the duration of vowel representing kasre izafat in the compound ...

119THE DURATION OF VOWEL REPRESENTING KASRE IZAFAT INTHE COMPOUND WORDS OF URDU LANGUAGEMIRZA FAHD ARSHED BAIGABSTRACTThe primary aim of this paper is to explorethe duration of the vowel involved in KasreIzafat in compound words of Urdu. Varioussentences, spoken by various speakers,were recorded and analyzed usingspecialized speech analysis software.Results suggest that the vowel is indeedshort. The paper also sheds light on thetopics of vowel duration and co-articulation.Furthermore, for the sake of analysis, stressrules of Urdu are also discussed in light ofearlier Urdu phonetic publications.1. INTRODUCTIONUrdu, along with Hindi, belongs to the familyof Indo-Aryan Languages. However unlikeHindi not much work, with regard toPhonetics and Phonology, is done on Urdu.In spite of this, among the few rules defined,there is a rule in Urdu that no word endswith a short vowel (Bukhari, 1985, p.18). Butin case of compound words, withpunctuations like the zair ( ) mark at theend of the first word, the possibility of a shortvowel existing at the intermediary wordboundary remains an open question. Thispaper aims to investigate this very situation.2. LITERATURE REVIEWTo investigate the vowels, specifically theirduration, it is vital to have knowledge of thefactors that affect vowel duration. Thereforevowel duration is reviewed. Furthermore acloser study of Urdu with respect to thosefactors is essential in order to successfullymeasure the duration.2.1 Vowel DurationVowels are greatly affected in duration by anumber of factors such as the identity of thefollowing consonant, the rate of speaking,the syllable stress, the number of syllablesin the word, the position of the word in thephrase or sentence, the type of word andthe importance or emphasis assigned to theword by the speaker (Pickett, 1999, p. 87).In languages it is commonly observed thatstress increases the duration significantly.However both the vowel quality and intensityonly contribute minimally to the increase.But this is certainly not true for all languagesas in the case of Estonian, whereunstressed vowels are longer than stressedones (Lehiste, 1970). Furthermore stressnot only increases duration but alsocorrelates to the intensity of formants(Napoli, 1996). Importance of the positionand word type on vowel duration can beseen from the fact that a result of anexperiment revealed that the longest vowelsare those that are stressed and occur at theend of phrases, clauses or sentences. Onthe other hand vowels appearing in thebeginning of a word are much shorter, induration, when stressed. This phenomenoncan be explained partly with the help of theword final lengthening effect that causes thefinal rhyme of a word to be less short then itactually may have been. Apart from theword final lengthening another reason canbe due to the characteristics of theconsonants following the vowel. Since it isobserved that voiceless stops shorten thepreceding vowel’s duration while, on thecontrary, nasal consonants lengthen theduration.Another factor affecting the vowel duration isthe rate of speaking. If the speed ofspeaking were fast then the time given tothe speech organs (tongue, lips and therest) to meet the targets required forproperly uttering the phonemes would beless then required. Thus the duration of thevowels and consonants too would decrease(but this decrease is more evident forvowels) (Pickett, 1999).

120Center for Research in Urdu Language Processing2.2 Co-ArticulationBesides the vowel duration factor anotherphenomenon that can potentially affectacoustic analysis is co-articulation. Coarticulationis a term used to refer to theinfluences of one sound on the articulationof other sounds in the same utterance. Thishappens mainly when the tongue articulationfails to attain certain targets. This failuremay be due to time shortage (if speakingfast) or by target anticipation. It isanticipatory because the movements for asequence of sounds, syllables and wordsseem to prepare for later articulatorypatterns. This results in current parts of theutterance, being spoken, to be affected. Infact Pickett says co-articulation is the rule inspeech production. Hence, it is common forconsecutive or adjacent phonemes to affecteach other (Pickett, 1999). Thus itbecomes imperative to keep co-articulationin mind during acoustic analysis of thespeech signal.2.3 Stress in UrduSince, stress is a key factor affecting vowelduration. So, a review of the rules regardingstress needs to be addressed. In Urdu,lexical stress doesn’t change the meaning ofthe word like it does in English. In Englishfor example, stress on the ‘in’ syllablemakes ‘insult’ a noun while the stress on the‘sult’ syllable makes ‘insult’ a verb.In Urdu the lengthening effect is not treatedas a supra-segmental feature but rather it,along with tone and stress, does affectintonation patterns. Irrespective of that,lengthening does play a vital role in therhythmic formation, as well as smooth flowof the sentence.In Urdu there are three short vowels [], []and [] for which punctuation marks paish( ), zair( ), zabar( ) are used respectively(Khan, 1997). One reason of why stress isnot distinctive in Urdu may be because thewords are not distinguished on the basis ofstress alone. Rather the stressing and destressingof syllables is linked to syllableweight. As a rule though, tense or stressedvowels are phonetically long (Kachru, 1987).According to another study done by Hussain(1997, p.121) lexical stress alters thephonetic properties of both vowels andconsonants (Among others, the duration ofboth, the vowels and consonants, alsoincreases). Furthermore long vowels in theinitial syllable are, on average, twice as longas short vowels in the initial syllable.Besides the durational increase due tostress is different for vowels in differentsyllables. Another result of the study wasthat the vowel duration increases least whenthe stressed vowel is in the initial syllableand most when it is in the last syllable (asreviewed earlier word final lengthening mayhave a role to play here) (Hussain, 1997, p.121).2.4 SummaryDuration of a vowel depends on manyfactors of which stress is very common andimportant. With respect to Urdu, stressdoes exist but may be unimportantphonetically as pointed out by some authors.Though stress does affect vowelsphonetically, by elongating that vowel.Co-articulation is also an important issuewhen acoustic analysis is being donebecause it is happening all the time. Theoccurrence of co-articulation can make itdifficult to identify the exact boundaries ofvowels accurately.3. MethodologyAs seen from the reviewed literature, in theprevious section, the zair ( ) punctuationmark would produce a short vowel (//)sound. On the other hand it is alsoreviewed that stress could elongate theshort vowel to some extent. Keeping thesethings in mind, an experiment is to bedevised such that the vowel (//) is verifiedas a short vowel.3.1 StimuliFor the proper verification of the short vowel,compound words were taken which had thezair ( ) punctuation mark at the end of thefirst word. Words were taken from the Urdulexicon and where possible highly familiarwords were selected. The experiment was

121designed with limited knowledge of thephonology and phonetics of Urdu.Furthermore ten most familiar words wereselected without any apparent considerationon their particular word structure. Table 1 isthe list of words used for the experiment.TABLE 1 Words, along with their phonetictranscription, used in the experiment.WordsQuaid-e-Azam(Founder ofPakistan)Rad-e-Amal(Reaction)Minar-e-Pakistan(A monument)Bagh-e-Jinnnah(Name of agarden)Aqwam-e-Mutahida(United Nations)Janab-e-Aali(Your Honour)Sitara-e-Imtiaz(Honorary medal)Azab-e-Qabr(Torment ofgrave)Dawat-e-Aam(Open invitation)Sahib-e-Hasiat(Capable person)Transcriptionqaid azmrd amalmnar pakstanba dnaaqvam mthdadnab alistara mtazazab qbrdawt amsahb hestThe bold symbol is the targeted short vowel.Each word was embedded in a carriersentence. Table 2 is the list of words thatcontained the selected long vowel (Theselected long vowel is bold and separatedwith space in order to clarify it).The transcription of the entire sentences isgiven in the Appendix A.microphone (low rated impedance (Z) of150-180Ω) with a frequency response of 80to 14000Hz. The microphone used, alsohad a built-in locking on/off switch on it.Furthermore a high quality integrated stereoamplifier supported the microphone. Allcoupled with a very high standard Sound-Blaster Card ensured minimum loss of datato noise and other interferences. Along withthat Praat 4. 0 sound recording system, TheSoundRecorder, was used (configured toproduce 16KHz WAV files).The Speech Analysis apparatus included thePraat 4.0 speech analyzer and the SpeechAnalyzer 1.5. Furthermore a pair of highperforming speakers was used. Beside thatMicrosoft Excel is used for storing the timedurations as well as doing the statisticalanalysis.TABLE 2 Words containing the selected longvowels.WordsAzadi(Independence)Shadeed(Extreme)Sair(Recreational tour)Bagh(Garden)Pantalis(fourty-five)Janab(Sir)Sitara(Star)Hamaen(We)Iftar(Breaking of fast)Ko(About)Transcriptionaz a did i ds æ rb a pent a lisdn a bst a rahm eft a rk o3.2 ApparatusThe apparatus used for the experiment canbe categorized into two broad categories:Recording apparatus and Speech Analysisapparatus. The recording apparatusincluded first and foremost a dynamic3.3 SpeakersSpeakers are an important part of theexperiment. Six male speakers are used inthe experiment, referred to as RS, AG, BA,AK, SS and GS. All of the six speakers’native language is Urdu. Besides, all of

122Center for Research in Urdu Language Processingthem are currently students pursuing abachelor’s degree in Computer Science fromNational University of Computer andEmerging Sciences (NUCES). All of thespeakers can understand Punjabi. AG andGS also speak Punjabi actively during theirdaily-life routine. All the speakers werejudged to have normal speech and hearing.All the speakers were naïve about thepurpose of the experiment. Two speakers,GS and SS, had no prior experience ofdoing any phonetic or speech experiment ofany sort. On the contrary the remainingfour have been participating in some sort ofphonetic or speech related experiments inthe current past.normalized vowel durations were alsocalculated and stored in thosespreadsheets.3.5 AnalysisFor analysis Praat 4.0 speech analysis tool,along with the Speech Analyzer was used.Both tools provide the facilities for inspectingthe spectrogram along with the properzooming and time duration calculationfacilities.3.4 ProceduresFirst, ten compound words of Urdu, with thezair ( ) punctuation placed at the end of thefirst word, are sought. These ten words aremade into ten sentences. In order to avoidemphatic stress and intonation, possiblyaffecting the vowel durations, thesesentences are created as simplestatements. All of these ten sentences arethen written legibly on ten cards (Each cardcontaining one sentence). All of the tencards are properly shuffled and presented tothe speaker.The speakers were advised to read thesentences out as if speaking them in theirdaily routine, keeping the sentences assimple statements and not introducing anysort of undue intonation or emphatic stress.Furthermore they were cautioned not torecite the sentences too slowly or too quick,but rather at their natural speaking pace.Each sentence was recorded one at a time.Once the ten sentences were recorded, thespeakers were given a short break while thecards were reshuffled. They then read thecards again. The whole procedure wasrepeated five times in a block-randomizeddesign. By doing this each speaker endedup recording five files per sentence. Thusthere were fifty files per speaker. And sincethere are six speakers so all in all there werethree hundred recorded files in total to beanalyzed. For each sentence, the durationof each speaker’s long and short vowels,measured in seconds, was recorded inMicrosoft Excel spread-sheets. TheFIGURE 1 Spectrogram, marking duration of //For analysis, time duration (in seconds) ofthe targeted vowel is measured alongsidewith the duration of any one of the longvowels spoken in the sentence (For thepurpose of proper comparison, the chosenlong vowel in a sentence remains the samefor each of the speaker). Due to the varyingspeaker speaking pace, care needs to betaken in selecting the long vowel. Forproper and effective interpretation thetargeted vowel duration is normalized that isits ratio with the long vowel is taken. Thiswould thus give us the relative length of thetargeted vowel with respect to the longvowel selected leading to a conclusiveinference of the result.For each vowel the time duration (inseconds) is calculated by subtracting thetime of the marker at the offset of the vowelwith the one on its onset. As in Figure1 theduration for the short vowel is measured bysubtracting the values indicated on the top

123of the window. The software shows bothtimes in seconds. It is to be noted that theduration marked in the figure is actuallydemarcated by properly zooming in to getthe proper limits of the vowel boundary.Furthermore co-articulation also comes intoplay, as a speaker’s natural speaking pacemay be fast enough to cause it. Asreviewed, the duration of the vowel in coarticulationis calculated betweenboundaries in which that vowel can bedistinctly heard and that the formants remainfairly constant. Figure 2 shows that case.Consequently, ten lowest values areextracted, one from each of the tensentences. Thus each speaker’s minimumNormalised Vowel Duration0.70.60.50.40.30.20.10RS AG BA AK SS GS AVGSpeakersMinimumMaximumAverageFIGURE 3 Maximum, minimum and averagevalues of Normalized Vowel Duration of eachspeaker.FIGURE 2 spectrogram, marking the duration of //when co-articulated with the vowel following it.4. ResultsFigure 3 shows the variation of vowelduration made by each of the six speakers.Firstly the Normalized Vowel Duration iscalculated by dividing the duration of theshort vowel, in seconds, with that of the longvowel, also in seconds. The figure alsoshows the minimum, average and themaximum values of the normalized vowel.Each speaker’s minimum value is calculatedby first selecting the smallest value of thefive recordings of every sentence.value is taken as the average of these tenvalues. Similarly the highest value amongthe five recordings of every sentence isextracted. This leads to ten highest valuesone from each sentence. The average ofthese ten values yields the averagemaximum value that the speaker achieved.Calculating the average value is just thesame. Instead of selecting the minimum ormaximum value, the average of the fiverecordings is calculated. Since there areten sentences, ten average values areextracted. The average of these tennumbers would give the average value perspeaker per sentence. Average bars arealso plotted alongside the six speaker bars.The minimum value is the average of the sixlowest values as plotted in the figure. It isthe case for the average’s maximum andaverage values.Figure 4 shows a line graph depicting themaximum, minimum and average voweldurations per sentence. For calculating themaximum value, an average of the fiverecordings is done for each of the sixspeakers. Then of those six averages themaximum is chosen. Calculating theminimum value is similar to that of themaximum value. Similarly for the average

124Center for Research in Urdu Language Processingvalue, the average of five recordings isdone. Finally the Average of the six averagevalues would yield an average value of allthe speakers per sentence.Furthermore the figure shows a thickhorizontal line just above the 0.5 mark.This is actually the average of the tenaverage values.Normalised VowelDuration0.90.80.70.60.50.40.35. Discussions1 2 3 4 5 6 7 8 9 10SentencesMaximumMinimumAverageAverage vowel lengthFIGURE 4: maximum, minimum and averagenormalized vowel duration of each sentence.The results suggest that the duration of theshort vowel // is on average less then thatof the long vowel. The long vowel is onaverage double in duration as compared tothe short vowel (approx 0.52 in ratio to along vowel).The range of duration of the short vowel,however, varied from speaker to speakerand also from sentence to sentence. Therange, though, was from approximately 0.3to 0.8, but that was by some speakers in ahandful of cases.By looking at the figures in the Appendix B(p.13-14) it can be seen that most ofspeakers altered the lengths of vowel evenfor the same sentence suggesting that theywere perhaps stressing the short vowel justa bit at times. Apart from this, the shortvowel comes right at the end of the firstword of the compound word suggesting thatword final lengthening may be playing itsrole, in elongating the last rhyme of theword, which in this case would be the shortvowel. However if the results of all thesentences are viewed, it can be seen thatthe duration on some instances (likesentence 4 and 6, Appendix B p. 13-14)remain small. This usually happened if thesecond word of the compound word startswith a consonant (sentence 5, Appendix B p.13-14, being a big exception). On thecontrary the duration of vowel altered greatlywhen the second word started with a vowel.Some speakers kept it smaller on someoccasions and elongated it a bit on otheroccasions. Another trend observed wasthat some speakers (usually AG and BA),keeping up with their natural pace, usuallypreferred to stop in between the two wordsthus elongating the short vowel just a bitwhile others didn’t. The others, wherepossible, carried on uttering the secondword resulting in co-articulation. Thevowels have to be different though as thiswasn’t observed in sentence 7 (Appendix Bp. 13-14) where most speakers preferred tobreak the compound after the first word.All in all, irrespective of variations, theduration of the vowel remained about halfthe size of a long vowel suggesting that itremained a short vowel (approximately 0.52in ratio to a long vowel). This further leadsto the conclusion that kasre Izafatrepresents a short vowel and in consonantvowel(CV) configuration (with ‘V’ being theshort vowel) can occur exceptionally at wordending in compound words.6. ReferencesBukhari, S. 1985. Phonology of UrduLanguage. Karachi, Pakistan.Hussain, S. 1997. Phonetic Correlates oflexical Stress in Urdu, Unpublished Ph.D.dissertation, Northwestern University, IL,USA.

125Kachru, Y. 1987. “Hindi-Urdu”, in The MajorLanguages of South Asia, The Middle Eastand Africa. Comrie, B (eds.). Routledge,London, UK.Khan, M. 1997. Urdu ka Sauti Kaida,Islamabad, Pakistan.Lehiste, I. 1970, Suprasegmentals.Cambridge: Massachusetts Institute ofTechnology Press, USA.Napoli, D. 1996. Linguistics an introduction.New York, USA.Pickett, J. 1999. The Acoustics of SpeechCommunication. Fundamentals, SpeechPerception Theory, And Technology.Massachusetts USA.

126Center for Research in Urdu Language Processing7. APPENDICIESAppendix A Transcription of sentences used for the experimentFollowing are the transcriptions of the sentences used in the experiment. The short vowel (//)and the long vowel used for normalization of vowel duration are shown in bold face.1. hme qid zm n zdi dli.(We got independence because of Quaid -e- Azam).2. us k red ml bhot did t.(His reaction was very extreme).3. hm sær kr næ mnr pkstn e.(We made a recreational trip to Minar-e-Pakistan).4. lhor ke mhur bo me ek b b dn hæ.Bagh-e-Jinnah is one of the famous gardens of Lahore).5. qvm mthd unis s pæntlis me kim hui.(United Nations was formed in 1945).6. dnb li un ko bizt bri kij de.(I request Your Honor to set him free).7. mude str mtz dij de.(Award me Sitara-e-Imtiaz).8. hme zb qbr se btn tahije.(We should save ourselves from the torments of the grave).9. un ki ftr prti me sb ko dvt m he.(There is an open invitation for their Iftar party).10. shb hest loo ko rib loo ki mdd krni thije.(Rich people should help the poor).

127Appendix B Results of the ExperimentBelow is the experiment result data. All values are durations in seconds except for thenormalization column, which is the ratio of short vowel to the long vowel.TABLE B.1 Sentence 1Speaker target vowel long vowel (azadi) normalisedRS 0.052 0.092 0.565AG 0.057 0.098 0.577BA 0.061 0.111 0.553AK 0.060 0.089 0.680SS 0.051 0.098 0.524GS 0.055 0.105 0.529Sentence Average 0.056 0.099 0.571TABLE B.2 Sentence 2Speaker target vowel long vowel (shadid) normalisedRS 0.056 0.094 0.594AG 0.072 0.115 0.627BA 0.059 0.114 0.523AK 0.053 0.103 0.513SS 0.046 0.092 0.506GS 0.065 0.125 0.519Sentence Average 0.058 0.107 0.547

128Center for Research in Urdu Language ProcessingTABLE B.3 Sentence 3Speaker target vowel long vowel (ser) normalisedRS 0.060 0.113 0.529AG 0.063 0.135 0.464BA 0.062 0.168 0.374AK 0.058 0.141 0.411SS 0.051 0.135 0.380GS 0.069 0.163 0.427Sentence Average 0.060 0.143 0.431TABLE B.4 Sentence 4Speaker target vowel long vowel (bagh) normalisedRS 0.069 0.166 0.418AG 0.079 0.161 0.493BA 0.075 0.186 0.397AK 0.081 0.176 0.463SS 0.081 0.183 0.449GS 0.090 0.200 0.453Sentence Average 0.079 0.179 0.445TABLE B.5 Sentence 5Speaker target vowel long vowel (-talis) normalisedRS 0.050 0.080 0.618AG 0.077 0.094 0.818BA 0.069 0.102 0.675AK 0.062 0.086 0.723SS 0.057 0.087 0.654GS 0.057 0.115 0.500Sentence Average 0.062 0.094 0.665

129TABLE B.6 Sentence 6Speaker target vowel long vowel (janab) normalisedRS 0.063 0.115 0.549AG 0.079 0.095 0.837BA 0.066 0.117 0.570AK 0.073 0.097 0.755SS 0.064 0.110 0.586GS 0.084 0.117 0.716Sentence Average 0.072 0.109 0.669TABLE B.7 Sentence 7Speaker target vowel long vowel (sitara-) normalisedRS 0.045 0.111 0.401AG 0.054 0.125 0.432BA 0.051 0.120 0.421AK 0.046 0.128 0.358SS 0.045 0.109 0.409GS 0.052 0.132 0.390Sentence Average 0.049 0.121 0.402TABLE B.8 Sentence 8Speaker target vowel long vowel (hamaen) normalisedRS 0.058 0.118 0.495AG 0.073 0.126 0.577BA 0.069 0.110 0.634AK 0.062 0.117 0.533SS 0.052 0.114 0.460GS 0.074 0.122 0.606Sentence Average 0.065 0.118 0.551

130Center for Research in Urdu Language ProcessingTABLE B.9 Sentence 9Speaker target vowel long vowel (iftar) normalisedRS 0.052 0.121 0.430AG 0.076 0.134 0.570BA 0.081 0.161 0.504AK 0.062 0.133 0.468SS 0.057 0.132 0.439GS 0.082 0.171 0.476Sentence Average 0.068 0.142 0.481TABLE B.10 Sentence 10Speaker target vowel long vowel (ko) normalisedRS 0.058 0.126 0.464AG 0.071 0.124 0.574BA 0.088 0.153 0.582AK 0.077 0.133 0.581SS 0.059 0.131 0.456GS 0.074 0.175 0.421Sentence Average 0.071 0.140 0.513

131Following are the distribution of normalized short vowel duration of each speaker withrespect to sentences:Speaker: RSNormalised vowelduration0.80.60.40.201 2 3 4 5 6 7 8 9 10SentencesIst Try2nd Try3rd Try4th Try5th TryFIGURE B.1: Variation of vowel duration of speaker RS across the ten sentences.Speaker: AG1Normalized vowelduration0.80.60.40.21st Try2nd Try3rd Try4th Try5th Try01 2 3 4 5 6 7 8 9 10SentencesFIGURE B.2: Variation of vowel duration of speaker AG across the ten sentences.Normalised VowelDuration0.90.80.70.60.50.40.30.20.10Speaker: BA1 2 3 4 5 6 7 8 9 10Sentences1st Try2nd Try3rd Try4th Try5th TryFIGURE B.3: Variation of vowel duration of speaker BA across the ten sentences.

132Center for Research in Urdu Language ProcessingNormalized VowelDuration0.90.80.70.60.50.40.30.20.10Speaker: AK1 2 3 4 5 6 7 8 9 10Sentences1st Try2nd Try3rd Try4th Try5th TryFIGURE B.4: Variation of vowel duration of speaker AK across the ten sentences.Normalized VowelDuration0.80.70.60.50.40.30.20.10Speaker: SS1st Try2nd Try3rd Try4th Try5th Try1 2 3 4 5 6 7 8 9 10SentencesFIGURE B.5: Variation of vowel duration of speaker SS across the ten sentences.Normalised VowelDuration0.90.80.70.60.50.40.30.20.10Speaker: GS1 2 3 4 5 6 7 8 9 10Sentences1st Try2nd Try3rd Try4th Try5th TryFIGURE B.6: Variation of vowel duration of speaker GS across the ten sentences.