RE: Emergent Semantics

Semantic Web Interest Group,
Artificial Intelligence Knowledge Representation Community Group,
Paola Di Maio,
Prachant Bradwell,

“In philosophy, systems theory, science, and art, emergence occurs when an entity is observed to have properties its parts do not have on their own” [1]. Also topical is adding semantics (triples, quads, predicate calculus expressions) to recursive tree-based and graph-based data structures.

We can envision a variety of tree data structure, such that nodes’ child nodes are sequences. The envisioned trees’ leaf nodes can be lexemes; the parent nodes of those can be phrases, then sentences, and so on. While such trees could be said to resemble parse trees, their root nodes needn’t represent sentences. Their root nodes could represent paragraphs or larger rhetorical structures. We can attach semantics to each tree node. The semantics on a node (e.g. a phrase or a sentence) can be said to include emergent semantics – semantics which are greater than the sum of the semantics of the parts.

When considering graph-based data structures, which can be considered as sets of edges, subgraphs come to be of particular interest as they, too, can have semantics attached to them. That is, sets of edges can have semantics attached to them, as can various subsets of edges. A specific type of graph, an event graph, is broached as topical for modeling narrative “fabula” [2]. In such a model, events can have semantics attached to them, as can graphs of events and subgraphs of events.

Populating the contents of the emergent semantics in the scenarios of natural language and narrative can be said to result from cognitive processes, processes occurring in the minds of humans or in intelligent systems. Resembling reader-response criticism [3], the contents of the emergent semantics which can be attached to lexemes, phrases, sentences, paragraphs and so on, and the contents of the emergent semantics which can be attached to events and event graphs, could result from a number of inferencing processes of human readers or intelligent systems [4][5][6]. That is, the contents of the emergent semantics, in the aforementioned scenarios, can be populated by processes of inferencing in human readers and in artificial intelligence systems.

The model broached for narrative “fabula” is, simply, an event graph where events, events graphs and subgraphs can all have semantics attached to them – semantics envisioned as populated by inferencing processes. A follow-up exercise could be modelling the “syuzhet” of argumentation and narrative, topics including stylistics shaping and guiding the aforementioned processes of inferencing.


Best regards,
Adam Sobieski

[1] https://en.wikipedia.org/wiki/Emergence
[2] https://en.wikipedia.org/wiki/Fabula_and_syuzhet
[3] https://en.wikipedia.org/wiki/Reader-response_criticism
[4] Graesser, Arthur C., Murray Singer, and Tom Trabasso. "Constructing inferences during narrative text comprehension." Psychological review 101, no. 3 (1994): 371.
[5] Goldman, Susan R., Kathryn S. McCarthy, and Candice Burkett. "Interpretive inferences in literature." Inferences during reading (2015): 386.
[6] Cardona-Rivera, Rogelio E., and R. Michael Young. "Desiderata for a Computational Model of Human Online Narrative Sensemaking." (2019).