Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Relations as an organising factor in memory
Poster Session B, Wednesday, September 30, 4:30 - 6:30 pm, Wangari Maathai
Fengyun Hou1,2, Alexander Anderson2, Markus Damian2, Nina Kazanina1; 1University of Geneva, 2University of Bristol
Current theories have considered knowledge representations as a semantic network of concepts linked via unlabelled associations (e.g. fox-animal, chick-animal; Collins & Quillian, 1969; Collins & Loftus, 1975). However, such associations are insufficient to capture the nature and structure of various relations binding entities in events (e.g. fox hid the chick’s toy, R(ACTION) [agent, object] = R(HIDE) [fox, toy]; R(POSSESSION) [possessor, possessed] = R(POSSESSION) [chick, toy]). Here, we investigate whether relations scaffold the organisation of novel events into memory. Twenty-four native Mandarin speakers memorised two types of sentences: (i) Event sentences, describing novel events in which five entities were bound via various relations (e.g. 狐狸把小鸡的玩具藏在毯子上的信封里. “The FOX hid the CHICK's TOY inside the ENVELOPE on the BLANKET”). In such sentences, relational distance among entities can be categorised as DIRECT (e.g. CHICK-TOY, directly related via a possessive relation) or INDIRECT (e.g. CHICK-BLANKET, linked only indirectly through intermediate entities). (ii) List sentences, simply enumerating five entities without specifying any relational structure (e.g. 这里有猴子，山羊，衬衫，柜子，灯笼. “Here are MONKEYS, GOATS, SHIRTS, CABINETS, and LANTERNS.”). Subsequently, participants completed a frequency-tagging EEG task, in which they viewed 60 s-long sequences of nouns, with each noun presented for exactly 166.7 ms (= 6 Hz). In the Event [List] sequence, every four adjacent nouns were sampled from a single Event [List] sentence and the 5th noun (<oddball>; oddball frequency = 1.2 Hz) was either a novel noun (not seen during the experiment, e.g. SOCK) or a ‘non-member’ noun that was previously seen but not from the same Event [List] sentence. This yielded four conditions of sequences: (1) Event-<Novel>: 小鸡-信封-玩具-毯子-[袜子] (CHICK-ENVELOPE-TOY-BLANKET-<SOCK> .....), (2) List-<Novel>: 柜子-猴子-山羊-灯笼-[袜子] (CABINET-MONKEY-GOAT-LANTERN-[<SOCK>....), (3) Event-<Non-member>: 小鸡-信封-玩具-毯子-[衬衫] (CHICK-ENVELOPE-TOY-BLANKET-<SHIRT>...), (4) List-<Non-member>: 柜子-猴子-山羊-灯笼-[狐狸] (CABINET-MONKEY-GOAT-LANTERN-<FOX>....). We predicted that the relations in the Event sentences would bind the five entities into a more tightly structured representation, whereas List sentences would be stored more fragmentarily. Thus, introducing a novel or non-member oddball should elicit stronger neural responses for Events than for Lists. Following EEG recording, participants completed a cued recall task in which they typed the memorised sentences based on cues randomly selected from the five entities of each sentence. EEG results revealed robust novelty-related oddball responses, but sentence type did not significantly modulate oddball neural responses. However, the cued recall results demonstrated better memory performance for Event sentences than for List sentences. Most critically, a Relational Distance Effect was found, i.e. in Event sentences, entities that were directly related to the cue were more likely to be retrieved than indirectly related ones. Taken together, these findings suggest that memory for complex events may involve parsing input into multiple structured propositional representations, in which relational information scaffolds various predicate-argument (role-filler) structures for participating entities. These findings call attention to the theoretical models that embed relations into their architecture (Anderson, 1983, 1993).
Topic Areas: Syntax and Combinatorial Semantics,