Quantitative assessment of second language learners' fluency: comparisons between read and spontaneous speech

Area of Science:

• Speech Science
• Linguistics
• Computational Linguistics

Background:

• Assessing second language (L2) fluency is crucial for language acquisition research and pedagogy.
• Perceived fluency is often subjectively evaluated by human raters.
• Objective measures of speech could offer a more standardized approach to fluency assessment.

Purpose of the Study:

• To investigate the relationship between objective speech properties and perceived fluency in non-native Dutch speakers.
• To determine the feasibility of developing objective fluency tests based on quantitative speech analysis.
• To compare the predictive power of objective measures for read versus spontaneous speech.

Main Methods:

• Two experiments were conducted with non-native Dutch speakers.
• Experiment 1 analyzed read speech fragments from 60 speakers.
• Experiment 2 analyzed spontaneous speech fragments from 57 speakers.
• Speech was rated for fluency by human judges and analyzed using a continuous speech recognizer for objective measures.

Main Results:

• Objective speech measures demonstrated predictive power for perceived fluency ratings.
• The predictive accuracy was significantly higher for read speech compared to spontaneous speech.
• The effectiveness of objective measures varied depending on the speech type and speaker's task.

Conclusions:

• Objective speech analysis holds promise for developing automated fluency assessment tools.
• Current objective measures are more reliable for evaluating read speech than spontaneous speech.
• Further research is needed to refine objective measures for spontaneous speech analysis and diverse speaking tasks.