Our theoretical background assumes the independence and semantic compatibility of verb meaning and construction meaning (García-Miguel 1995:24-25, Goldberg 1995). We think that the global meaning of a sentence combines the meaning of lexical items and the meaning of grammatical constructions in a non deterministic way, but in a process of partial compositionality (Langacker 2000:152). We also adhere to some tenets of frame semantics, and particularly to some practices of the FrameNet project, although there are also some important differences. Put simply, we think that the syntactic structure of the clause must be explained through semantics.
The verb evokes a complex conceptual representation that includes some basic participants in a scene. The syntactic alternations with the same verb provide alternate construals of the scene focusing on different facets of the situations. With this problems in mind, ADESSE aims to become a data base for the empirical study of the interaction between verb meaning and construction meaning.
The conceptual scenario evoked by a verb can be accounted for by describing the set of potential semantic arguments which that verb can be combined with. For example, the conceptual frame of escribir 'write' can be described by making use of four semantic roles: 0-Writer, 1-Text, 2-Recipient and 3-Topic. Though sometimes it is possible to express the whole set of semantic arguments, as in (a), syntactic constructions usually select a subset, profiling them in different ways and leaving the rest unexpressed, as in (b) or (c):
(a) Juan  le escribió una carta  a su madre  sobre sus recuerdos de infancia 
(b) Juan  escribió una carta 
(c) Juan  le escribió a su madre 
What definitively proves that syntax is not enough is that, sometimes, the same syntactic construction can be mapped with different configurations of semantic arguments. Compare examples (b) and (c) below, from the verb sustituir 'substitute, replace', <0-Agent, 1-Substituted (Old Entity), 2-Substitute (New Entity)>, where the syntactic pattern Subj DObj corresponds to two semantic schemas (0-1 and 2-1):
(a) Rijkaard  sustituyó a Xavi  por Deco 
(b) Rijkaard  sustituyó a Xavi 
(c) Deco  sustituyó a Xavi 
Finally, it is possible that one verb evokes, in different instances, frames corresponding to different semantic domains. For example, the verb enseñar admits uses as the following ones:
It seems clear that we must distinguish two frames, one corresponding to the domain of Cognition (examples a and c, roughly equivalent to English teach, despite the differences in syntactic construction ) and the other to Perception (example b, English show, despite the fact that the constructions is similar to that in a). In cases such as this one, we need different sets of semantic roles for labelling verb arguments <0-Teacher, 1-Thing taught, 2-Learner> vs. <0-Shower, 1-Thing shown, 2-Seer>, so we postulate two different verb senses.
In order to account for these and other similar facts, the design of our database takes a structure, whose main tables and relations are depicted in Figure 1
Each record of the BDS ("Clausula" in fig. 1) is linked to a table of syntactic schemata ("tblEsqSinSem") where we map each syntactic function with a participant index (the equivalent of "0", "1", and "2" in the examples above). Each schema, in turn, is linked to a verb sense (tblVerbSens), associated with a set of participant roles, and ascribed to one or more semantic classes (tblVerbClases).
(c) Proyecto ADESSE