Back in 1951, Karl Lashley pointed out that the production and comprehension of word sequences obey a complex Combinatorics and require a specific representation format. In an article published in 2015 in the journal Neuron and entitled " The Neural Representation of Sequences: From Transition Probabilities to Algebraic Patterns and Linguistic Trees ", my colleagues and I reviewed the full range of mechanisms for the cerebral representation of temporal sequences. Many animal species can represent probabilities and transition times from one item to another; group them into blocks (chunks); represent their ordinal structure (1st,2nd,3rd), and even identify algebraic patterns (AAB). However, none of these mechanisms is sufficient to account for the organization of syntactic structures. Syntactic structures form embedded tree-like structures called syntagms (or sentences ), which obey very specific recursive rules.
There is no shortage of arguments in favor of the existence of these structures. First, there are numerous documented cases of syntactic ambiguity, in which the same sequence of words can be represented by two distinct embeddings (e.g. " [black cab] driver " and " black [cab driver] ". Even an isolated word can be ambiguous (e.g. " unlockable " = [un-lock]able, which can be unlocked, or un[lock-able], which cannot be locked). These examples show that the human brain does not stop at the superficial sequence of syllables or words, but recodes them using a hierarchy of embedded trees. A second argument comes from the observation of ellipses or substitutions. A phrase can be recognized by the fact that it can be deleted or replaced by a shorter word, and in some sentences, embedded structures can be recognized by the fact that each of them can be abbreviated (e.g. " he [drove [to [this [big house]]] " = " he drove to this ", "he drove to it ", " he drove there ", " he did. "). A third argument comes from syntactic movement: syntactics can be moved out of their original position, for example when forming questions, cleaved sentences, etc., but always respecting the embedding of the sentence's word groups. Finally, long-distance dependencies, such as subject-verb agreement or the linking of a pronoun to a noun, cannot be correctly described at the level of the sentence's sequential structure, but only by considering the embedding of its syntagms. In " Les voitures qui doublent ce camion roulent trop vite ", the verb agrees in the plural, as required by the subject " les ", without the agrammality of the sequence " ce camion roulent " even being detected.
