Journal o f Psycholinguistie Research, Vol. 1, No. 2, 19 72
Rate and Pause Characteristics of Oral Reading R. S. Brubaker 1
Received January. 20, 1971
Twenty-three male subjects each read a paragraph aloud and graphic level recordings o f their readings were measured in order to determine the average sentence rate, phrase rate, and duration o f phrase and sentence pauses. It was found that noninitial sentences and phrases were significantly faster in rate and that pauses between sentences were significantly longer in duration. The findings are consistent with a reduction of uncertainty hypothesis that rate increases and pause decreases as uncertainty decreases.
INTRODUCTION In speaking or reading aloud, the production of speech does not proceed at an even pace. Even disregarding small unintentional "hems and haws" that occasionally disrupt fluency, there are frequent interruptions in the stream of speech. Obviously the speaker must pause to breathe. Within such breath groups are other s~ght pauses used to set off groups of words into phrases. In addition, there are other slight hesitations which check the smooth flow of speech, such as the momentary hiati associated with stop consonants or internal open junctures. Nor does speech proceed at a uniform rate. In general, stressed syllables are uttered more slowly. Vowels and diphthongs tend to be longer than consonants and vary also as a function of their vertical place of articulation (House and Fairbanks, 1953). Environmentally, vowels are lengthened or This research was supported in part by the Research Office of the College of Liberal Arts, The Pennsylvania State University. IDepartment of Speech, The Pennsylvania State University, University Park, Pennsylvania. 141
~) 1972 Plenum Publishing Corporation, 227 West 171Ih Street, New York, N.Y. 1001 I.
142
Brubaker
shortened depending upon whether the following consonant is voiced or voiceless (Delattre, 1962; Peterson and Lehiste, 1960). As a result, vocal rate varies from moment to moment during the act of speaking. However, another source of variation is reported in the research of Goldman-Eisler (1958). In studies of pauses and hesitations, she found shorter pauses prior to points of high transitional probability and faster rates subsequent to such points. Conversely, pauses occurring before relatively unexpected (low probability) sequences were longer and the sequences uttered more slowly. Hence, both pause and rate may vary according to the degree of uncertainty encountered by the speaker at the moment of encoding. The present study is an examination of rate and pause characteristics of speech in oral reading. The primary purpose is to study the variation of rate and pause as related to sentences and phrases in a prose reading passage.
PROCEDURE
A number of young adult males were chosen from randomly selected sections of a beginning speech course required of all students at The Pennsylvania State University. These men were requested to read the first paragraph of the "Rainbow" (Fairbanks, 1957) passage for audio tape recording. Recording was accomplished using a high-quality microphone-tape recorder system located in a standard two-room, acoustically treated laboratory. The subjects were instructed to read the material silently once and then aloud, as if to a classroom audience of 20 people. The latter reading allowed the researcher to adjust equipment levels and to correct any errors in pronunciation. After an opportunity for questions concerning the nature of the task, the subjects recorded the passage. Next, the recordings were monitored by the investigator and an assistant to select the performances to be analyzed. Only those subjects who read the material without error or mispronunciation, who spoke general American dialect, and whose speech was free from clinical deviations were retained for study. The audio recordings were converted into visual traces by means of a graphic sound level recorder which yielded a time display of the voltage variations of the voice recording. Previous to this step, the recorder system had been calibrated and found to have an average error of 3%. The tracings for each subject were inspected and those found to be noise-free during silent intervals were selected for measurement. As a result of this selective process, the recordings of 23 subjects were analyzed for this study.
Rate and Pause Characteristics of Oral Reading
143
The durations of pause 2 and vocalization as traced on the level recordings were measured by two assistants working independently. Discrepancies between the two sets of measures were resolved by a third measurement. Following this, the visual record was compared with the audio record in order to identify all utterances and pauses. Finally, the durations were tabulated for statistical analysis.
RESULTS AND DISCUSSION Sentence Rate Here the data consisted of values indicating the duration of each sentence for each subject. From these values the rate in words per minute for each sentence was calculated for each subject. Table I presents the mean sentence rate and sigma for the six sentences of the passage read. With the exception of the third sentence, there is a progressive increase in rate from the first to the sixth, and final sentence. An analysis of variance, summarized in Table II, was performed to determine if the sentences differed significantly in rate. The obtained F ratio 2pause refers to intended interruptions of vocal output. In the error-free readings analyzed, pauses occurred at comma and period locations.
Table I. Average Sentence Rate and Sigma (in wpm) for Six-Sentence Reading Passage (N = 23) Sentence Ave. rate S~ma
1
2
215.96 17.24
220~9 14.17
3
207.14 16.95
4
5
6
228.86 23.76
253.24 26.67
272.00 22.60
Table II. Summary of Analysis of Variance of Sentence Rates Source
df
Sum of squares
Between Ss Within Ss Sentence Residual Total
22 115 5 110 137
39,690.97 87~654.85 70,758.44 16,896.41 127,345.82
aF 0.99 (5,110)= 3.20.
Mean square
F
14,151.69 153.60
92.13 a
144
Brubaker
of 92.13, well beyond the 0.01 level, supports the conclusion that nonchance differences exist among sentences. Since the F statistic indicates the presence of real differences, the Newman-Keuls 0/diner, 1961) procedure was employed to find which sentences differed beyond chance. The results of this procedure are summarized in Table III, where it may be seen that of the 15 one-way comparisons possible, only 3 failed to achieve the 0.01 level of confidence. On the basis of the statistical findings, one can conclude that the rate increases were nonchance effects related to the position of sentences in the passage. Faster rates occurring toward the end of the passage suggest a goal gradient effect. That is to say, the subjects tended to speed up in their performance as they neared the end of the passage, presumably in order to terminate the laboratory task more quickly. On the other hand, it may be that passage uncertainty decreases progressively as the reader approaches the completion of a passage. As a result of increased awareness of what the passage entails, the reading performance becomes more rapid, a Phrase Rate
For each subject phrase rate was calculated by dividing the number of words in a phrase by its duration and multiplying by 60 to express the result 31t is poss~le that the goal gradient effect may be due to increased certainty as to the outcome o f a performance.
Table III. Newman-Keuls Test of Differences Between Pa~s of Sentence Rate Means Sentences TotMs 3. 4764.17 1. 4967.05 2. 5073.47 4. 5263.87 5. 5824.43 6. 6255.92
3 4764.17
1 4967~5 202.88
9.99(r,110) 3.71 ~n MSres)zA = 59.44 (n MSres)89= 9.99(r,110) i
II
2
4
5
6
5073A7 309.30 a 106A2
5263.87 499.70 a 296.82 a 190A0
5824.43 1060.26 a 857.38 a 750~6 a 560.56 a
6255.92 1491.75 a 1288.87 a 1182.45 a 992.05 a 431.49 a
4.21 221.12
4.52 250.12
ii
asignificant at the 0.01 level o f confidence.
4.73 268.67
4.89
290.66
281.15 i i
Rate and Pause Characteristics of Oral Reading
145
Table IV. Mean Phrase Rate and Sigma (in wpm)
Sentence One (N = 16) First phrase Second phrase Sentence Four (N = 14) First phrase Second phrase Sentence Five (N = 17) First phrase Second phrase Sentence Six (N = 13) First phrase Second phrase
Mean
Sigma
211.83 285.85
21.I 1 62.25
3.82a
212.96 277.12
22.79 17.62
10.53a
221.79 315.84
37.24 24.43
8.53a
269.03 322.19
16.82 19.34
16.74a
gsignificant at the 0.01 level.
in words per minute. The data for subjects who phrased a given sentence alike were grouped and the average rates calculated. Since some readers phrased sentences differently, all did not contribute to the values reported in Table IV, nor are the same subjects represented in each sentence. For example, in the table above are found the average rates on the first and second phrase for Sentences One, Four, Five and Six. Only 16 of the 23 subjects furnished the data for the first sentence. On the other hand, the fifth sentence was phrased alike by 17 subjects. Not all of the 16 subjects providing data for Sentence One contributed to the data for Sentence Five. In general, within a sentence the subjects tended to pause at comma locations. Of the four sentences considered here, over half of the subjects paused after the word "air" in Sentence One, after "legend" in Sentence Four, after "look" in Sentence Five and after "reach" in Sentence Six. The major point of interest in Table IV is the significant increase in rate associated with the second phrase. Generally, these latter rates are approximately one-third faster than those of the first phrase. The differences between first-phrase and second-phrase rates range from about 50 words per minute in the last sentence to over 90 words per minute in the fifth sentence. Reference to the t values at the right of Table IV shows that all differences exceed the 0.01 level; demonstrating that the first and second phrases differ beyond expectations attributable to chance. Several possibilities suggest themselves to explain the positional effects seen in Table IV. Uncontrolled phonetic factors peculiar to the reading
146
Brubaker
material or its phrase construction may have contributedthe effects.4 Then, too, limitations on the breath supply available for the production of second phrases may have produced faster rates on second phrases, but in view of the slower rates commonly observed in the speech of patients with emphysema, this explanation seems unlikely. An interesting alternative explanation is that speakers' uncertainties diminished as they read. The increase in rate associated with the second phrase may be related to reduction of uncertainty as one reads a thought unit aloud. This explanation does not deny such phonetic influences as stress, etc. which were mentioned earlier. However, it is possible that in reading aloud one's sureness of the o u t c o m e increases as one proceeds through each sentence. Knowledge of later words reduces a reader's uncertainty as to the meaning of earlier words in any sentence and faster rate may be the response. Also, as a subject reads aloud, his oral performance lags behind his silent reading potentiality. It may be that oral rate increases when a subject completes his silent reading on a sentence because at that point he becomes reasonably sure of the full meaning of the unit. If this view is correct, then one can advance the hypothesis that in oral reading rate increases as uncertainty about the meanings of sentences decreases. Pause Duration Finally, this investigation explored the pauses separating phrases and sentences. Table V presents the means and sigmas of the average phrase and sentence pauses for all 23 subjects. As shown in the table, the average pause between sentences was slightly less than twice as long as the average pause separating phrases. The t statistic reveals that the difference is significant at the 0.01 level. This difference, coupled with intonation, should be linguisti4Currently, research is being conducted to investigate first and second phrase effects using an experimental passage in which the order of the phrases can be reversed.
Table V. Average and Sigma of Phrase and Sentence Pause Durations (in see) i
Ave. phrase pause Ave. sentence pause ,i,
aSignificant at 0.01 level.
I
ii
9
Mean
Sigma
0.36 0.69
0.1039 0.1264
t
532.44 a
Rate and Pause Characteristics of Oral Reading
147
tally useful to listeners, the shorter pauses signaling to auditors that there is more to come. However, the findings with respect to rate also suggest an alternative view. The longer pauses which separate sentences may be regarded as phenomena initiating the material which foUows. So viewed, the larger duration of pauses between sentences "fit" the hypothesis that rate increases as uncertainty is reduced, for as Goldman-Eisler pointed out, longer pauses preceded high entropy verbal sequences while shorter pauses seem to be associated with more predictable sequences. Thus, if, as postulated above, the first phrase of a sentence represents material of low certainty, an oral reader will pause longer before starting and will proceed at a slower rate on the initial portion of a sentence.
CONCLUSIONS The results of this study of rate and pause in oral reading indicate that sentences near the end of a paragraph are read aloud at a significantly faster rate than are those occurring earlier. The second phrase of two-phrase sentences occurring in the paragraph was also read at a significantly faster rate than the first phrase. Comparisons of the durations of pauses occurring between sentences with those occurring between phrases revealed that the former are significantly longer than the latter. These variations in rate and durations of pauses support the hypothesis that, although phonetic factors cannot be discounted, a major determinant of rate and pause is the degree of uncertainty a reader confronts from moment to moment during his oral performance. Thus, rate and duration of pauses in oral reading may not be controlled by the demands of the communicative necessities of interpreting meaningfully for listeners, but by the certainty of the performer during the act of reading.
REFERENCES Delattre, P. (1962). I. Acoust. Soc. Amer. 34, 1141L. Fairbanks, G. (1957). Voice and Articulation Drillbook. Harper & Row, New York, p. 127. Goldman-Eisler, F. (1958). Quart. J. Exper. Psych. 10, 96. House, A., and Fairbanks, G. (1953). Z Acoust. Soc. Amer. 25, 105. Peterson, G. and Lehiste, I. (1960). J. Acoust. Soc. A, mer. 32, 693. Winer, B. (1961). Statistical Principles in Experimental Design. McGraw-Hill, New York, p. 114.
Rate and Pause Characteristics of Oral Reading R. S. Brubaker 1
Received January. 20, 1971
Twenty-three male subjects each read a paragraph aloud and graphic level recordings o f their readings were measured in order to determine the average sentence rate, phrase rate, and duration o f phrase and sentence pauses. It was found that noninitial sentences and phrases were significantly faster in rate and that pauses between sentences were significantly longer in duration. The findings are consistent with a reduction of uncertainty hypothesis that rate increases and pause decreases as uncertainty decreases.
INTRODUCTION In speaking or reading aloud, the production of speech does not proceed at an even pace. Even disregarding small unintentional "hems and haws" that occasionally disrupt fluency, there are frequent interruptions in the stream of speech. Obviously the speaker must pause to breathe. Within such breath groups are other s~ght pauses used to set off groups of words into phrases. In addition, there are other slight hesitations which check the smooth flow of speech, such as the momentary hiati associated with stop consonants or internal open junctures. Nor does speech proceed at a uniform rate. In general, stressed syllables are uttered more slowly. Vowels and diphthongs tend to be longer than consonants and vary also as a function of their vertical place of articulation (House and Fairbanks, 1953). Environmentally, vowels are lengthened or This research was supported in part by the Research Office of the College of Liberal Arts, The Pennsylvania State University. IDepartment of Speech, The Pennsylvania State University, University Park, Pennsylvania. 141
~) 1972 Plenum Publishing Corporation, 227 West 171Ih Street, New York, N.Y. 1001 I.
142
Brubaker
shortened depending upon whether the following consonant is voiced or voiceless (Delattre, 1962; Peterson and Lehiste, 1960). As a result, vocal rate varies from moment to moment during the act of speaking. However, another source of variation is reported in the research of Goldman-Eisler (1958). In studies of pauses and hesitations, she found shorter pauses prior to points of high transitional probability and faster rates subsequent to such points. Conversely, pauses occurring before relatively unexpected (low probability) sequences were longer and the sequences uttered more slowly. Hence, both pause and rate may vary according to the degree of uncertainty encountered by the speaker at the moment of encoding. The present study is an examination of rate and pause characteristics of speech in oral reading. The primary purpose is to study the variation of rate and pause as related to sentences and phrases in a prose reading passage.
PROCEDURE
A number of young adult males were chosen from randomly selected sections of a beginning speech course required of all students at The Pennsylvania State University. These men were requested to read the first paragraph of the "Rainbow" (Fairbanks, 1957) passage for audio tape recording. Recording was accomplished using a high-quality microphone-tape recorder system located in a standard two-room, acoustically treated laboratory. The subjects were instructed to read the material silently once and then aloud, as if to a classroom audience of 20 people. The latter reading allowed the researcher to adjust equipment levels and to correct any errors in pronunciation. After an opportunity for questions concerning the nature of the task, the subjects recorded the passage. Next, the recordings were monitored by the investigator and an assistant to select the performances to be analyzed. Only those subjects who read the material without error or mispronunciation, who spoke general American dialect, and whose speech was free from clinical deviations were retained for study. The audio recordings were converted into visual traces by means of a graphic sound level recorder which yielded a time display of the voltage variations of the voice recording. Previous to this step, the recorder system had been calibrated and found to have an average error of 3%. The tracings for each subject were inspected and those found to be noise-free during silent intervals were selected for measurement. As a result of this selective process, the recordings of 23 subjects were analyzed for this study.
Rate and Pause Characteristics of Oral Reading
143
The durations of pause 2 and vocalization as traced on the level recordings were measured by two assistants working independently. Discrepancies between the two sets of measures were resolved by a third measurement. Following this, the visual record was compared with the audio record in order to identify all utterances and pauses. Finally, the durations were tabulated for statistical analysis.
RESULTS AND DISCUSSION Sentence Rate Here the data consisted of values indicating the duration of each sentence for each subject. From these values the rate in words per minute for each sentence was calculated for each subject. Table I presents the mean sentence rate and sigma for the six sentences of the passage read. With the exception of the third sentence, there is a progressive increase in rate from the first to the sixth, and final sentence. An analysis of variance, summarized in Table II, was performed to determine if the sentences differed significantly in rate. The obtained F ratio 2pause refers to intended interruptions of vocal output. In the error-free readings analyzed, pauses occurred at comma and period locations.
Table I. Average Sentence Rate and Sigma (in wpm) for Six-Sentence Reading Passage (N = 23) Sentence Ave. rate S~ma
1
2
215.96 17.24
220~9 14.17
3
207.14 16.95
4
5
6
228.86 23.76
253.24 26.67
272.00 22.60
Table II. Summary of Analysis of Variance of Sentence Rates Source
df
Sum of squares
Between Ss Within Ss Sentence Residual Total
22 115 5 110 137
39,690.97 87~654.85 70,758.44 16,896.41 127,345.82
aF 0.99 (5,110)= 3.20.
Mean square
F
14,151.69 153.60
92.13 a
144
Brubaker
of 92.13, well beyond the 0.01 level, supports the conclusion that nonchance differences exist among sentences. Since the F statistic indicates the presence of real differences, the Newman-Keuls 0/diner, 1961) procedure was employed to find which sentences differed beyond chance. The results of this procedure are summarized in Table III, where it may be seen that of the 15 one-way comparisons possible, only 3 failed to achieve the 0.01 level of confidence. On the basis of the statistical findings, one can conclude that the rate increases were nonchance effects related to the position of sentences in the passage. Faster rates occurring toward the end of the passage suggest a goal gradient effect. That is to say, the subjects tended to speed up in their performance as they neared the end of the passage, presumably in order to terminate the laboratory task more quickly. On the other hand, it may be that passage uncertainty decreases progressively as the reader approaches the completion of a passage. As a result of increased awareness of what the passage entails, the reading performance becomes more rapid, a Phrase Rate
For each subject phrase rate was calculated by dividing the number of words in a phrase by its duration and multiplying by 60 to express the result 31t is poss~le that the goal gradient effect may be due to increased certainty as to the outcome o f a performance.
Table III. Newman-Keuls Test of Differences Between Pa~s of Sentence Rate Means Sentences TotMs 3. 4764.17 1. 4967.05 2. 5073.47 4. 5263.87 5. 5824.43 6. 6255.92
3 4764.17
1 4967~5 202.88
9.99(r,110) 3.71 ~n MSres)zA = 59.44 (n MSres)89= 9.99(r,110) i
II
2
4
5
6
5073A7 309.30 a 106A2
5263.87 499.70 a 296.82 a 190A0
5824.43 1060.26 a 857.38 a 750~6 a 560.56 a
6255.92 1491.75 a 1288.87 a 1182.45 a 992.05 a 431.49 a
4.21 221.12
4.52 250.12
ii
asignificant at the 0.01 level o f confidence.
4.73 268.67
4.89
290.66
281.15 i i
Rate and Pause Characteristics of Oral Reading
145
Table IV. Mean Phrase Rate and Sigma (in wpm)
Sentence One (N = 16) First phrase Second phrase Sentence Four (N = 14) First phrase Second phrase Sentence Five (N = 17) First phrase Second phrase Sentence Six (N = 13) First phrase Second phrase
Mean
Sigma
211.83 285.85
21.I 1 62.25
3.82a
212.96 277.12
22.79 17.62
10.53a
221.79 315.84
37.24 24.43
8.53a
269.03 322.19
16.82 19.34
16.74a
gsignificant at the 0.01 level.
in words per minute. The data for subjects who phrased a given sentence alike were grouped and the average rates calculated. Since some readers phrased sentences differently, all did not contribute to the values reported in Table IV, nor are the same subjects represented in each sentence. For example, in the table above are found the average rates on the first and second phrase for Sentences One, Four, Five and Six. Only 16 of the 23 subjects furnished the data for the first sentence. On the other hand, the fifth sentence was phrased alike by 17 subjects. Not all of the 16 subjects providing data for Sentence One contributed to the data for Sentence Five. In general, within a sentence the subjects tended to pause at comma locations. Of the four sentences considered here, over half of the subjects paused after the word "air" in Sentence One, after "legend" in Sentence Four, after "look" in Sentence Five and after "reach" in Sentence Six. The major point of interest in Table IV is the significant increase in rate associated with the second phrase. Generally, these latter rates are approximately one-third faster than those of the first phrase. The differences between first-phrase and second-phrase rates range from about 50 words per minute in the last sentence to over 90 words per minute in the fifth sentence. Reference to the t values at the right of Table IV shows that all differences exceed the 0.01 level; demonstrating that the first and second phrases differ beyond expectations attributable to chance. Several possibilities suggest themselves to explain the positional effects seen in Table IV. Uncontrolled phonetic factors peculiar to the reading
146
Brubaker
material or its phrase construction may have contributedthe effects.4 Then, too, limitations on the breath supply available for the production of second phrases may have produced faster rates on second phrases, but in view of the slower rates commonly observed in the speech of patients with emphysema, this explanation seems unlikely. An interesting alternative explanation is that speakers' uncertainties diminished as they read. The increase in rate associated with the second phrase may be related to reduction of uncertainty as one reads a thought unit aloud. This explanation does not deny such phonetic influences as stress, etc. which were mentioned earlier. However, it is possible that in reading aloud one's sureness of the o u t c o m e increases as one proceeds through each sentence. Knowledge of later words reduces a reader's uncertainty as to the meaning of earlier words in any sentence and faster rate may be the response. Also, as a subject reads aloud, his oral performance lags behind his silent reading potentiality. It may be that oral rate increases when a subject completes his silent reading on a sentence because at that point he becomes reasonably sure of the full meaning of the unit. If this view is correct, then one can advance the hypothesis that in oral reading rate increases as uncertainty about the meanings of sentences decreases. Pause Duration Finally, this investigation explored the pauses separating phrases and sentences. Table V presents the means and sigmas of the average phrase and sentence pauses for all 23 subjects. As shown in the table, the average pause between sentences was slightly less than twice as long as the average pause separating phrases. The t statistic reveals that the difference is significant at the 0.01 level. This difference, coupled with intonation, should be linguisti4Currently, research is being conducted to investigate first and second phrase effects using an experimental passage in which the order of the phrases can be reversed.
Table V. Average and Sigma of Phrase and Sentence Pause Durations (in see) i
Ave. phrase pause Ave. sentence pause ,i,
aSignificant at 0.01 level.
I
ii
9
Mean
Sigma
0.36 0.69
0.1039 0.1264
t
532.44 a
Rate and Pause Characteristics of Oral Reading
147
tally useful to listeners, the shorter pauses signaling to auditors that there is more to come. However, the findings with respect to rate also suggest an alternative view. The longer pauses which separate sentences may be regarded as phenomena initiating the material which foUows. So viewed, the larger duration of pauses between sentences "fit" the hypothesis that rate increases as uncertainty is reduced, for as Goldman-Eisler pointed out, longer pauses preceded high entropy verbal sequences while shorter pauses seem to be associated with more predictable sequences. Thus, if, as postulated above, the first phrase of a sentence represents material of low certainty, an oral reader will pause longer before starting and will proceed at a slower rate on the initial portion of a sentence.
CONCLUSIONS The results of this study of rate and pause in oral reading indicate that sentences near the end of a paragraph are read aloud at a significantly faster rate than are those occurring earlier. The second phrase of two-phrase sentences occurring in the paragraph was also read at a significantly faster rate than the first phrase. Comparisons of the durations of pauses occurring between sentences with those occurring between phrases revealed that the former are significantly longer than the latter. These variations in rate and durations of pauses support the hypothesis that, although phonetic factors cannot be discounted, a major determinant of rate and pause is the degree of uncertainty a reader confronts from moment to moment during his oral performance. Thus, rate and duration of pauses in oral reading may not be controlled by the demands of the communicative necessities of interpreting meaningfully for listeners, but by the certainty of the performer during the act of reading.
REFERENCES Delattre, P. (1962). I. Acoust. Soc. Amer. 34, 1141L. Fairbanks, G. (1957). Voice and Articulation Drillbook. Harper & Row, New York, p. 127. Goldman-Eisler, F. (1958). Quart. J. Exper. Psych. 10, 96. House, A., and Fairbanks, G. (1953). Z Acoust. Soc. Amer. 25, 105. Peterson, G. and Lehiste, I. (1960). J. Acoust. Soc. A, mer. 32, 693. Winer, B. (1961). Statistical Principles in Experimental Design. McGraw-Hill, New York, p. 114.
