Introduction

What is the mechansim behind sentence production? Analysis of verbal behavior, for example speech error analysis, has produced fruitful results for production in spoken language. Here we aim to study the characteristics of speech disfluencies in the utterance of people with the Attention Deficit Hyperactivity Disorder (ADHD) condition to see whether attention deficits affects speech production. Specifically, I will compare two audio recordings of ten-minute-long monologue produced by an adult with ADHD condition when she was on medication and not on medication respectively. Taking medication is assumed to imply that the person will have no significant cognitive difference with those of people without ADHD. The hypothesis is that, when the participant is not on medication, there will be more speech disfluencies, both specifically stuttering-like ones and general disfluencies including meaningless repetition. This is motivated from the finding that ADHD children tend to produce significantly more stuttering-like disfluencies [4] and this effect might be perserved for grown-ups. I have also collected basic statistics of verbal behavior and looked for occurances of classic patterns found in laboratories like the tendency to to delay long noun phrases to the end of sentences, syntactic priming of passive voices, and put shorter phrases before longer ones. Results are mixed, but for disfluencies specifically, it might be that people with ADHD without taking medication will tend to produce more meaningless repetition in their speech.

Method

Analysis Goal

Analysis 1. The number of words, disfluencies, speech errors, long pauses, and sentence. The number of words in every minute and in every sentence.

Analysis 2. The Heavy-NP shift (the tendency to move long noun phrases to the end of a sentence, e.g., “I gave the book to the boy who I had met in class yesterday” [5]), syntactic priming of active and passive voices (the tendency to repeat active or passive voice of the preceding sentence when producing a new one), and short-before-long tendency (the tendency to prefer short-before-long word or phrase structures over long-before-short ones, e.g., “Ross and Rachel” [5]).

These goals are decided before the behavioral coding.

The Participant

A college student (female) with ADHD condition. The participant is usually on medication for study and work. Recruited through the author’s personal connection. The participant talked about her trajectory of psychology interest change and one research project on the day without medicine and talked about another two research experiences on the day with usual medicine.

Procedure

The participant was seated in front of the experimenter and instructed to put on audio recorder and start talking continuously for around 10 minutes until she receive a signal from the experimenter. The procedure is repeated under two conditions on two days. The control condition is when the participant is on medication as usual, while the treatment condition is when the participant did not take medicine on that day.

A 10-minute recording of continuous spontaneous speech is collected for each condition (00:00:06 - 00:10:04 in the treatment condition audio and 00:00:03 - 00:10:08 in the control condition audio). Then, the two recordings are first transcribed by OpenAI Whisper with time stamps saved, which are imported into ELAN [1] and manually polished. The resulting transcript is automatically analyzed by a Python script and manually examined. The Python script for importing time-aligned transcript into ELAN and analyzing is available at https://drive.google.com/file/d/1-VYiXHPIZ7EBaIidkTfAUwPztpBbHK6L/view?usp=sharing.

Transcription

The rules for transcription employed solely for the purpose of thisproject are the following:

A word is an unit in written language that cannot be split by spaces. For example, “what’s” count as a single word.
A sentence is a communication units. A communication unit is an independent main clause with none or some dependent subordinate clauses that would not make sense without the main clause, where a clause is a statement containing a noun phrase and a verb phrase [2]. For example, “Actually I really wanted to go to law school um because when I was in high school I's the I I did like mock trial which is like fake trials and I really liked that”. Here the text in bold is the main clause and the rest are dependent clauses. No punctuation other than period is used, since clause syntactic structure is not part of the analysis goals. Disfluencies and speech errors are not taken into account when parsing seentences.
A long pause is a space between words that last too longer than the typical pause time when the utterance is produced by reading the corresponding written script, usually longer than one second. In the transcript, a long pause will be marked as “[long_pause]”.
A disfluency is a meaningless filler word. For example, “um” (short pauses) and “uh” (long pauses). It allows the speaker to have more time to think. A repetition that is not due to none of a speech error, meaning error, nor pronunciation error is also taken as a meaningless filler that gives the speaker to think longer and therefore, count as a special kind of disfluency. In the transcript, it will be enclosed by a pair of square brackets, e.g., “But then [um but then] I think”. Long pauses do not count as disfluencies because they are silent.