I. ntr oduc i Figure 3 [basm
1
Acoustic study to determine whether word-final nasals are devoiced, deleted or maintained in Romanian. • Petrovici (1930), in work on Romanian nasals claimed that word-final nasals following /s/, e.g., basm ‘fairytale’, are devoiced about 50% of the time. • Vasiliu (1965) in his work on Romanian phonology said that all nasals in Romanian that are word-final and follow a consonant are partially devoiced. • Devoicing occurs in the /sm/ sequence. • Devoicing occurs in the /mn/ sequence in this environment except in one speaker. • The method in which speakers realize the /mn/ sequence varies across speakers. • Articulatory data in combination with acoustic data is necessary to determine which part of the nasal sequence is the /m/ and which is the /n/, and to confirm the original of the broadband release noises. References Petrovici, Emile. 1930. De la nasalité en Roumain. Institul de Arte Grafice ‘Ardealul’. Cluj Tucker, Benjamin V. (2003) Moraic structure in Romanian. Presentation at annual meeting of the Linguistic Society of America. Vasiliu, Emanuel. 1965. Fonologia Limbii Române. Editura tiinific. Bucureti. * [email protected] I am grateful to the native Romanian speakers that were willing to donate their time to this research, and to Brad James for his design skills Figure 1 Voiced Nasal Duration 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 Final /mn/ Final /m/ or /n/ Medial /mn/ Location and type of Nasal Duration (sec) Figure 2 [basm] b a s gap m • A consistent release was identified at the end of the /mn/ sequence (Figures 4-8). •A silent period between the end of voicing and this nasal release provides evidence for the presence of a devoiced nasal (Figure 7). • Speakers vary in how they realize final /mn/ sequences (Figures 4-8). Time (s) 0 0.806677 0 5000 b a s m Figure 3 [basm ] Time (s) 0 0.698413 0 5000 I. Introduction • 7 Native Romanian speakers • Words read in frame sentences. • The target word is utterance final. • Duration of coda consonants, preceding vowels, and nasal releases measured. II. Methods Figure 4 An example from S6, who typically has a voiced portion of the nasal, then a voiceless period, then a brief low-amplitude broadband noise that is taken to be the oral release of the final nasal consonant. S2 and S4 show a very similar pattern (Figures 5-6). Figure 5 An example from S1. This speaker shows a pattern similar to the previous three, except that he also consistently has an additional low-amplitude brief broadband noise in the middle of the /mn/ sequence. This is probably the release of the /m/, so this speaker has a release noise for each nasal separately. Figure 6 S7 shows voicing nearly throughout the entire sequence. • Because of low amplitude and devoicing, it is difficult to identify the boundary between the /m/ and the /n/, so rather than measuring duration of the /m/ and the /n/, duration of the voiced part of the nasal sequence, of the voiceless part, and of the release noise (if present) are measured. III. Results IV. Discussion • Perception test using non-native listeners to determine what can be heard. • Ultrasound and video data to determine the timing of articulations of /mn/ sequence. • Oral and nasal air flow measurements to determine the the origin of the word-final /mn/ release and to confirm that nasals are devoiced rather than deleted. • Electroglottograph data to determine the exact timing of voicing in relation to the spectrogram. Word-final /sm/ sequences • 80.4% of /m/ tokens in the /sm/ sequences are devoiced. • Most voiced tokens are from one speaker (S5) (Figure 2). Data from another environment (where a vowel-initial word follows the word-final nasal sequence) suggest that even this speaker normally devoices (Figure 3), but was using very careful speech in this subset of the data. • 91.7% devoicing in /sm/ sequences if data from S5 is not considered. Word-final /mn/ sequences • Figure 1 indicates that /n/ is not deleted in word-final /mn/ sequences because the mean duration of /mn/ is longer than that of word-final /n/ or /m/. Time (s) 0 0.762086 0 5000 Figure 4 [domn ] S6 d o m n IV. Conclusions IV. Future Research Time (s) 0 0.792245 0 5000 Figure 7 [domn ] S1 d o m n release Time (s) 0 0.693968 0 5000 Figure 8 [domn ] S7 release release release /m/ release? d o m n devoicing devoicing Time (s) 0 1.01098 0 5000 Figure 5 [demn ] S2 release d e m n devoicing Time (s) 0 0.570748 0 5000 Figure 6 [lemn ] S4 release l e m n devoicing
Transcript of I. ntr oduc i Figure 3 [basm
Acoustic study to determine whether word-final nasals are devoiced, deleted or maintained in Romanian.
• Petrovici (1930), in work on Romanian nasals claimed that word-final nasals following /s/, e.g., basm ‘fairytale’, are devoiced about 50% of the time.
• Vasiliu (1965) in his work on Romanian phonology said that all nasals in Romanian that are word-final and follow a consonant are partially devoiced.
• Devoicing occurs in the /sm/ sequence.• Devoicing occurs in the /mn/ sequence in this environment
except in one speaker.• The method in which speakers realize the /mn/ sequence varies
across speakers.• Articulatory data in combination with acoustic data is necessary
to determine which part of the nasal sequence is the /m/ and which is the /n/, and to confirm the original of the broadband release noises.
ReferencesPetrovici, Emile. 1930. De la nasalité en Roumain. Institul de Arte Grafice‘Ardealul’. ClujTucker, Benjamin V. (2003) Moraic structure in Romanian. Presentation at annual meeting of the Linguistic Society of America.Vasiliu, Emanuel. 1965. Fonologia Limbii Române. Editura �tiin�ific�. Bucure�ti.
* [email protected] I am grateful to the native Romanian speakers that were willing to donate their time to this research, and to Brad James for his design skills
Figure 1 Voiced Nasal Duration
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
Final /mn/ Final /m/ or /n/ Medial /mn/
Location and type of Nasal
Dura
tion (s
ec)
Figure 2 [basm]
b a s gap m
• A consistent release was identified at the end of the /mn/ sequence (Figures 4-8).
• A silent period between the end of voicing andthis nasal release provides evidence for the presence of a devoiced nasal (Figure 7).
• Speakers vary in how they realize final /mn/sequences (Figures 4-8).
Time (s)0 0.806677
0
5000
b a s m�
Figure 3 [basm �]
Time (s)0 0.698413
0
5000I. Introduction
• 7 Native Romanian speakers• Words read in frame sentences.• The target word is utterance final.• Duration of coda consonants, preceding vowels, and nasal
releases measured.
II. Methods
Figure 4 An example from S6, who typically has a voiced portion of the nasal, then a voiceless period, then a brief low-amplitude broadband noise that is taken to be the oral release of the final nasal consonant. S2 and S4 show a very similar pattern (Figures 5-6).
Figure 5 An example from S1. This speaker shows a pattern similar to the previous three, except that he also consistently has an additional low-amplitude brief broadband noise in the middle of the /mn/ sequence. This is probably the release of the /m/, so this speaker has a release noise for each nasal separately.
Figure 6 S7 shows voicing nearly throughout the entire sequence.
• Because of low amplitude and devoicing, it is difficult to identify the boundary between the /m/ and the /n/, so rather than measuring duration of the /m/ and the /n/, duration of the voiced part of the nasal sequence, of the voiceless part, and of the release noise (if present) are measured.
III. Results
IV. Discussion
• Perception test using non-native listeners to determine what can be heard.
• Ultrasound and video data to determine the timing of articulations of /mn/ sequence.
• Oral and nasal air flow measurements to determine the the origin of the word-final /mn/ release and to confirm that nasals are devoiced rather than deleted.
• Electroglottograph data to determine the exact timing of voicing in relation to the spectrogram.
Word-final /sm/ sequences• 80.4% of /m/ tokens in the /sm/ sequences are devoiced.• Most voiced tokens are from one speaker (S5) (Figure 2).
Data from another environment (where a vowel-initial word follows the word-final nasal sequence) suggest that even this speaker normally devoices (Figure 3), but was using very careful speech in this subset of the data.
• 91.7% devoicing in /sm/ sequences if data from S5 is not considered.
Word-final /mn/ sequences• Figure 1 indicates that /n/ is not deleted in word-final /mn/
sequences because the mean duration of /mn/ is longer than that of word-final /n/ or /m/.
Time (s)0 0.762086
0
5000
Figure 4 [domn �] S6
d o m n�
IV. Conclusions
IV. Future Research
Time (s)0 0.792245
0
5000Figure 7 [domn �] S1
d o m n� release
Time (s)0 0.693968
0
5000Figure 8 [domn �] S7
release
release
release
/m/ release?
d o m n�
devoicing
devoicing
Time (s)0 1.01098
0
5000Figure 5 [demn �] S2
release
d e m n�
devoicing
Time (s)0 0.570748
0
5000Figure 6 [lemn �] S4
release
l e m n�
devoicing
