Agreement is found in a high proportion of the world’s languages, and its prevalence tends to make us forget what a puzzling phenomenon it is. Basically, agreement is the expression of information in the ‘wrong place’. This can be seen even in English sentences as trivial as: the system works intermittently versus the systems work intermittently. Why would we signal the number of systems redundantly on the verb (works versus work) when this is quite clear from system/systems? The information is available in the ‘right place’ (on the noun system) and redundantly also in the ‘wrong place’ (on the verb works). However there are cases where it appears only in the wrong place, as in the sheep is grazing versus the sheep are grazing. Agreement in English is in some ways rather straightforward. In other languages the phenomenon is much more complex.

The database encodes information on agreement in fifteen genetically diverse languages. For each language, we define its agreement in terms of controllers, targets, domains, categories and conditions. The database contains pointers to examples illustrating the instances of agreement. In addition, there are language reports describing the languages, giving sources and enabling the user to see how decisions were made. The database holds extensive data on a small sample of languages and is intended for typological investigation rather than primarily for statistical use.


The 'Languages' panel of the search interface allows you to refine a query by either language or language family, by making a selection from the relevant list. A query can be run on this selection alone or combined with search criteria from the Features and/or Conditions panels.

When using this panel, you should select either a language or a language family, but not both. Selecting an option from both menus results in an 'AND search' and will therefore only return results if the correct language family is selected, which by definition makes an 'AND search' in this context redundant.

Querying by different elements of agreement

In the database, we define agreement in terms of agreement controllers, targets, categories, domains and conditions. The search interface provides the user with drop-down menus for each of those five elements of agreement. In addition, categories are associated with targets and with controllers so that we allow for a mismatch between the two.

A simply query would be to search for a particular agreement controller, e.g. noun. For this query the database will return all records where the controller is a noun.

Further details of the drop-down menus available are provided below.

Controller properties

Comitative Noun Phrase
Construction of the type ‘Kim with Sandy’ where both elements can influence agreement.
Nominal Predicate
This term is used in instances of ‘back agreement’ where a verb agrees with the nominal predicate and not with the subject.
Non-canonical controllers
We have identified three types of non-canonical controllers, i.e. No Overt Controller, Defective Controller and No Possible Controller.
No Overt Controller
A controller which is not overtly present but could be supplied for emphatic or topic purpose. No Overt Controllers are, for example, used to describe pro-drop languages where the controller noun phrase or pronoun can be omitted.
Defective Controller
An overt controller which lacks the canonical features for controlling the target. For example clausal or infinitival subjects are defective controllers. In the database, nine subtypes of defective controllers are distinguished, i.e. cl (clause), attr phr(attributive phrase), Cnjd Int (Conjoined Interjection), dat subj (dative subject), inf phr (infinitive phrase) Intjctn (interjection) Nomnlz (nominalization), postp (postposition), that cl (that clause). In all instances in the database the defective controllers trigger ‘default’ agreement on the target.
No Possible Controller
An absent controller which cannot have any surface expression as for example the Italian weather verb piove which cannot have an overt subject controlling it.
Noun Phrase, Noun and Personal Pronoun
In the database a distinction is made between Noun Phrases, Nouns and Personal Pronouns. Hence, Personal Pronouns are not included under Noun Phrase. Where a controller consists of a Noun or Noun plus modifier, this is treated as a Noun Phrase controller. Within the Noun Phrase, the controller is given as Noun.
Various types of interrogative pronouns.

Target properties

Adverbials are verbal forms having adverbial suffixes (see Tsakhur language report).
A morphosyntactic device, which can convert all lexical categories and syntactic phrases in nominal modifiers (see Tsakhur language report).
Distributive Quantified Noun
Noun expressions such as fivesome, dozen. (see Tsakhur language report).
Finite Attributive
Attributive verb forms which are syntactically finite (see Tsakhur language report).
Noun Phrase, Noun and Personal Pronoun
In the database a distinction is made between Noun Phrases, Nouns and Personal Pronouns. Hence, Personal Pronouns are not included under Nooun Phrase. Where a controller consists of a Noun or Noun plus modifier, this is treated as a Noun Phrase controller. Within the Noun Phrase, the controller is given as Noun.
The target numeral is subdivided into Numeral_1 (‘one’), Numeral_2 (‘two’), and Numeral_3/4 (‘three’ and ‘four’). These are distinguished only when their agreement behaviour is different.
Predicate Adjective SF/LF
Long form and short form predicate adjectives. Their agreement patterns differ.
Special Pronominal Adjectives
The target special pronominal adjectives has been introduced in the database to deal with instances of agreement with items such as oneself, alone and alone, where they have unique agreement properties.
Various types of interrogative pronouns.

Contextual properties


In languages with ergative alignment, a distinction is made between subjects and antecedent which are S (subject of an intransitive clause) and A (subject of a transitive verb).


The abbreviations O and IO are used for object and indirect object respectively in the Antecedent-Anaphor domain.

Allocutive agreement is found in Basque, where a finite verb may be marked with a morph, known as an ‘allocutive marker’, which references the gender of addressee, when that addressee is 2nd person singular familiar/intimate and may or may not be an argument of the verb.
For pronominal affixes, where there may or may not be an antecedent, the meaning of Antecedent-Anaphor is somewhat stretched. We use it to include the function of these affixes where there is strictly speaking no antecedent but direct exophoric reference.
Agreement between Noun Phrases fulfilling the same syntactic function.
Special domain name for less usual agreement domains limited to the clause.
Abbreviation used for second predicates.
Verbal Complex
A clause containing a number of verbal forms which agree in final inflection (see Kayardild language report).


Conditions suffixed with an _M stand for morphological conditions. They are a kind of cooccurrence restrictions in the sense that a particular morphosyntactic category is a prerequisite for another. For example, in Russian number is a prerequisite for gender agreement which only occurs in the singular.

Distance between target and controller in number of words.
Measure terms such as liter.
Place of attachment
Target can be controlled by different controllers depending on where it is attached.
There is variability, but we don’t know what the condition on it is.

Refining searches

Searches can be refined by choosing values from more than one field. For instance, it is possible to search for a particular controller-target combination, or to search for target-category combinations.