Theoretical Background

The Role of Construction Grammar

As the name indicates, Fluid Construction Grammar shares a lot of common ground with other construction grammar theories. The common aspects of construction grammars, dubbed 'Vanilla Construction Grammar' by [Croft2005], can therefore be considered as essential foundations of FCG. The most important are:

1. Lexical and grammatical inventories consist of constructions, that is, pairings of form and meaning. Cognitive linguists have generalized this notion of a construction to representations that range from purely syntactic units to complex symbolic form-meaning associations. This allows for a uniform way of representing all grammatical knowledge, better known as the syntax-lexicon continuum. This view is very different from the sharp distinction between lexicon and grammar as proposed by many other syntactic theories;
2. A speaker's knowledge about her language is seen as a structured inventory or taxonomy of constructions. In FCG this structure is represented by the fact that - even though all the linguistic knowledge is represented in exactly the same way - some constructions may differ in their function when licensing a construct. This enables us to label the different levels of structure to facilitate linguistic analysis.
3. Construction Grammar is a usage-based model of language [Langacker2000]. The inventories available to speakers and hearers consist of constructions which can be highly specialised, perhaps only pertaining to a single case, or much more abstract, covering a wide range of usage events. There is no sharp distinction therefore between idiomatic and general constructions.

Fluid Construction Grammar has been implemented as a unification-based formalism, which has also been proposed for other types of construction grammar [Kay and Fillmore1999]. Unification is one of the core mechanisms of FCG together with the merge operation.

Additional Design Principles

Because Fluid Construction Grammar supports 'Vanilla Construction Grammar', it can be used as a theory-independent tool for linguistic research. This enables the experimenter to test different alternative hypotheses and shape the formalism to theory-specific needs. However, FCG does have some additional properties that might place it closer to particular branches of construction grammars or cognitive linguistics.

First of all, FCG stresses the uniform representation of linguistic knowledge by having the exact same grammatical inventory for both parsing and production. In other words, every construction has to be bi-directional. This bi-directionality is a tough but necessary challenge because we are interested in agents capable of both parsing and production.

FCG has also been designed to be totally open-ended with respect to the set of syntactic and semantic categories that are used by the grammar. This does not mean that the experimenter can not implement a given set of categories, but FCG itself does not assume any kind of universal categories. Constraints on feature values are expressed as propositions or predicate-argument clauses so that the set of categories can be expanded at any time. In our experiments there are typically hundreds or even thousands of new categories built. This openness of categories is in line with the Radical Construction Grammar approach which argues that linguistic categories are not universal, but construction-specific and subject to evolution [Croft2005].

FCG further subscribes to the notion of emergent grammar [Hopper1987]. Languages are not seen as stable entities, but as complex adaptive systems with a great deal of variation in all levels. Several conventions may be in competition with each other, or some might not be conventionalized yet. This is implemented by assigning scores to constructions that are adopted based on the communicative success of the interactions that use these conventions. Moreover, the system should still come up with a reasonable parse for ungrammatical or unknown parts in an utterance, or a consistent production of new meanings and situations.

Strongly connected to that principle, FCG offers a usage-based model of language from a multi-agent perspective. The convention of constructions and their success in communication depends on the communicative interactions and needs of the language users. This emphasizes the role of the actual speakers in language change instead of treating language in its idealized form. The multi-agent aspect also means that the agents might not have completely the same language inventories and that linguistic knowledge is always local to an agent.