Effects of accent typicality and phonotactic frequency on nonword immediate serial recall performance in Japanese

Yuuki Tanida, Taiji Ueno, Satoru Saito, Matthew A. Lambon RalphInterspeech 2010

TL;DR

This study investigates how our long-term knowledge of Japanese phonotactics (sound sequences) and pitch accent patterns influences our ability to recall sequences of unfamiliar words (nonwords). They found that while common sound patterns improve recall of the sounds themselves, typical pitch accent patterns are crucial for recalling the accent pattern accurately, suggesting these two types of linguistic knowledge operate somewhat independently in short-term memory.

What Problem Does This Paper Address?

Understanding how our brains store and retrieve information in short-term memory is a fundamental question in cognitive science. Previous research has shown that our existing knowledge of language, like the common sound combinations in a language (phonotactics) and the meaning of words, significantly aids short-term memory recall. For instance, real words are recalled better than made-up ones, and nonwords with common sound sequences are recalled better than those with rare sequences. However, the role of prosodic features, such as pitch accent, in short-term memory recall has been less explored. Japanese is a particularly interesting case because pitch accent is not just a subtle feature but can distinguish word meanings (e.g., "rain" vs. "candy"). This paper aims to bridge this gap by examining how both phonotactic frequency and accent typicality (how common a particular accent pattern is) affect immediate serial recall of nonwords in Japanese.

Methods

The researchers designed an experiment using immediate serial recall of three-mora Japanese nonwords. They manipulated two key factors:

  • Phonotactic Frequency: Nonwords were created to have either high or low phonotactic frequency, based on the commonality of their sound sequences in Japanese.
  • Accent Typicality: Nonwords were produced with three different pitch accent patterns: "flat" (low-high-low), "type 1" (high-low-low), and "type 2" (low-high-low). These patterns were chosen because their typicality in Japanese three-mora nouns is well-established (flat being most common, type 2 least common).

Participants heard sequences of three nonwords and were then asked to recall them orally in the correct order. Two scoring methods were used:

  • Phoneme Accuracy: Scored correct if all target phonemes were recalled in the correct serial order.
  • Accent Accuracy: Scored correct if the recalled nonword had the correct pitch accent pattern.

This factorial design allowed them to analyze the independent and interactive effects of phonotactic frequency and accent type on recall performance.

Datasets

The study did not use pre-existing datasets for training. Instead, they selected and created nonwords based on:

  • Phonotactic Frequency Data: Derived from a lexical corpus of Japanese newspapers to quantify the frequency of sound sequences.
  • Accent Typicality Data: Based on previous analyses of Japanese corpora that established the relative frequencies of different accent patterns in common nouns.

The nonwords themselves were recorded by a native Japanese speaker, creating a custom stimulus set for the experiment.

Key Insights and Ideas

  • Phonotactics Help Sounds, Accents Help Accents: A crucial finding is the apparent dissociation between phonotactic knowledge and accent knowledge. Phonotactic frequency significantly improved the recall of the sounds of the nonwords, but had no effect on recalling the accent pattern. Conversely, accent typicality influenced accent recall but not phoneme recall.
  • Accent Typicality Matters for Accent Recall: The more typical an accent pattern is in Japanese, the better participants were at recalling that specific accent pattern on the nonwords. This suggests that our long-term knowledge of common accent structures directly supports the short-term memory representation of those prosodic features.
  • "Regularization" of Accents: The most common errors in accent recall were "accent changes" from less typical patterns (especially type 2) towards more typical patterns (like flat). This "regularization" effect suggests a bias towards familiar, common accent structures when memory is challenged.
  • Independent Knowledge Systems: The results strongly suggest that long-term knowledge about phoneme sequences and long-term knowledge about accent patterns might operate as relatively independent systems, contributing to short-term memory performance in distinct ways.

Key Results

  • Phoneme Accuracy: Nonwords with high phonotactic frequency were recalled significantly better than those with low phonotactic frequency. There was also a significant effect of accent type, with flat-accented nonwords recalled less accurately than type 1 or type 2. However, there was no significant interaction between phonotactic frequency and accent type for phoneme accuracy.
  • Accent Accuracy: Accent typicality had a significant effect. Nonwords with the least typical accent pattern (type 2) were recalled less accurately than flat or type 1 patterns. Phonotactic frequency did not significantly affect accent accuracy, nor did it interact with accent type.
  • Error Analysis: The most frequent accent errors involved a shift from the least typical accent (type 2) towards the most typical accent (flat).

Limitations and Future Work

  • Limited Scope of Nonwords: The nonwords were all three moras long and followed a CVCVCV structure. Future research could explore if these findings generalize to nonwords of different lengths or structures.
  • Single Speaker: The nonwords were recorded by a single male speaker. While this controlled for speaker variability, it would be beneficial to test with multiple speakers to ensure the findings are not speaker-specific.
  • Independent Operation Hypothesis: While the results suggest independent operation of phonotactic and accent knowledge, the authors acknowledge this needs further examination. Future work could explore potential interactions or shared mechanisms more deeply, perhaps using different experimental paradigms or computational modeling.
  • Generalizability: The study focused on Japanese, a pitch-accent language. Investigating similar effects in non-accent languages or languages with different prosodic features would be valuable for understanding universal versus language-specific mechanisms of short-term memory.

Summary updated

BibTeX
@inproceedings{tanida2010effects,
  author = {Yuuki Tanida and Taiji Ueno and Satoru Saito and Matthew A. Lambon Ralph},
  title = {{Effects of accent typicality and phonotactic frequency on nonword immediate serial recall performance in Japanese}},
  booktitle = {Interspeech},
  year = {2010},
}
Figure