There is a fundamental misunderstanding concerning the meanings of the terms 'semantics' and 'syntax'. I call this cognitive error the 'fundamental misunderstanding of linguistics'. Briefly, it is the idea that language is a group of meaning-bearing sounds. In fact, the main role of language is internal, ie thought is made from linguistic units- we hesitate to call them words because of the obvious confusion this would cause.
Endel Tulving created a model of memory he called ‘monohierarchical’ because it consisted of three general memory types, nested like russian dolls, such that each next level formed a single specialisation from the previous one. Procedural memory, the lowest category, had Semantic memory as its sole specialised subtype. Semantic memory in turn had Episodic (autobiographical) memory as its sole specialised subtype. This situation is depicted in the figure below-
But what did Tulving mean by the term ‘semantic’, which usually connotes linguistic matters? He said that...
Semantic memory is the memory necessary for the use of language. It is a mental thesaurus, organized knowledge a person possesses about words and other verbal symbols...(Episodic and semantic memory, Tulving E & Donaldson W, Organization of Memory, 1972, New York: Academic Press).
It is reasonable to replace his use of 'semantic' with 'linguistic', according to the quote above, as well as further reading of his work. That is, we can reasonably assume that both semantic and syntactic forms of mental representation represent special subtypes of the simpler, procedural kind of memory, the type created when one learns to ride a bike for the first time.
Consider the model depicted in the figure above. It consists of two levels, an upper semantic level, and a lower syntactic level. It also consists of two sides, that is, it is lateralised just like our brain, and for similar reasons. The left side consists of single or unique entities, while the right side allows for multiple, similar but not identical entities. The left upper side contains episodic or autobiographical data structures, representations of one individual's (ie the subject's) experiences. The right upper side ALSO includes the experiences of other subjects, which the primary subject has learned about from third party observation (in animals and humans), as well as linguistic communication (in humans only). Because of the automatic upwards entailment property of the entries (percept categories) in this memory, this upper right side becomes a useful store of general knowledge, ie facts. While the left side contains records of experiences (facts plus the circumstances of their acquisition), the right side only contains facts that are mutually concurrent. The right side cannot contain the learning experiences associated with all the newly acquired facts because this one individual only has access to its own experiences. Summarising, left ('wet') side stores memory content plus its personal context, while right ('dry') side stores memory content only.
The Venn-style (sets and subsets) diagram above depicts a possible implementation of a Tulving memory. The specialisation function at each level is obtained by grouping a specialised subset of nth level entities into a higher (n+1)th level entity. For example, several (say 10%).procedural memory atoms (lets call them 'posture changes', 'skills' or 'reflexes'), labelled 'P', have evolved into groups to make semantic level entities labelled 'S'. Imagine four translations joined by 90 degree rotations (say, 4 x P[90]). The group forms a rectangle, which is a semantic atom S[rect] - [1].
The next step, from semantic atoms to episodic records, might possibly occur as a side effect of the conversion of STM placeholders into LTM neural bias settings, a sleep-based process called memory consolidation. Finally, the process of entailment from retrograde episodic records to anterograde beliefs (involuntarily/internally produced) and/or reasoned thoughts (voluntarily/externally guided) occurs when all episodic traces are made labile, permitting pluripotent cause-effect inferences to be made.
1. The reader is encouraged to think of more biologically valuable action groupings, I have used this simple vector graphics example just to make the point.