Where is the length effect? A cross-linguistic study.

Anne‐Catherine Bachoud‐Lévi, Emmanuel Dupoux, Laurent Cohen, Jacques Mehler

Abstract

Many models of speech production assume that one cannot begin to articulate a word before all its segmental units are inserted into the articulatory plan. Moreover, some of these models assume that segments are serially inserted from left to right. As a consequence, latencies to name words should increase with word length. In a series of five experiments, however, we showed that the time to name a picture or retrieve a word associated with a symbol is not affected by the length of the word. Experiments 1 and 2 used French materials and participants, while Experiments 3, 4 and 5 were conducted with English materials and participants. These results are discussed in relation to current models of speech production, and previous reports of length effects are reevaluated in light of these findings. We conclude that if words are encoded serially, then articulation can start before an entire phonological word has been encoded.

Introduction

When an individual identifies a concept and prepares to speak, the brain performs a complex sequence of tasks: retrieving the phonological code and building a motor plan for the vocal tract. Traditional models of speech production often assume serial encoding—the planning of a word's sounds from left to right. If these models also assume that the entire "articulatory plan" must be completed before speech begins, then longer words (e.g., hippopotamus) should take significantly longer to initiate than short ones (e.g., cat). This predicted delay is known as the length effect.

While some early studies reported finding this effect, Bachoud-Lévi and colleagues suspected those results were confounded by variables like word frequency or familiarity. This study seeks to determine if a pure length effect exists when these variables are strictly controlled.

Methods

The researchers conducted five experiments using two primary paradigms:

  1. Picture Naming: Participants named line drawings from the Snodgrass and Vanderwart (1980) set as quickly as possible.
  2. Symbol Naming: To eliminate visual recognition noise, participants learned to associate arbitrary symbols with specific words before being tested on their retrieval speed.
The vocabulary of symbols used in the symbol-naming task to control for visual processing noise.

Experimental Controls

Words were meticulously matched for frequency using the Brulex database for French and the Kuçera and Francis corpus for English. Crucially, stimuli were matched by their initial phoneme to ensure voice-activated microphones triggered consistently. The study spanned both French (syllable-timed) and English (stress-timed) to ensure cross-linguistic validity.

Key Results and Insights

The Frequency vs. Length Paradox

The experiments consistently revealed a significant frequency effect—participants were much faster to name common words—but failed to find a significant length effect. As shown in the data, the reaction time (RT) difference between monosyllabic and bisyllabic words was often negligible or even favored longer words, while the gap between high and low-frequency words remained robust.

Mean reaction times (RT) and error rates (Err) demonstrating a clear frequency effect but no significant length effect.

Across the studies, the findings were remarkably consistent:

  • French Results: In Experiment 1, monosyllables took 565ms to initiate, while bisyllables took 561ms.
  • English Results: In Experiment 3, monosyllables took 601ms vs. 602ms for bisyllables.
  • Statistical Power: Post-hoc analysis confirmed the sample size was sufficient to detect an effect as small as 11–15 milliseconds, yet no such effect emerged even when increasing complexity to three syllables.

The "Pipelining" Hypothesis

The lack of a length effect suggests that even if speech planning is serial, the brain likely uses a "pipeline" approach. Articulation of the first syllable begins as soon as it is ready, while the remainder of the word is compiled in the background. Previous reports of length effects likely measured the difficulty of retrieving rare or complex words rather than the time taken to plan their phonology.

Conclusion and Future Directions

While no length effect was found for individual words, it is well-documented that such an effect exists for sentences. This suggests a "minimal planning unit" that is larger than a phoneme but likely no larger than a single word. Future research may investigate if much longer words (4-5 syllables) eventually trigger a delay or if the planning unit is tied to specific prosodic structures like the metrical foot.

Summary updated

Abstract

Many models of speech production assume that one cannot begin to articulate a word before all its segmental units are inserted into the articulatory plan. Moreover, some of these models assume that segments are serially inserted from left to right. As a consequence, latencies to name words should increase with word length. In a series of five experiments, however, we showed that the time to name a picture or retrieve a word associated with a symbol is not affected by the length of the word. Experiments 1 and 2 used French materials and participants, while Experiments 3, 4 and 5 were conducted with English materials and participants. These results are discussed in relation to current models of speech production, and previous reports of length effects are reevaluated in light of these findings. We conclude that if words are encoded serially, then articulation can start before an entire phonological word has been encoded.
BibTeX
@inproceedings{bachoudlevi1998where,
  author = {Anne‐Catherine Bachoud‐Lévi and Emmanuel Dupoux and Laurent Cohen and Jacques Mehler},
  title = {{Where is the length effect? A cross-linguistic study.}},
  booktitle = {CogPrints},
  year = {1998},
}
Figure