. This means that grammars with such errors can be created by the customization system, but they will not load in lkb and cannot be compiled to work with PET or ACE. An example of this would be the following:

NounType 1
Type name: 2nd-sg
supertype: 3rd-sg
Feature: PERSON 2nd

NounType 2
Type name: 3rd-sg

Feature: PERSON 3rd

The 2nd-sg noun will inherit the property [PERSON 3rd] from the 3rd-sg noun, but have the additional property of [PERSON 2nd] from its own definition. Assuming that 3rd and 2nd are conflicting values, the grammar contains an ill-defined type. The following hierarchy would be possible:

NounType 1
Type name: 2nd-sg
supertype: 3rd-sg
Feature: PERSON 2nd

NounType 2
Type name: 3rd-sg

Feature: PERSON 3rd, 2nd

In this case, Noun Type 2 will have a value for PERSON that is an underspecification of 2nd and 3rd. This value can unify with the value of Noun Type 1.

There is a bug report for this problem, but it is extremely hard to implement so that it works at the time of validation. So for now and the near future: pay attention while defining inheritance!

Features

The Feature option for verbs has three fields name, value. You can select a feature under name and its value(s) under value.

The drop-boxes will show features and values based on what was filled out at the pages Person, Number, Gender, Case, and Other Features. The first four libraries (Person, Number, Gender, Case) each define features typically appropriate for nouns. Other Features allows you to define customized features. These features can be assigned to nouns, verbs or both. Features that are defined for both categories or for nouns will appear in the drop-box.

Select the feature you want to specify from the left drop-box. Appropriate values will appear in the right drop-box. Each value is preceded by a check box. You can select as many values as you want for each feature (but note that selecting all values leads to the same behavior as not specifying a feature value pair at all).

If you select more than one value for a feature, the noun class will assign an underspecified value to the feature that may unify with any of the selected values. The customization system will adapt the type hierarchy of the feature value accordingly. See the analysis section for an example of the impact on the type hierarchy.

Determiners

This is the only obligatory question for noun classes, validation indicates an error when this question is not answered, while other questions for a given noun class are.

You must indicate whether determiners are optional, obligatory or impossible. Nouns with optional determiners can either form a noun phrase by themselves or by being combined with a determiner. If determiners are obligatory, they can only form a noun phrase (and hence only occur in a parsed sentence) when combined with a determiner. If determiners are impossible, the nouns form NPs by themselves and cannot be combined with determiners.

If you indicated on the Word Order page that the grammar does not have determiners, you must select impossible (else validation will indicate there is an error).

Stems

Here you can define the actual stems of nouns belonging to a given class. Note that classes that are meant to serve as a supertype for other noun classes should (in most cases) not have stems. You can add as many noun stems per class as you want. It is not the fastest way to define a large scale lexicon, so you probably want to limit it to a hand-full that allows you to build decent test suites.

For each noun stem, spelling and a predicate must be provided. After you fill out the spelling and hit tab, a predicate name based on the stem and MRS standards is suggested. This has the following form:

Spelling: myStem     Predicate: _myStem_n_rel

You can change the predicate if you like, but it is recommended to stick to the conventions of starting with _ and ending nominal relations by _n_rel.

Pronouns are typical examples where predicates are not based on the spelling of the stem (e.g. _pronoun_n_rel, _pro_n_rel for all pronouns, INDEX features indicate the person, number, gender and other distinctive properties it may have).

Morphotactic constraints

This part of the lexicon library allows you to define interactions with morphotactics, which morphemes are required and which are forbidden to combine with a given lexical type.

When you click either of the buttons, a constraint description appears with a drop-down menu. The position classes and morphological rules defined for nouns on the morphology page appear in the drop-down. Select a class to define complete the constraint. You can define as many constraints as necessary.

A Require constraint makes sure the lexical item is only accepted as a word after combining with a morpheme belonging to the selected position class or an instance of the selected lexical rule (depending on whether a position class or lexical rule was selected).

A Forbid constraint makes sure the lexical item cannot combine with a morpheme belonging to the selected position class or an instance of the selected lexical rule (depending on whether a position class or lexical rule was selected).

Verbs

The following information can be provided for each noun type:

A more detailed description of each of these properties follows below.

Type name

Defines the name of the class of verbs (similar to nouns). It should be noted that:

Supertypes

This dropbox allows you to select supertypes of the verb-class. The dropbox will contain all other verb types you have defined, the order is not relevant. You can select several supertypes. This is useful if you want to cross-classify properties such as argument structure, case marking on arguments and subject or object agreement (if not handled by morphotactics).

A verb inherits the feature values of its supertype. For instance:

In the resulting grammar, all verbs belonging to the class of verb_type5 will be intransitive verbs with nominative subjects, all verbs belonging to the class of verb_type6 will be transitive verbs with nominative subjects and accusative objects.

Next to Verb Types, you find a(n experimental) link saying visualize verb hierarchy. It will display the supertype relations that are defined at that moment as a hierarchy.

Cycles of inheritance are not allowed (and they do not make sense anyway). So if verb_type2 is a supertype of verb_type1, verb_type1 may not be a supertype of verb_type2. Note that the supertype-subtype relation is transitive, so if a := b indicates a supertype relation where b is supertype of a, the following is forbidden as well:

a := b.
b := c.
c := a.

In the supertype hierarchy above, c is a supertype of a through b. So a cannot be a supertype of c

CAREFUL!!!!!!

Validation does not check whether constraints on supertypes and subtypes are compatible. This means that grammars with such errors can be created by the customization system, but they will not load in lkb and cannot be compiled to work with PET or ACE. An example of this would be the following:

VerbType 1
Type name: past
supertype: present-v
Feature: TENSE past

NounType 2
Type name: present-v

Feature: TENSE present

The past verb will inherit the property [TENSE present] from the future verb, but have the additional property of [TENSE past] from its own definition. Assuming that past and future are conflicting values, the grammar contains an ill-defined type. The following hierarchy would be possible:

VerbType 1
Type name: past
supertype: past-v
Feature: TENSE past

VerbType 2
Type name: present

Feature: TENSE present, past

In this case, Verb Type 2 will have a value for TENSE that is an underspecification of present and past (typically this would be non-future defined on verbal features). This value can unify with the value of Noun Type 1.

There is a bug report for this problem, but it is extremely hard to implement so that it works at the time of validation. So for now and the near future: pay attention while defining inheritance!

Features

The Feature option for verbs has three fields: name, value, and specified on. The name of the feature can be selected under name, its value(s) under value and where the feature-value pair applies to under specified on. Further explanation follows below.

The first two drop-boxes will show features and values based on what was filled out at the pages Person, Number, Gender, Case, Tense-Mood-Aspect, and Other Features. The first four libraries (Person, Number, Gender, Case) each define features typically appropriate for nouns. Tense-Mood-Aspect library provides the means to define typical verbal features (including grammatical features such as finiteness). Other Features allows you to define customized features, which can be assigned to nouns, verbs or both. All features, regardless of whether they are defined for nouns, verbs or both categories will appear in the drop-box.

The third drop-box provides five options: the verb, the subject NP, the object NP, the higher-ranked NP and the lower-ranked NP. Typical verbal properties such as Tense, Aspect, Mood and verb form should be assigned to the verb. Agreement between verbs and their arguments can be modeled by assigning properties to an NP. Be careful to only assign properties that are defined for verbs or both categories to the verb and only properties that are defined for nouns or both to the NPs. Currently validation does not check for such errors. You will therefore be allowed to do define properties this way on the questionnaire, but your grammar will not load. A bug report has been filed to fix validation.

If your grammar has optional arguments and you defined the related properties on Argument Optionality page of the questionnaire, you can define which arguments are optional and which are not by defining feature value pair [OPT +] and [OPT -], respectively, for the arguments in question. In other words, an optional object is defined by assigning [OPT +] to the Object NP. The OPT feature cannot be defined for the verb. Validation checks for this.

Select the feature you want to specify from the left drop-box. Appropriate values will appear in the right drop-box. Each value is preceded by a check box. You can select as many values as you want for each feature (but note that selecting all values leads to the same behavior as not specifying a feature value pair at all). Assign the feature value pair to the appropriate element, keeping in mind the comments made above.

If you select more than one value for a feature, the verb class will assign an underspecified value to the feature that may unify with any of the selected values. The customization system will adapt the type hierarchy of the feature value accordingly. See the analysis section for an example of the impact on the type hierarchy.

Please not that you cannot define argument-structure here, argument-structure should be defined by the special argument-structure drop-down explained below. The option argument-structure is intended for the morphemes only (to be defined on the morphology page).

Argument Structure

The argument structure drop-box allows you to indicate whether a verb is transitive or intransitive. Depending on the definitions you provided on the Case page, the drop-box also provides the standard case marking of your language's subjects and objects (e.g. intransitive (nom) transitive (nom-acc), intransitive (erg), transitive (erg-abs)). For split case marking, no basic marking will be given for the case marking on the argument structure.

If you want to define a verb with non-standard case marking (e.g. a verb with dative objects in a nominative-accusative language, nominative-nominative case marking for copula verbs), you can select the option case unspecified. You can then specify the case marking on subject and object using 'features'.

For verbs that will undergo nominalization, choose "case unspecified" and then constrain the output of the morphological rule associated with nominalization appropriately.

For verbs that will take clausal complements, you have three options:

Argument structure is obligatory for all verb classes.

Stems

Here you can define the actual stems of verbs belonging to a given class. Note that classes that are meant to serve as a supertype for other verb classes should (in most cases) not have stems. You can add as many verb stems per class as you want. It is not the fastest way to define a large scale lexicon, so you probably want to limit it to a hand-full that allows you to build decent test suites.

For each verb stem, spelling and a predicate must be provided. After you fill out the spelling and hit tab, a predicate name based on the stem and MRS standards is suggested. This has the following form:

Spelling: myStem     Predicate: _myStem_v_rel

You can change the predicate if you like, but it is recommended to stick to the conventions of starting with _ and ending verbal relations by _v_rel.

Bipartite stems

"The term bipartite stem was coined by William Jacobsen (1980) to describe a pervasive type of verb formation in Washo, a Hokan language of western Nevada in which [...] bound morphemes neither of which can constitute a verb stem on its own, are combined to form a “bipartite” stem." (DeLancey, p.1)

Above 'Stems' you will find the line If this verb class includes bipartite stems, select the position class for the affix portion of the stems: followed by a dropbox. Here, you can select a position class defined on the Morphology page. Position classes define whether an affix is a prefix or suffix and the morphemes (stems or other affixes) it is preceded or followed by.

If the verbs in the language have positions specific to the affix of the bipartite stem, you should go to the Morphology page and define this position. You can then select it for the verb class on the Lexicon page.

If you click on Add a bipartite Stem three boxes appear, one for the root, one for the affix and one for the predicate. As for regular stems, a predicate is suggested as soon as you enter the root:

Root spelling: myRoot       Affix spelling: myAffix        Predicate: _myroot_v_rel

Morphotactic Constraints

This part of the lexicon library allows you to define interactions with morphotactics, which morphemes are required and which are forbidden to combine with a given lexical type.

When you click either of the buttons, a constraint description appears with a drop-down menu. The position classes and morphological rules defined for verbs on the Morphology page appear in the drop-down. Select a class to define the constraint. You can define as many constraints as necessary.

A Require constraint makes sure the lexical item is only accepted as a word after combining with a morpheme belonging to the selected position class or an instance of the selected lexical rule (depending on whether a position class or lexical rule was selected).

A Forbid constraint makes sure the lexical item cannot combine with a morpheme belonging to the selected position class or an instance of the selected lexical rule (depending on whether a position class or lexical rule was selected).

Adjectives

The following information can be provided for each adjective type:

A more detailed description of each of these properties follows below.

Type name

Defines the name of the class of adjectives (similar to nouns and verbs). It should be noted that:

Supertypes

This dropbox allows you to select supertypes of the adjective-class. The dropbox will contain all other adjective types you have defined, the order is not relevant. You can select several supertypes. This is useful when defining agreement features, such as gender or number.

An adjective inherits the feature values of its supertype. For instance:

In the resulting grammar, there will be three leaf adjective types, one for each of the possible combinations of the two two-valued features, which can easily have lexical instances (lexicon entries) added to. Doing this helps the grammar remain structured, and to easily place additional constraints on types, say if it is discovered that plural nouns are inherently feminine.

Next to Adjective Types, you find a(n experimental) link saying visualize adjective hierarchy. It will display the supertype relations that are defined at that moment as a hierarchy.

Cycles of inheritance are not allowed (and they do not make sense anyway). So if adj_type2 is a supertype of adj_type1, adj_type1 may not be a supertype of adj_type2. Note that the supertype-subtype relation is transitive, so if a := b indicates a supertype relation where b is supertype of a, the following is forbidden as well:

a := b.
b := c.
c := a.

In the supertype hierarchy above, c is a supertype of a through b. So a cannot be a supertype of c

CAREFUL!!!!!!

Validation does not check whether constraints on supertypes and subtypes are compatible. This means that grammars with such errors can be created by the customization system, but they will not load in lkb and cannot be compiled to work with PET or ACE. An example of this would be the following:

AdjType 1
Type name: regular-both
supertype: regular-attributive

AdjType 2
Type name: regular-attributive
This adjective is attributive.

The regular-both adjective will inherit the property Behavoir:attributive from the regular-attributive adjective, but have the additional property of Behavoir:both from its own definition. This results in an ill-defined type, and should be avoided.

There is a bug report for this problem, but it is extremely hard to implement so that it works at the time of validation. So for now and the near future: pay attention while defining inheritance!

Syntactic behavoir

There are four options for syntactic behavior:

Examples of attributive adjectives include:

English: a tall man
Apatani: Abraham 1985: 23
  aki atu
  dog small
  'a small dog'
French:
  le chien noir
  DET.DEF dog black
  'the black dog'

Examples of predicative adjectives include:

Many adjectives in many languages can be both attributive and predicative, such as in English:

the big dog barks
the dog is big

The unspecified choice is for adjectives that either agree with only their subject or modificand (but not both), or are optionally the complement of a copula. The unspecified option is currently only useful for these situations, and should be avoided otherwise.

Features

The Feature option for adjectives has three fields: name, value, and specified on. The name of the feature can be selected under name, its value(s) under value and where the feature-value pair applies to under specified on. Further explanation follows below.

The first two drop-boxes will show features and values based on what was filled out at the pages Person, Number, Gender, Case, Tense-Mood-Aspect, and Other Features. The first four libraries (Person, Number, Gender, Case) each define features typically appropriate for nouns. Tense-Mood-Aspect library provides the means to define typical predicate or 'verbal' features (including grammatical features such as finiteness). Other Features allows you to define customized features, which can be assigned to nouns, verbs or both. All features, regardless of whether they are defined for nouns, verbs, or both categories will appear in the drop-box.

The third drop-box provides four options: the modified noun, the subject, both positions, and the adjective. Typical verbal properties such as Tense, Aspect, Mood and form should be assigned to the adjective. Agreement between adjectives and their modificand or subject can be modeled by assigning properties to one of the other options. Be careful to only assign properties that are defined for verbs or both categories to the adjective and only properties that are defined for nouns or both to the NPs.

Note that both positions is typologically more common, and both positions can and should be used for attributive only or predicative only adjectives. The options the modified noun and the subject are for cases where the adjective can appear in either position but only agrees with one position, or agrees differently with each argument. For instance, German adjectives only agree in attributive constructions, so they might be specified to agree with the modified noun as opposed to both positions.

Select the feature you want to specify from the left drop-box. Appropriate values will appear in the right drop-box. Each value is preceded by a check box. You can select as many values as you want for each feature (but note that selecting all values leads to the same behavior as not specifying a feature value pair at all). Assign the feature value pair to the appropriate element, keeping in mind the comments made above.

If you select more than one value for a feature, the adjective class will assign an underspecified value to the feature that may unify with any of the selected values. The customization system will adapt the type hierarchy of the feature value accordingly. See the analysis section for an example of the impact on the type hierarchy.

On the use of the subject and the modified noun, in order to enable this functionality, the system automatically does two things on the selection of one of these options:

Modification direction

This choice is only relevant for attributive or both adjectives, and specifies whether the adjective appears before or after the noun it modifies, or either. The configurations are as follows:

after the adjective: adj n
before the adjective: n adj
either position: adj n OR n adj

While some free word order languages allow adjectives to appear in any discourse constrained position, the word order library does not currently support this.

Unique modification

This choice is only relevant for attributive or both adjectives, and specifies whether the adjective can be the only modifier of the noun it modifies. That is, the string adj adj n is illegal in the target language because two adjectives modify the same noun.

Predicative behavior

This choice is only relevant for predicative or both adjectives, and specifies whether an adjective obligatorily, optionally, or impossibly appears as a copula complement. These are described below:

Stems

Here you can define the actual stems of verbs belonging to a given class. Note that classes that are meant to serve as a supertype for other verb classes should (in most cases) not have stems. You can add as many verb stems per class as you want. It is not the fastest way to define a large scale lexicon, so you probably want to limit it to a hand-full that allows you to build decent test suites.

For each verb stem, spelling and a predicate must be provided. After you fill out the spelling and hit tab, a predicate name based on the stem and MRS standards is suggested. This has the following form:

Spelling: myStem     Predicate: _myStem_a_rel

You can change the predicate if you like, but it is recommended to stick to the conventions of starting with _ and ending adjectival relations by _a_rel.

Morphotactic Constraints

NOTE: morphotactic constraints on adjectives are still under development and may not work as intended.

This part of the lexicon library allows you to define interactions with morphotactics, which morphemes are required and which are forbidden to combine with a given lexical type.

When you click either of the buttons, a constraint description appears with a drop-down menu. The position classes and morphological rules defined for verbs on the Morphology page appear in the drop-down. Select a class to define the constraint. You can define as many constraints as necessary.

A Require constraint makes sure the lexical item is only accepted as a word after combining with a morpheme belonging to the selected position class or an instance of the selected lexical rule (depending on whether a position class or lexical rule was selected).

A Forbid constraint makes sure the lexical item cannot combine with a morpheme belonging to the selected position class or an instance of the selected lexical rule (depending on whether a position class or lexical rule was selected).

Auxiliaries

Auxiliaries represent verbs that take other verbs as their complement. A basic description is provided on the Lexicon page. You can add an auxiliary by clicking on the Add an Auxiliary Type. This opens up a range of relevant fields and questions to define auxiliaries.

They are:

A more detailed description of each of these properties follows below (note: currently under development)

Type Name

Defines the name of the class of auxiliaries (similar to nouns and verbs). It should be noted that:

No predicate or An independent predicate

By selecting a radio button, you indicate whether the auxiliaries belonging to a class have an independent predicate (e.g. English can, may) or not (e.g. English will contributing future tense through a feature).

Auxiliary features

The Feature option for auxiliaries has three fields name, value and specified on. The name of the feature can be selected under name, its value(s) under value and where the feature-value pair applies to under specified on. Further explanation follows below.

The first two drop-boxes will show features and values based on what was filled out at the pages Person, Number, Gender, Case, Tense-Mood-Aspect, and Other Features. The first four libraries (Person, Number, Gender, Case) each define features typically appropriate for nouns. Tense-Mood-Aspect provides the means to define typical verbal features (including grammatical features such as finiteness). Other Features allows you to define customized features, which can be assigned to nouns, verbs (including auxiliaries) or both. All features, regardless of whether they are defined for nouns, verbs or both categories will appear in the drop-box.

The third drop-box provides five options: the auxiliary, the subject NP, the object NP, the higher-ranked NP and the lower-ranked NP. Typical verbal properties such as Tense, Aspect, Mood and verb form should be assigned to the auxiliary. Agreement between verbs and their arguments can be modeled by assigning properties to an NP. Be careful to only assign properties that are defined for auxiliaries or both categories to the auxilaries and only properties that are defined for nouns or both to the NPs. Currently validation does not check for such errors. You will therefore be allowed to do define properties this way on the questionnaire, but your grammar will not load. A bug report has been filed to fix validation.

If your grammar has optional arguments and you defined the related properties on Argument Optionality page of the questionnaire, you can define which arguments are optional and which are not by defining feature value pair [OPT +] and [OPT -], respectively, for the arguments in question. In other words, an optional object is defined by assigning [OPT +] to the Object NP. The OPT feature cannot be defined for the verb. Validation checks for this.

Select the feature you want to specify from the left drop-box. Appropriate values will appear in the right drop-box. Each value is preceded by a check box. You can select as many values as you want for each feature (but note that selecting all values leads to the same behavior as not specifying a feature value pair at all). Assign the feature value pair to the appropriate element, keeping in mind the comments made above.

If you select more than one value for a feature, the auxiliary class will assign an underspecified value to the feature that may unify with any of the selected values. The customization system will adapt the type hierarchy of the feature value accordingly. See the analysis section for an example of the impact on the type hierarchy.

Please not that you cannot define argument-structure here, the argument-structure for auxiliaries is currently defined on the Word Order page, because its specifications are mostly seen in word order. In principle, all that can be defined on argument structure for auxiliaries is which arguments of the verbal complement are raised, and which are picked up by the verbal complement itself. The option argument-structure is intended for the morphemes only (to be defined on the Morphology page).

Subject Type

Here you can define the kind of subject the auxiliary takes. There are three options concerning case marking of nouns and the option to select an adpositional phrase as a subject. Case options include no case marking, the case the verbal complement assigns to its subject, i.e. similar to a raising analysis and receiving the following case from the auxiliary, which allows you to define a specific case the auxiliary will assign to the subject.

Complement Features

Click on the button Add a complement feature to define features on the verbal complement. As mentioned on the page, a feature for FORM is required. If no feature values for form where defined, the options finite and nonfinite will be available for FORM.

The features in the drop-down menu are all suitable for verbal complements.

Stems

Here you can define the actual stems of auxiliaries belonging to a given class. You can add as many stems per auxiliary class as you want.

For each auxiliary stem must be provided. Depending on the option you filled out regarding predicates (No predicate or An independent predicate) you may also need to define a predicate (validation will throw an error if you define a predicate for a predicateless auxiliary or no predicate for an auxiliary requiring one). Since not all auxiliaries have their own predicate, no predicate form is suggested for auxiliaries. The recommended form is however similar to that of nouns and main verbs:

Spelling: myStem     Predicate: _myStem_v_rel

You can provide any predicate you like, but it is recommended to stick to the conventions of starting with _ and ending verbal relations by _v_rel.

Morphotactic Constraints

This part of the lexicon library allows you to define interactions with morphotactics, which morphemes are required and which are forbidden to combine with a given lexical type.

When you click either of the buttons, a constraint description appears with a drop-down menu. The position classes and morphological rules defined for verbs on the morphology page appear in the drop-down. Note that position classes for auxiliaries are included in position classes for verbs. Select a class to define the constraint. You can define as many constraints as necessary.

A Require constraint makes sure the lexical item is only accepted as a word after combining with a morpheme belonging to the selected position class or an instance of the selected lexical rule (depending on whether a position class or lexical rule was selected).

A Forbid constraint makes sure the lexical item cannot combine with a morpheme belonging to the selected position class or an instance of the selected lexical rule (depending on whether a position class or lexical rule was selected).

Copulas

NOTE Currently, the Grammar Matrix customization system only supports copulas with adjectival complements. However, if these sorts of copulas fit your needs for other complement types, you can use the Grammar Matrix customization system to define your copula types and lexical items, and then edit the output grammar.

The following values can be defined on Copulas.

Each of these is covered in detail below.

Type Name

Defines the name of the class of copulas (similar to nouns and verbs). It should be noted that:

Supertypes

This dropbox allows you to select supertypes of the copula-class. The dropbox will contain all other copula types you have defined, the order is not relevant. You can select several supertypes. This is useful when defining agreement features, such as gender or number.

An adjective inherits the feature values of its supertype. For instance:

In the resulting grammar, there will be three leaf copula types, one for each of the possible combinations of the two two-valued features, which can easily have lexical instances (lexicon entries) added to. Doing this helps the grammar remain structured, and to easily place additional constraints on types, say if it is discovered that plural nouns are inherently feminine.

Next to Copula Types, you find a(n experimental) link saying visualize copula hierarchy. It will display the supertype relations that are defined at that moment as a hierarchy.

Cycles of inheritance are not allowed (and they do not make sense anyway). So if cop_type2 is a supertype of cop_type1, cop_type1 may not be a supertype of cop_type2. Note that the supertype-subtype relation is transitive, so if a := b indicates a supertype relation where b is supertype of a, the following is forbidden as well:

a := b.
b := c.
c := a.

In the supertype hierarchy above, c is a supertype of a through b. So a cannot be a supertype of c

CAREFUL!!!!!!

Validation does not check whether constraints on supertypes and subtypes are compatible. This means that grammars with such errors can be created by the customization system, but they will not load in lkb and cannot be compiled to work with PET or ACE. An example of this would be the following:

CopType 1
Type name: regular-adj-comp-pl
supertype: regular-adj-comp-sg

CopType 2
Type name: regular-adj-comp-sg
This copula constrains its subject to be singular.

The regular-adj-comp-pl copula will inherit the property Feature:subject:singular from the regular-adj-comp-sg copula, but have the additional property of Feature:subject:plural from its own definition. This results in an ill-defined type, and should be avoided.

There is a bug report for this problem, but it is extremely hard to implement so that it works at the time of validation. So for now and the near future: pay attention while defining inheritance!

Complement Type

Choose the part of speech type that can appear as a complement to this copula type. Options are NPs, or Noun Phrases; PPs, or Prepositional Phrases; and APs, or Adjective Phrases. Currently, the only available option is APs, which generates a copula type in the output grammar with its complement constrained to be an adjective.

Features

Stems

Here you can define the actual stems of copulas belonging to a given class. You can add as many stems per auxiliary class as you want.

Copulas are analyzed as having empty semantics, so there is not an option for copula predicates.

Morphotactic Constraints

NOTE: morphotactic constraints on copulas are still under development and may not work as intended.

This part of the lexicon library allows you to define interactions with morphotactics, which morphemes are required and which are forbidden to combine with a given lexical type.

When you click either of the buttons, a constraint description appears with a drop-down menu. The position classes and morphological rules defined for verbs on the Morphology page appear in the drop-down. Select a class to define the constraint. You can define as many constraints as necessary.

A Require constraint makes sure the lexical item is only accepted as a word after combining with a morpheme belonging to the selected position class or an instance of the selected lexical rule (depending on whether a position class or lexical rule was selected).

A Forbid constraint makes sure the lexical item cannot combine with a morpheme belonging to the selected position class or an instance of the selected lexical rule (depending on whether a position class or lexical rule was selected).

Determiners

The following information can be provided for each determiner:

Type name

Defines the name of the class of auxiliaries (similar to nouns, verbs and determiners). It should be noted that:

Stems

Here you can define the actual stems of determiners belonging to a given class. You can add as many determiner stems per class as you want.

For each determiner stem, spelling and a predicate must be provided. After you fill out the spelling and hit tab, a predicate name based on the stem and MRS standards is suggested. This has the following form:

Spelling: myStem     Predicate: _myStem_q_rel

You can change the predicate if you like, but it is recommended to stick to the conventions of starting with _ and ending quantifier relations by _q_rel.

Features

The Feature option for determiners has two fields name and value. You can select a feature under name and its value(s) under value.

The drop-boxes will show features and values based on what was filled out at the pages Person, Number, Gender, Case, and Other Features. The first four libraries (Person, Number, Gender, Case) each define features typically appropriate for nouns, which can also be assigned to determiners. Other Features allows you to define customized features. These features can be assigned to nouns, verbs or both. Features that are defined for both categories or for nouns will appear in the drop-box.

Select the feature you want to specify from the left drop-box. Appropriate values will appear in the right drop-box. Each value is preceded by a check box. You can select as many values as you want for each feature (but note that selecting all values leads to the same behavior as not specifying a feature value pair at all).

If you select more than one value for a feature, the determiner class will assign an underspecified value to the feature that may unify with any of the selected values. The customization system will adapt the type hierarchy of the feature value accordingly. See the analysis section for an example of the impact on the type hierarchy.

Morphological constraints

This part of the lexicon library allows you to define interactions with morphotactics, which morphemes are required and which are forbidden to combine with a given lexical type.

When you click either of the buttons, a constraint description appears with a drop-down menu. The position classes and morphological rules defined for determiners on the Morphology page appear in the drop-down. Select a class to define the constraint. You can define as many constraints as necessary.

A Require constraint makes sure the lexical item is only accepted as a word after combining with a morpheme belonging to the selected position class or an instance of the selected lexical rule (depending on whether a position class or lexical rule was selected).

A Forbid constraint makes sure the lexical item cannot combine with a morpheme belonging to the selected position class or an instance of the selected lexical rule (depending on whether a position class or lexical rule was selected).

Case Marking Adpositions

This part of the Lexicon page allows you to define case marking adpositions. Unlike the other lexical items on this page, you cannot define an adposition class, but only individual adpositions with grammmatical properties.

You can provide the following information for each adposition:

Note that these adpositions are meant to fill the role of case markers. They thus have no predicates of their own. If the adposition is optional, check the check box before 'optional' in the sentence following Spelling.

You can indicate whether the adposition is a preposition or postposition by indicating whether it appears before or after the noun phrase.

Features

The Feature option for case-marking adpositions has two fields name and value. You can select a feature under name and its value(s) under value.

The drop-boxes will show features and values based on what was filled out at the pages Person, Number, Gender, Case, and Other Features. The first four libraries (Person, Number, Gender, Case) each define features typically appropriate for nouns, which can all be defined on adpositions. Other Features allows you to define customized features. These features can be assigned to nouns, verbs or both. Features that are defined for both categories or for nouns will appear in the drop-box.

Select the feature you want to specify from the left drop-box. Appropriate values will appear in the right drop-box. Each value is preceded by a check box. You can select as many values as you want for each feature (but note that selecting all values leads to the same behavior as not specifying a feature value pair at all).

If you select more than one value for a feature, the adposition will assign an underspecified value to the feature that may unify with any of the selected values. The customization system will adapt the type hierarchy of the feature value accordingly. See the analysis section for an example of the impact on the type hierarchy.

Analyses

General: Introduced underspecification of feature values

If you select more than one value for a feature, the noun class will assign an underspecified value to the feature that may unify with any of the selected values. The customization system will adapt the type hierarchy of the feature value accordingly.

For instance, the basic customized hierarchy for the option 'first-second-third' person and 'none' first person distinction will look as follows:

person := avm.
1st := person.
2nd := person.
3rd := person.

If you define the following Feature-value pair:

Feature: PERSON value [x] 1st []2nd [x]3rd

The hierarchy will look like this:

person := avm.
2nd := person.
non-2nd := person.
1st := non-2nd.
3rd := non-2nd.

The resulting noun class will bear the feature-value pair [PERSON non-2nd]

Nouns

Generally included: Argument structure

A basic type for noun-phrases is introduced in the grammar. It looks as follows:

noun-lex := basic-noun-lex & basic-one-arg & no-hcons-lex-item &
   [ SYNSEM.LOCAL [ CAT.VAL [ SPR < #spr & [ LOCAL.CAT.HEAD det ] >,
                              COMPS < >,
                              SUBJ < >,
                              SPEC < > ] ],
     ARG-ST < #spr > ].

It inherits from three matrix.tdl types: basic-noun-lex, basic-one-arg and no-hcons-lex-item.

The argument structure makes it select for a specifier of category determiner ([HEAD det]). Note that this also applies to languages that do not have determiners: in these languages, all nouns will use the bare-np-phrase rule to cancel the determiner from the argument list. The application of the bare-np-rule is regulated through the feature [ OPT bool ] on the synsem of the specifier. The bare-np-rule can only apply when the determiner on the argument list is [ OPT + ], the head-specifier rule combining determiners with nouns, requires [ OPT - ] (which is added to the rule if the language has determiners, and nouns that cannot take one are defined in the questionnaire).

If all nouns in the language have the same requirements towards determiners (i.e. they are always obligatory, always impossible or always optional) the required constraint is regulated on the basic definition of noun-lex as follows (in partial structures of the noun-lex type presented above):

always obligatory determiners:
noun-lex := ...
 [ SYNSEM.LOCAL [ CAT.VAL [ SPR < #spr & [ LOCAL.CAT.HEAD det,
                                           OPT - ] >,
  .... ].

determiners are never allowed:
noun-lex := ...
 [ SYNSEM.LOCAL [ CAT.VAL [ SPR < #spr & [ LOCAL.CAT.HEAD det,
                                           OPT + ] >,
  .... ].

determiners are always optional:
noun-lex := ...
 [ SYNSEM.LOCAL [ CAT.VAL [ SPR < #spr & [ LOCAL.CAT.HEAD det ] >,
  .... ].

In the first case, the bare-np-phrase cannot apply, but the spr-head-phrase can combine the noun with a determiner. The [ OPT + ] feature blocks the spr-head-phrase, forcing the bare-np-phrase to apply to create NPs (even if the language has determiners, nouns can only be part of bare-np-phrases in this case). If they are always optional, the [ OPT bool ] feature remains underspecified, so that both rules can apply.

If restrictions towards determiners vary, subtypes of noun-lex introducing the right constraints are introduced:

If obligatory determiners occur:
obl-spr-noun-lex := noun-lex &
  [ SYNSEM.LOCAL.CAT.VAL.SPR < [ OPT - ] > ].

If nouns for which determiners are impossible occur:
no-spr-noun-lex := noun-lex &
  [ SYNSEM.LOCAL.CAT.VAL.SPR < [ OPT + ] > ].

Generally included: Case Marking

If the language has case, the feature [ CASE case ] is added to the head-type noun as an addendum:

noun :+ [ CASE case ]

Noun Class Specific: Type Name

The name provided to the class will serve as the type identifier in the lexicon with the suffix indicated above.

If a type has supertypes identified on the Lexicon page, these will be the only supertypes it receives. Else, the following supertypes will be provided:

Noun Class Specific: Supertypes

The supertypes of a noun class depend on:

The noun class will only have noun-lex as supertype if either of the following applies:

Else if no supertypes are defined, it will inherit from the obl-spr-noun-lex or no-spr-noun-lex, depending on which is applicable according to its behavior towards determiners.

If specific supertypes are defined on the questionnaire, they will appear as supertypes of the noun-class. Additionally, the noun class may take obl-spr-noun-lex or no-spr-noun-lex as a supertype. This happens when all supertypes have underspecified behavior towards determiners, and the noun class itself either has obligatory determiners or cannot occur with them at all.

Noun Class Specific: Feature values

All features associated with the noun class will be added to the language specific type definition.

See General: Introduced underspecification of feature values to see how the hierarchy is modified to include underspecified features, if you define multiple values to a feature. The created underspecified value is assigned to the feature in this case.

Noun Class Specific: Stems and Predicates

For each stem 'my_stem' and predicate 'my_pred' belonging to noun class 'my_nclass' will be added to the lexicon:

my_stem := my_nclass &
[ STEM < "my_stem" >,
  SYNSEM.LKEYS.KEYREL.PRED "my_pred" ].

If several words with the same stem occur in the language, the stem identifiers (my_stem left of the assignment sign := in the representation above) receive numbers.

Verbs

General: basic type definitions

If the language has auxiliaries, the addenda include:

head :+ [ AUX bool ].

The following basic types for verbs are included:

verb-lex := non-mod-lex-item &
                 [ SYNSEM.LOCAL.CAT.HEAD verb ].

main-verb-lex := verb-lex & basic-verb-lex &
       [ SYNSEM.LOCAL [ CAT [ VAL [ SPR < >,
                                    SPEC < >,
                                    SUBJ < #subj > ],
                              HEAD.AUX - ],
                        CONT.HOOK.XARG #xarg ],
         ARG-ST < #subj &
                  [ LOCAL [ CAT.VAL [ SPR < >,
                                      COMPS < > ],
                            CONT.HOOK.INDEX #xarg ] ], ... > ].
aux-lex := verb-lex &
                [ SYNSEM.LOCAL.CAT.HEAD.AUX + ].

Else, if there are no auxiliaries, only the following will be included:

verb-lex := verb-lex & basic-verb-lex &
       [ SYNSEM.LOCAL [ CAT.VAL [ SPR < >,
                                  SPEC < >,
                                  SUBJ < #subj > ],
                        CONT.HOOK.XARG #xarg ],
         ARG-ST < #subj &
                  [ LOCAL [ CAT.VAL [ SPR < >,
                                      COMPS < > ],
                            CONT.HOOK.INDEX #xarg ] ], ... > ].

basic-verb-lex is a definition from matrix.tdl, defining standard semantics [ KEY-REL event-relation ] and [ HEAD verb ].

The basic types for intransitive and transitive verbs are the following (where mainorverbtype stands for main-verb-lex when the language has auxiliaries and verb-lex when it does not:

intransitive-verb-lex := mainorverbtype  & intransitive-lex-item &
       [ SYNSEM.LOCAL.CAT.VAL.COMPS < > ].


transitive-verb-lex := mainorverbtype & transitive-lex-item &
       [ SYNSEM.LOCAL.CAT.VAL.COMPS < #comps >,
         ARG-ST < [ ],
                  #comps &
                  [ LOCAL.CAT [ VAL [ SPR < >,
                                      COMPS < > ] ] ] > ].

Verb Class Specific: Supertypes

The supertypes of a verb class depend on:

The verb class will only have main-verb-lex (if the language has auxiliaries) or verb-lex (if the languages does not have auxiliaries) as supertype if no supertypes are defined for the verb

Else if no supertypes are defined, it will inherit from the obl-spr-noun-lex or no-spr-noun-lex, depending on which is applicable according to its behavior towards determiners.

If specific supertypes are defined on the questionnaire, they will appear as supertypes of the noun-class. Additionally, the noun class may take obl-spr-noun-lex or no-spr-noun-lex as a supertype. This happens when all supertypes have underspecified behavior towards determiners, and the noun class itself either has obligatory determiners or cannot occur with them at all.

General: case marking

If the language has case marking, subtypes representing the basic case patterns of the languages are added to the language. A few examples are given below:

nominative accusative language:

nom-intransitive-lex := intransitive-lex-item &
  [ ARG-ST.FIRST.LOCAL.CAT.HEAD noun & [ CASE nom ]].

nom-acc-transitive-lex := transitive-lex-item &
  [ ARG-ST < [ LOCAL.CAT.HEAD noun & [ CASE nom ] ], [ LOCAL.CAT.HEAD noun & [ CASE acc ] ] > ].


ergative language:

abs-intransitive-lex := intransitive-lex-item &
  [ ARG-ST.FIRST.LOCAL.CAT.HEAD noun & [ CASE abs ]].

erg-abs-transitive-lex := transitive-lex-item &
  [ ARG-ST < [ LOCAL.CAT.HEAD noun & [ CASE erg ] ], [ LOCAL.CAT.HEAD noun & [ CASE abs ] ] > ].

Verb class specific: supertypes

The supertype of a verb depends on its argument structure and the supertypes defined on the questionnaire. Verb classes inherit from the argument-structure types described above as well as the user-defined supertypes.

Auxiliaries

Depending on the definition of the complement of the verbal complement of auxiliaries on the word order page, one of the following basic types will be provided:

sentential complements:
aux-lex := basic-one-arg &
              [ SYNSEM.LOCAL.CAT.VAL [ SUBJ < >,
                                       COMPS < #comps > ],
                ARG-ST < #comps & [ LOCAL.CAT [ VAL [ SUBJ < >,
                                                      COMPS < >,
                                                       SPR < >,
                                                       SPEC < > ],
                                                HEAD verb ]] > ].


vp complements:
aux-lex := trans-first-arg-raising-lex-item  &
                     [ SYNSEM.LOCAL.CAT.VAL.COMPS < #comps >,
                       ARG-ST < [ ],
                                #comps &
                                [ LOCAL.CAT [ VAL [ SUBJ < [ ] >,
                                                    COMPS < >,
                                                    SPR < >,
                                                    SPEC < > ],
                                              HEAD verb ]] > ].

v complements:
aux-lex := basic-two-arg &
             [ SYNSEM.LOCAL.CAT.VAL.COMPS < #comps . #vcomps >,
               ARG-ST < [ ],
                      #comps &
                      [ LIGHT +,
                        LOCAL [ CAT [ VAL [ SUBJ < [ ] >,
                                            COMPS #vcomps ],
                                      HEAD verb ],
                                CONT.HOOK.XARG #xarg ]] > ].

For vp-complements, the subject is raised (which is defined on the supertype it takes).

The subject of verbs taking v-complements is underspecified. Its definition depends on the case restrictions defined for subjects of the auxiliary.

Determiners

The following basic type is introduced for determiners:

determiner-lex := basic-determiner-lex & basic-zero-arg &
          [ SYNSEM.LOCAL.CAT.VAL [ SPR < >,
                                   COMPS < >,
                                   SUBJ < > ]].

The type inherits from basic-zero-arg (the customization system only covers determiners that do not take any arguments for now) and basic-determiner-lex in matrix.tdl. The definition of basic-determiner-lex in matrix.tdl is the following:

basic-determiner-lex := norm-hook-lex-item &
  [ SYNSEM [ LOCAL [ CAT [ HEAD det,
                           VAL.SPEC.FIRST.LOCAL.CONT.HOOK [ INDEX #ind,
                                                            LTOP #larg ]],
                     CONT [ HCONS <! qeq &
                                   [ HARG #harg,
                                     LARG #larg ] !>,
                            RELS <! relation !> ] ],
             LKEYS.KEYREL quant-relation &
                   [ ARG0 #ind,
                     RSTR #harg ] ] ].

The most important property of this type is that it defines the quantifier relation of the determiner over the noun. The index of the determiner [ ARG0 #ind ] and [ SPEC....INDEX #ind ] on the noun are identical. The HARG (corresponding to the RSTR imposed by the semantics of the determiner) has scope over its LARG, which is identical to the LTOP of the noun.

Case Marking Adpositions

The basic type for case marking adpositions is presented below:

case-marking-adp-lex := basic-one-arg & raise-sem-lex-item &
          [ SYNSEM.LOCAL.CAT [ HEAD adp & [ CASE #case, MOD < > ],
                               VAL [ SPR < >,
                                     SUBJ < >,
                                     COMPS < #comps >,
                                     SPEC < > ]],
            ARG-ST < #comps & [ LOCAL.CAT [ HEAD noun & [ CASE #case ],
                                            VAL.SPR < > ]] > ].

The restruction on the SPR of the ARG-ST makes sure the adposition only takes NPs (not Ns) as an argument. Note that these adpositions are pure case-markers, they do not assign CASE to their NP complement, but they share a CASE value with the noun they are combined with.

References

Delancey, S. (2009). Bipartite Verbs in Languages of Western North America. In A. Filchenko and O. Potanina, (eds.), Time and Space in Languages of Various Typology. Proceedings of the XXV International Conference "Dulson Readings”. Tomsk: Tomsk State Pedagogical University.

Drellishak, S. (2009). Widespread but Not Universal: Improving the Typological Coverage of the Grammar Matrix. PhD thesis, University of Washington.

Jacobsen, W. H. (1980). Washo bipartite verb stems. In K. Klar, M. Langdon, & S. Silver (Eds.), Trends in Linguistics: Studies and Monographs 16: American Indian and Indoeuropean Studies: Papers in Honor of Madison S. Beeler (pp. 85-99). The Hague: Mouton.

MatrixDoc/Lexicon (last edited 2018-01-03 23:55:31 by OlgaZamaraeva)

(The DELPH-IN infrastructure is hosted at the University of Oslo)