Center Embedded and Cross Serial Dependency Learning with Rhyme and Reason

Artificial grammar learning (AGL) studies have long compared the learnability of abstracted grammars, such as those employing center-embedded (CE) and cross-serial (CS) dependencies. This area remains relevant because experimental findings contradict predictions based on grammatical complexity alone: CE structures are found to be either more difficult or not reliably different to process than CS ones, despite being grammatically simpler. Another area of AGL research has investigated the role of cue salience, finding that semantic relatedness or and prosodic marking between lexical elements can affect performance. This thesis synthesizes these two areas of AGL research, using two vocabularies with identical lexical material to compare the learnability of CE and CS grammars in which dependencies are determined by either semantic relatedness, or rhyme. Findings support the notion that performance on CE vs CS AGL tasks is shaped not only by formal structure, but also the nature of the relationship linking dependent elements.