vocal tract

大连外国语学院2006年攻读硕士学位研究生入学考试试题学科专业：英语语言文学、外国语言学及应用语言学（英语）考试科目：语言学（A卷）考生请注意：答案必须写在答题纸上，将标号写清楚。

写在试卷上不给分。

I. Mark the following statements with T if they are true or F if they false(20 points)1. The design features of human languages include only creativity and displacement.2. The bow-wow theory, the pooh-pooh theory and the “yo-he-ho” theory are well-established theory with adequate supportive evidences.3. Recreational function and metalingual function are the most important functions of language.4. Phonetics, phonology, morphology, syntax, and semantics are branches of linguistics.5. Macrolinguistics is of interdisciplinary nature.6. Articulatory phonetics, acoustic phonetics and auditory phonetics study speech sounds.7. Fictive is among the eleven places of articulation which are distinguished on the IPA chart.8. A physically definable unit, the common factor underlying a set of forms, and a grammatical unit are used to discuss the definition of “word”.9. Broadening, narrowing, meaning shift, class shift and semantic change contribute to change in sentence meaning.10. In Saussure’s view, language is a system of sings, each of which consists of two parts: SIGNIFIED (sound image) and SIGNIFER (concept).11. Syntax refers to the study of the rules governing the ways sounds are combined to form sentences in a language.12. There are generally three kinds of sense relations recognized－sameness relation, oppositeness relation and inclusiveness relation with the technical terms of synonymy, antonymy and hyponymy respectively.13. Pairs of words like buy-sell and lend-borrow belong to complementary antonymy.14. Cognitive psycholinguistics is concerned with making inferences about the content of the human mind.15. The theory of CONTEXT OF SITUATION can be summarized as follows: the relevant features of the participants, the relevant object and the effects of the verbal action.16. Austin claims that there are two types of sentences: performatives and constatives.17. The part of linguistics that studies the language of literature is termed literary stylistics.18. CAI aims at seeing educational problems on the part of the teacher, whereas CAL emphasizes the use of a computer in both teaching and learning in order to help the learner achieve educational objectives.19. Communicative Competence has two components: appropriateness and performance.20. An important contribution of the Prague School to linguistics is that it sees language in terms of FUCTION.II. Fill in each of the following blanks with an appropriate word with the hint of the initial letter (20points)1. A______ of language makes it potentially creative, and conventionality of language makes learning a language laborious. For learners of a foreign language , it is this feature of language that is more worth noticing than its arbitrariness.2. Speaking of the manners of articulation, p______ is complete closure of the articulators involved so that the airstream cannot escape through the mouth..3. V______ is made with the back of the tongue and the soft palate. An example in English is [k] as in “cat”.4. The principle of a______ creation can account for the co-existence of two forms, regular and irregular, in the conjugation of some English verbs.5. G______ is a type of control over the form of some words by other words in certain syntactic constructions.6. P______ logic is the study of the truth conditions for propositions: how the truth ofa composite proposition is determined by the truth of value of its constituent propositions and the connections between them.7. The principle of r______ is defined as every act of ostensive communication communicates the presumption of its own optimal relevance.8. Assonance describes s______ with a common vowel (cVc).9. The type of l______ constructed by second or foreign language learners who are still in the process of learning a language is often referred to as interlanguage.10. E______ is the grammatically incorrect form; mistake appears when the language is correct grammatically but improper in a communicational context.III. Mark the choice that best completes the statement (20points)1. The dog barks wow wow in English but wangwangwang in Chinese. This indicates the ______ relationship between the sound of a morpheme and its meaning.A. arbitraryB. dualC. creativeD. displaced2. The _______function of language is one of the most powerful uses of language because it is so crucial in changing the emotional status of an audience for or against someone or something.A. recreationalB. emotiveC. interpersonalD. performative3. ______ are produced by constricting or obstructing the vocal tract at some place to divert, impede, or completely shut off the flow of air in the oral cavity.A. V owelsB. Semi-vowelsC. ConsonantsD. Semi-consonants4. There are two possibilities of assimilation. If a following sound is influencing a preceding sound, we call it ____ assimilation.A. progressiveB. precedingC. precedingD. regressive5. The term ___ refers to those words that consist of more than one lexical morpheme, or the way to join two separation words to produce a single form.A. compoundB. derivationC. inventionD. blending6. ___ refers to a type of word-forming where a shorter word is derived by deleting an imagined affix from a longer form already in the language.A. BorrowingB. Back-formationC. AcronymD. Analogical creation7. ___, also known agreement, maybe defined as the requirement that the forms of two or more words in a syntactic relationship should agree with each other in terms of some categories.A. tenseB. aspectC. concordD. government8. The ___ construction is defined as a construction whose distribution is not functionally equivalent to any of its constituents, such as “on the shelf”.A. endocentricB. exocentricC. subordinateD. coordinate9. The theory of meaning which relates the meaning of a word to the thing it refers to, or stands for, is known as the ___ theory.A. integratedB. sense relationC. subordinateD. coordinate10. Ogden and Richards proposes a theory that the relation between a word and a thing it refers to is not direct, but it is mediated by ____.A. triangleB. conceptC. meaningD. referential11. ___ describes the additional ease with which a word is accessed due to its more frequent usage in the language.A. Cohort TheoryB. Frequency EffectC. Recency EffectD. Context Effect12. General context effect occurs when our ___ knowledge about the world influences language comprehension.A. generalB. specificC. discourseD. text13. Language may determine our thinking patterns. This is part of the ___.A. Sapir-Whorf HypothesesB. Cross-Cultural Communication theoryC. Context of situation theoryD. Ethnography of Communication theory14. ____ may be said to equivalent to speaker’s meaning, contextual meaning, or extra meaning.A. LocutionaryB. Perloculationary ActC. Illocutionary principleD. Performatives15. Which of the following are not the four categories of maxims introduced by Grice?A. quantityB. qualityC. relationD. obscurity16. Make your contribution necessary and say no more than you must. This is the ___ proposed by Horn.A. Q-principleB. R-principleC. Cooperative principleD. Q-based principle17. The person who tells the story may also be a character in the fictional world of the story, relating the story after the event. In this case, the critics call the narrator ___. A. third-person narrator B. I-narratorC. First narratorD. Indirect narrator18. ___ deals with the principles and practice of using a large body of machine-readable texts in language study.A. SociolinguisticsB. PsycholinguisticsC. PragmaticsD. Corpus linguistics19. Classified by different aims, there are four types of test. Which is not one of them?A. AptitudeB. Attitude testC. Proficiency testD. achievement test20. Chomsky believes that language is somewhat innate, and that children are born with what he calls a ___.DB.MTC. S→R R→SD. ESTⅣ. Match each in Column A with a term in Column B that best describes it (10 points)Ⅴ. Write out the scholar that is closely connected with the concept or theory (10 points)Ⅵ. Analyze two possible interpretations of the following ambiguous expressions by IC Analysis (10 points) Leave the boy at homeⅦ. Explain the following concepts or theories (20 points)1. Language and parole (4 points)2. Morpheme (4 points)3. Explain the three major strands of psycholinguistics research comprehension, production and acquisition (4 points)4. Stress (at word level) (4 points)5. Computational linguistics (4 points)Ⅷ. Essay questions (40 points)1. Illustrate arbitrariness of language with examples. (10 points)2. What is “compound”? Illustrate it with examples from English. (10 points)3. Illustrate Austin’s claim about the type of sentences “performtives”. (10 points)4. What are the contributions of sociolinguistics to language teaching? (10 points)。

语言学名词解释胡壮麟语言学名词解释总结1.Design feature: are features that define our human languages, such as arbitrariness, duality, creativity,displacement, cultural transmission, etc.2.Function: the use of language to communicate, to think, etc. Language functions include informativefunction, interpersonal function, performative function, emotive function, phatic communion,recreational function and metalingual function.3.Synchronic: a kind of description which takes a fixed instant (usually, but not necessarily, thepresent), as its point of observation. Most grammars are of this kind.4.Diachronic: study of a language is carried through the course of its history.5.Prescriptive: a kind of linguistic study in which things are prescribed how ought to be, i.e. layingdown rules for language use.6.Descriptive: a kind of linguistic study in which things are just described.7.Arbitrariness: one design feature of human language, which refers to the face that the forms oflinguistic signs bear no natural relationship to their meaning.8.Duality: one design feature of human language, which refers to the property of having two levels ofare composed of elements of the secondary level and each of the two levels has its own principles of organization.9.Displacement: one design feature of human language,which means human language enable theirusers to symbolize objects, events and concepts which are not present c in time and space, at the moment of communication.10.Phatic communion: one function of human language, which refers to the social interaction oflanguage.11.Metalanguage: certain kinds of linguistic signs or terms for the analysis and description of particularstudies./doc/1716061846.html,petence: language user’s underlying knowledge about the system of rules.13.Performance: the actual use of language in concrete situation./doc/1716061846.html,ngue: the linguistic competence of the speaker.15.Parole: the actual phenomena or data of linguistics (utterances).16.Phonetics: it studies how speech sounds are produced, transmitted, and perceived. (The study ofsounds which are used in linguistic communication is called phonetics.)17.Phonology: the study of the sound patterns and sound systems of languages. It aims to discover theprinciples that govern the way sounds are organized in languages, and to explain the variations that occur.18.Articulatory phonetics: the study of production of speech sounds.19.Phoneme: the abstract element of sound, identified as being distinctive in a particular language.20.Manner of articulation: in the production of consonants, manner of articulation refers to the actualrelationship between the articulators and thus the way in which the air passes through certain parts of the vocal tract.21.Place of articulation: in the production of consonants, place of articulation refers to where in thevocal tract there is approximation, narrowing, or the obstruction of air./doc/1716061846.html,plementary distribution: the relation between tow speech sounds that never occur in the sameenvironment. Allophones of the same phoneme are usually in complementary distribution.23.Suprasegmental: suprasegmental features are those aspects of speech that involve more than singlesound segments. The principal supra-segmental features are syllable, stress, tone, and intonation. 24.Morpheme: the smallest unit of language in terms of relationship between expression and content, aunit that cannot be divided into further small units without destroying or drastically altering themeaning, whether it is lexical or grammatical.25.Inflection: the manifestation of grammatical relationship through the addition of inflectional affixes,such as number, person, finiteness, aspect and case, which do not change the grammatical class of the stems to which they are attached.26.Affix: the collective term for the type of formative that can be used only when added to anothermorpheme (the root or stem).27.Derivation: different from compounds, derivation shows the relation between roots and affixes.28.Root: the base from of a word that cannot further be analyzed without total lass of identity.29.Stem: any morpheme or combination of morphemes to which an inflectional affix can be added.30.Bound morpheme: an element of meaning which is structurally dependent on the world it is added to,e.g. the plural morpheme in “dog’s”.31.Free morpheme: an element of meaning which takes the form of an independent word.32.Lexicon: a list of all the words in a language assigned to various lexical categories and provided withsemantic interpretation.33.Grammatical word: word expressing grammatical meanings, such conjunction, prepositions, articlesand pronouns.34.Lexical word: word having lexical meanings, that is, those which refer to substance, action andquality, such as nouns, verbs, adjectives, and verbs.35.Blending: a relatively complex form of compounding, in which two words are blended by joining theinitial part of the first word and the final part of the second word, or by joining the initial parts of the two words.36.Loanword: a process in which both form and meaning are borrowed with only a slight adaptation, insome cases, to eh phonological system of the new language that they enter.37.Loan blend: a process in which part of the form is native and part is borrowed, but the meaning isfully borrowed.38.Loan shift: a process in which the meaning is borrowed, but the form is native.39.Acronym: is made up of the first letters of the name of an organization, which has a heavily modifiedheadword.40.Loss: the disappearance of the very sound as a morpheme in the phonological system.41.Back-formation: an abnormal type of word-formation where a shorter word is derived by deleting animagined affix from a long form already in the language.42.Assimilation: the change of a sound as a result of the influence of an adjacent sound, which is morespecifically called. ”contact” or”contiguous” assimilation.43.Linguistic determinism: one of the two points in Sapir-Whorf hypothesis, i.e. language determinesthought.44.Linguistic relativity: one of the two points in Sapir-Whorf hypothesis, i.e. there’s no limit to thestructural diversity of languages.45.Performative: an utterance by which a speaker does something does something, as opposed to aconstative, by which makes a statement which may be true or false.46.Constative: an utterance by which a speaker expresses a proposition which may be true or false.47.Locutionary act: the act of saying something; it’s an act of conveying l iteral meaning by means ofsyntax, lexicon, and phonology. Namely, the utterance of a sentence with determinate sense andreference.48.Illocutionary act: the act performed in saying something; its force is identical with the speaker’sintention.49.Perlocutionary act: the act performed by or resulting from saying something, it’s the consequence of,or the change brought about by the utterance.50.Conversational implicature: the extra meaning not contained in the literal utterances,understandable to the listener only when he shares the speaker’s knowledge or knows why and how he violates intentionally one of the four maxims of the cooperative principle.。

Vocal Tract Normalization as Linear Transformation of MFCCMichael Pitz and Hermann NeyChair of Computer Science VI (Lehrstuhl f¨u r Informatik VI)RWTH Aachen –University of Technology52056Aachen,Germany{pitz,ney }@informatik.rwth-aachen.deAbstractWe have shown previously that vocal tract normalization (VTN)results in a linear transformation in the cepstral domain.In this paper we show that Mel-frequency warping can equally well be integrated into the framework of VTN as linear transformation on the cepstrum.We show examples of transformation matri-ces to obtain VTN warped Mel-frequency cepstral coefficients (VTN-MFCC)as linear transformation of the original MFCC and discuss the effect of Mel-frequency warping on the Jaco-bian determinant of the transformation matrix.Finally we show that there is a strong interdependence of VTN and Maximum Likelihood Linear Regression (MLLR)for the case of Gaussian emission probabilities.1.IntroductionV ocal tract normalization (VTN)tries to compensate for theeffect of speaker specific vocal tract lengths by warping the frequency axis of the power spectrum of the speech signal [1,2,3,4].The frequency axis is scaled by a warping func-tiong α:[0,π]→[0,π](1)ω→˜ω=g α(ω)and the warped spectrum is defined as |{X (ω)}|= ˜X (g α(ω))where the warping function g αis assumed to be invertible,i.e.strictly monotonic and continuous.The frequency ω=πcorresponds to the Nyquist frequency and the domain and co-domain are chosen to conserve bandwidth and information contained in the original spectrum.We have shown in [5,6]that in the framework of cepstral signal analysis VTN amounts to a linear transformation in the cepstral space for any arbitrary invertible warping function with domain and co-domain as given in Eq.(1).In that work we exemplary derived analytical solutions for the transformation matrices of piece-wise linear,quadratic,and bilinear warping functions.The warped cepstral coefficients ˜c n (α),n =1...N can be obtained by a linear transformation of the original cepstral coefficients c k ,k =1...K with a transformation matrix A (α)of dimension N ×K :This work was partially funded by the European Commission un-der the Human Language Technologies project CORETEX (IST-1999-11876),and by the DFG (Deutsche Forschungsgemeinschaft)under contract NE 572/4-1.A nk (α)=2s kππZ 0d ˜ωcos(˜ωn )cos(g (−1)α(˜ω)k )(2)withs k =(12:k =01:elseIn the case of continuous spectra,there may be no upper limit for N and K .We have assumed that the original spectrum can be represented by a finite number of cepstral coefficients,for instance if it has been cepstrally smoothed already.In prac-tice,however,we work with discrete spectra.Hence,N and K will be finite and equal to the number of spectral lines of the discrete Fourier spectrum.This number can be further reduced for cepstral smoothing.In the following we will show that VTN warped Mel-frequency cepstral coefficients (VTN-MFCC)can also be ob-tained by a linear transformation of either the original plain cepstral coefficients or the original MFCC for arbitrary invert-ible warping functions.We will exemplary discuss transforma-tion matrices obtained for a piece-wise linear warping function.Finally we will discuss a consequence of VTN being a linear transformation of the MFCC,namely a strong interdependence of VTN and Maximum Likelihood Linear Regression (MLLR).This interdependence can explain previous experimental results that improvements obtained by VTN and subsequent MLLR were not additive [7].2.Integration of Mel Frequency ScaleMel frequency warping is applied during signal analysis to ad-just the spectral resolution to the human ear [8]:f mel =2595·lg1+f700Hz .There are two possible ways to include Mel frequency warping into the framework of VTN as linear transformation:A.)to express the VTN-MFCC as a linear function of theoriginal,unwarped plain cepstral coefficients (CC)orB.)to express the VTN-MFCC as a linear function of theMFCC.In the following we will calculate the MFCC directly on the power spectrum as described in [9]rather than using a filter-bank.2.1.From Plain CC to VTN-MFCCWe have shown in[5,6]that a frequency warping of the spec-trum with an arbitrary invertible function results in a linear transformation of the cepstral coefficients.Mel frequency warp-ing can be considered as one special case of such a frequency warping and thus results in a linear transformation as well. Therefore the combination of VTN and subsequent Mel warp-ing still amount to a linear transformation in the cepstral do-main.VTN is typically applied before Mel scale warping;hence the combination of both warping steps becomesg mel(gα(ω)):ω→˜ωmel=B·lg1+gα(ω)·f s(3)where gα(ω)denotes the VTN warping function as before,f s denotes the sampling frequency,and B is defined asB=πlg1+f s2·700Hzto meet the requirement g mel(π)=π.Inserting Eq.(3)into Eq.(2)leads toA nk(α)=2s k ππZd˜ωmel cos(˜ωmel n)cosg(−1)αg(−1)mel(˜ωmel)k(4)Thus we can express the cepstral coefficients of the VTN-Mel-warped spectrum as linear transformation of the original,un-warped cepstral coefficients.2.2.From MFCC to VTN-MFCCWe will see in Section4that VTN is equivalent to a param-eterized constrained MLLR transformation.MLLR is a lin-ear transformation of model parameters(means and variances) which were typically been estimated from MFCC feature vec-tors.Thus more interesting and of practical relevance is to ex-press the VTN-Mel-warped cepstral coefficients as a function of the MFCC(i.e.without VTN)instead of the plain cepstral coefficients.The difficulty in the present case is that VTN is typically applied before Mel warping.We start with the defini-tion of the VTN-Mel-warped cepstral coefficients˜c mel n(α)˜c mel n(α)=s kππZd˜ωmel lnˆX(˜ωmel)cos(˜ωmel n).(5)VTN is usually applied to original,i.e.non-Mel-scaled,spec-trum(˜ωmel denotes the VTN-Mel-warped frequency)˜ωmel=g mel◦gα(ω)and the warped spectrum is given asnˆX(˜ωmel)o=nXg(−1)αg(−1)mel(˜ωmel)o=nX(ω)o.We now expand the spectrum as function of the Mel-warped frequencyωmel in terms of unnormalized(i.e.not VTN-warped) cepstral coefficients c mel kln|X(ω)|2=lnˆX(ωmel)2=2KXk=0c mel k cos(ωmel k).(6)As before,inserting Eq.(6)into Eq.(4)results in˜c mel n(α)=KXk=0c mel k2s kπZd˜ωmel cos(ωmel k)·cos(˜ωmel n)We now need to express the unnormalized Mel-scale frequencyωmel as function of the VTN-warped Mel-scale frequency˜ωmel:ωmel=g mel(ω)=g mel◦g(−1)α◦g(−1)mel(˜ωmel).Finally,we obtain˜c mel n(α)=KXk=0A mel nk(α)c mel kwithA mel nk(α)=2s kππZd˜ωmel cos(˜ωmel n)cos(g mel◦g(−1)α◦g(−1)mel(˜ωmel)k).(7)Hence,the cepstral coefficients˜c mel n(α)of the VTN-warpedMel-scale spectrum can be computed by a linear transforma-tion of the unnormalized cepstral coefficients c mel k(without VTNwarping).Because of the non-linear transformation the integralin Eq.(7)may hardly be solved analytically.Nevertheless,thetransformation matrix can be calculated numerically.We have calculated the transformation matrix numericallyfor a piece-wise linear warping function(dashed line in Fig.1)ω→˜ω=gα(ω)=8><>:αω:ω≤ω0αω0+π−αω0π−ω0(ω−ω0):ω>ω0(8)We choose the inflexion pointω0,where the slope of the warp-ing function changes,as follows:ω0=8<:78πα≤178·απα>1The resulting warping function g eff:=g mel◦g(−1)α◦g(−1)melreadsg eff(˜ωmel):=g mel◦g(−1)α◦g(−1)mel(˜ωmel)=8>>><>>>:B·log[1+1α(10˜ωmel/B−1)]:ωmel≤g mel(ω0)B·logh1+f s˜ω02·700Hz1α−π−α−1˜ωπ−˜ω0+π−α−1˜ω0π−˜ω0(10˜ωmel/B−1)i:ωmel>g mel(ω0)(9)and is shown in Fig.1(straight line).If we expand the effectivewarping function g eff(˜ωmel)forωmel≤g mel(ω0)in a Taylor se-ries aboutα=1(forωmel≥g mel(ω0)the expansion is similar)g eff(˜ωmel)=˜ωmel−B1−10−˜ωmel/Bln(10)(α−1)+O(α−1)2we see that the linear term dominates the expansion because theterm1−10−˜ωmel/Bln(10)is small for0≤˜ωmel≤π.Thus,g eff(˜ωmel)ωmel = g mel °g α°g (-1)mel (ωmel ) ωorig= g α(ωorig)πωω~πα>1α<1~~Figure 1:mel α(−1)mel tion of the Mel frequency ωmel (straight)in comparison to the warping function g αfor plain CC as function of the original frequency ωorig (dashed)for α=0.9and α=1.1can be approximated by a linear function with an appropriate choice of an effective warping factor αeff .The cepstral coefficients ˜c mel n (α)obtained by a linear trans-formation of MFCC with the matrix defined by Eq.(7)are iden-tical to those calculated by explicitly warping the spectrum dur-ing signal analysis as presented in [9].Transformation matrices for MFCC using piece-wise lin-ear warping with warping factors (α=0.9and α=1.1)are shown in Fig.2and 3.These matrices were calculated by solv-ing Eq.(7)numerically without approximations.Figure 2:Matrix for piecewise linear warping function,α=0.9,Mel scaleFigure 3:Matrix for piecewise linear warping function,α=1.1,Mel scaleIn order to study the effect of the Mel-frequency warping on the transformation matrix,we will compare these matrices with those calculated analytically for computing VTN-CC as linear transformation of the plain CC (Fig.4and 5).We observe that the matrices for the Mel scale are more diagonally domi-nant than those for the original paring the resultingwarping function ˜ωmel =g mel ◦g α◦g (−1)mel (ωmel )(straight linein Fig.1)as function of the Mel frequency with the warpingfunction g αfor plain CC (dashed line in Fig.1)as function ofthe original frequency,we see that g mel ◦g α◦g (−1)mel is much closer to identity than g α.Therefore the transformation matrix for MFCC is more diagonally dominant than those obtained for plain cepstral coefficients.The general structure of the warp-ing matrices as well as possible approximations are discussed in more detail in [6],also for bilinear and quadratic warping functions.Figure 4:Matrix for piecewise linear warping function,α=0.9,no Mel scaleFigure 5:Matrix for piecewise linear warping function,α=1.1,no Mel scale3.Discussion of Jacobian DeterminantTo estimate the unknown warping factor α,we proceed as fol-lows:For each speaker r we are given labeled training data (X r ,W r )where X r denote the sequence of acoustic data and W r the sequence of spoken words.In recognition,a preliminary hypothesis of the unknown word sequence W r can be obtained by a first recognition pass.Typically,we apply a maximum likelihood estimation of αˆαr =argmax αp (X r |W r ;θ,α)(10)=argmax αp 0(f α(X r )|W r ;θ0)· d f α(X r )dX r(11)In VTN the speaker normalization is usually not performed as atransformation of the acoustic vectors but by warping the power spectrum during signal analysis instead.Hence,the Jacobian determinant can hardly be calculated.In virtually all experi-mental studies the second term in Eq.(11),the Jacobian deter-minant,is neglected.Whether this is a good approximation or not will depend very much on how much the Jacobian determi-nant depends on α.Therefore it is good to study the second term as function of α.Expressing VTN as a matrix transforma-tion of the acoustic vector (x →A x )now enables us study the Jacobian determinant |det A |of the transformation.A plot showing the dependency of the Jacobian determi-nant on the warping factor αhas been computed numerically for piece-wise linear warping (Fig.6).The dependence of theJacobian determinant onαis weaker for MFCC(dashed line in Fig.6)because the effective warping function g mel◦gα◦g(−1)mel16cepstral coefficients as function ofα.Straight line:original frequency scale,dashed line:Mel frequency scale4.Interdependence of VTN and MLLR Most of today’s automatic speech recognition systems make use of Hidden Markov Models(HMM)with Gaussian emission probability distributionsN(x|µ,Σ)=1pdet(2πΣ)exp−12(x−µ)TΣ−1(x−µ).with state dependent parametersµandΣ.If the acoustic fea-ture vector(essentially the cepstral coefficients)x is normalized with the VTN matrix A,the Gaussian distribution changes tox→y=A x:N(x|µ,Σ)→N(y|µ,Σ)=N(x|A−1µ,A−1ΣA−1T)=N(x|ˆµ,ˆΣ)withˆµ=A−1µandˆΣ=A−1ΣA−1T.(12) Thus,a linear transformation of the observation vector x is equivalent to a linear transformation of the mean vectorµand an appropriate transformation of the covariance matrixΣ.The transformations in Eq.(12)describe a special case of MLLR which is called constrained MLLR[10,11](con-strained refers to the use of the same matrix A for the transfor-mation of the mean and variance).In[7],Uebel and Woodland have found experimentally that improvements obtained by constrained MLLR and VTN were not additive.As we have shown,VTN may be viewed as a special case of constrained MLLR adaptation with an restriction to only one adjustable parameter(the warping parameter)which determines the matrix elements.The experiments were based on a MF-PLP signal analysis.The difference between MFCC and MF-PLP is mainly caused by different types of smoothing. This difference is not expected to effect the equivalence of VTN and linear transformations.Hence,the experiments support the analytic result that VTN is a special case of constrained MLLR.5.ConclusionWe have shown in this work that VTN warped Mel-frequency cepstral coefficients(VTN-MFCC)can also be obtained by a linear transformation of either the original plain cepstral co-efficients or the original MFCC for arbitrary invertible warp-ing functions.The numerical values of VTN-MFCC computed with the presented approach were identical to those obtained by explicitly warping the spectrum during signal analysis.Ex-pressing VTN as matrix transformation of MFCC allows us to compute the Jacobian determinant of the transformation,which has typically been neglected so far.Finally,we have shown that for the case of Gaussian emission probabilities VTN and MLLR are strongly interdependent,which can explain previous experimental results that improvements obtained by VTN and subsequent constrained MLLR were not additive.6.References[1] E.Eide and H.Gish,“A parametric approach to vocal tractlength normalization,”in Proc.IEEE Int.Conf.on Acous-tics,Speech,and Signal Processing,vol.1,Atlanta,GA, May1996,pp.346–349.[2]L.Lee and R.Rose,“Speaker normalization using effi-cient frequency warping procedures,”in Proc.IEEE Int.Conf.on Acoustics,Speech,and Signal Processing,vol.1, Atlanta,GA,May1996,pp.353–356.[3]H.Wakita,“Normalization of vowels by vocal tract lengthand its application to vowel identification,”in Proc.IEEE Int.Conf.on Acoustics,Speech,and Signal Processing, vol.ASSP-25,No.2,Apr.1977,pp.183–192.[4]S.Wegmann, D.McAllaster,J.Orloff,and B.Pe-skin,“Speaker normalization on conversational telephone speech,”in Proc.IEEE Int.Conf.on Acoustics,Speech, and Signal Processing,vol.1,Atlanta,GA,May1996, pp.339–341.[5]M.Pitz,S.Molau,R.Schl¨u ter,and H.Ney,“V ocaltract normalization equals linear transformation in cep-stral space,”in Proc.ISCA Europ.Conf.on Speech Com-munication and Technology,vol.4,Aalborg,Denmark, Sept.2001,pp.2653–2656.[6]M.Pitz and H.Ney,“V ocal tract normalization equals lin-ear transformation in cepstral space,”submitted to IEEE Trans.on Speech and Audio Processing,2003.[7]L.F.Uebel and P.C.Woodland,“An investigation intovocal tract length normalisation,”in Proc.ISCA Europ.Conf.on Speech Communication and Technology,vol.6, Budapest,Hungary,Sept.1999,pp.2527–2530.[8]S.J.Young,HTK:Hidden Markov Model Toolkit V1.4.User Manual,Cambridge,England,Feb.1993.[9]S.Molau,M.Pitz,R.Schl¨u ter,and H.Ney,“Comput-ing mel-frequency cepstral coefficients on the power spec-trum,”in Proc.IEEE Int.Conf.on Acoustics,Speech,and Signal Processing,vol.1,Salt Lake City,UT,June2001, pp.73–76.[10]V.Digalakis,D.Rtischev,and L.Neumeyer,“Speakeradaptation using constrained estimation of gaussian mix-tures,”IEEE Trans.on Speech and Audio Processing, vol.3,no.5,pp.357–366,Sept.1995.[11]M.J.F.Gales,“Maximum likelihood linear transfor-mations for HMM-based speech recognition,”Computer Speech and Language,vol.12,no.2,pp.75–98,Apr.1998.。

胡壮麟《语言学教程》术语表第一章phonology音系学grammar语法学morphology形态学syntax句法学lexicology词汇学general linguistics普通语言学theoretical linguistics理论语言学historical linguistics历史语言学descriptive linguistics描写语言学empirical linguistics经验语言学dialectology方言学anthropology人类学stylistics文体学signifier能指signified所指morphs形素morphotactics语素结构学/形态配列学syntactic categories句法范畴syntactic classes句法类别序列sub-structure低层结构super-structure上层结构open syllable开音节closed syllable闭音节checked syllable成阻音节rank 等级level层次ding-dong theory/nativistic theory本能论sing-song theory唱歌说yo-he-ho theory劳动喊声说pooh-pooh theory感叹说ta-ta theory模仿说animal cry theory/bow-wow theory模声说Prague school布拉格学派Bilateral opposition双边对立Mutilateral opposition多边对立Proportional opposition部分对立Isolated opposition孤立对立Private opposition表缺对立Graded opposition渐次对立Equipollent opposition均等对立Neutralizable opposition可中立对立Constant opposition恒定对立Systemic-functional grammar系统功能语法Meaning potential意义潜势Conversational implicature会话含义Deictics指示词Presupposition预设Speech acts言语行为Discourse analysis话语分析Contetualism语境论Phatic communion寒暄交谈Metalanguage原语言Applied linguistics应用语言学Nominalism唯名学派Psychosomatics身学第二章trachea/windpipe气管tip舌尖blade舌叶/舌面front舌前部center舌中部top舌顶back舌后部dorsum舌背root舌跟pharynx喉/咽腔laryngeals喉音laryngealization喉化音vocal cords声带vocal tract声腔initiator启动部分pulmonic airstream mechanism肺气流机制glottalic airstream mechanism喉气流机制velaric airstream mechanism腭气流机制Adam’s apple喉结Voiceless sound清音Voiceless consonant请辅音Voiced sound浊音Voiced consonant浊辅音Glottal stop喉塞音Breath state呼吸状态Voice state带音状态Whisper state耳语状态Closed state封闭状态Alveolar ride齿龈隆骨Dorsum舌背Ejective呼气音Glottalised stop喉塞音Impossive内爆破音Click/ingressive吸气音Segmental phonology音段音系学Segmental phonemes音段音位Suprasegmental超音段Non-segmental非音段Plurisegmental复音段Synthetic language综合型语言Diacritic mark附加符号Broad transcription宽式标音Narrow transcription窄式标音Orthoepy正音法Orthography正字法Etymology词源Active articulator积极发音器官Movable speech organ能动发音器官Passive articulator消极发音器官Immovable speech organ不能动发音器官Lateral边音Approximant [j,w]无摩擦延续音Resonant共鸣音Central approximant中央无摩擦延续音Lateral approximant边无摩擦延续音Unilateral consonant单边辅音Bilateral consonant双边辅音Non-lateral非边音Trill [r]颤音trilled consonant颤辅音rolled consonant滚辅音Labal-velar唇化软腭音Interdental齿间音Post-dental后齿音Apico-alveolar舌尖齿龈音Dorso-alveolar舌背齿龈音Palato-alveolar后齿龈音Palato-alveolar腭齿龈音Dorso-palatal舌背腭音Pre-palatal前腭音Post-palatal后腭音Velarization软腭音化Voicing浊音化Devoicing清音化Pure vowel纯元音Diphthong二合元音Triphthong三合元音Diphthongization二合元音化Monophthongization单元音化Centring diphthong央二合元音Closing diphthong闭二合元音Narrow diphthong窄二合元音Wide diphthong宽二合元音Phonetic similarity语音相似性Free variant自由变体Free variation自由变异Contiguous assimilation临近同化Juxtapostional assimilation邻接同化Regressive assimilation逆同化Anticipatory assimilation先行同化Progressive assimilation顺同化Reciprocal assimilation互相同化Coalescent assimilation融合同化Partial assimilation部分同化Epenthesis插音Primary stress主重音Secondary stress次重音Weak stress弱重音Stress group重音群Sentence stress句子重音Contrastive stress对比重音Lexical stress词汇重音Word stress词重音Lexical tone词汇声调Nuclear tone核心声调Tonetics声调学Intonation contour语调升降曲线Tone units声调单位Intonology语调学Multilevel phonology多层次音系学Monosyllabic word多音节词Polysyllabic word单音节次Maximal onset principle最大节首辅音原则第三章词汇liaison连音contracted form缩写形式frequency count词频统计a unit of vocabulary词汇单位a lexical item词条a lexeme词位hierarchy层次性lexicogrammar词汇语法morpheme语素nonomorphemic words单语素词polymorphemic words多语素词relative uninterruptibility相对连续性a minimum free form最小自由形式the maximum free form最大自由形式variable words 可变词invariable words不变词paradigm聚合体grammatical words(function words)语法词/功能词lexical words(content words)词汇词/实义词closed-class words封闭类词opened-class words开放类词word class词类particles小品词pro-form代词形式pro-adjective(so)代形容词pro-verb(do/did)代副词pro-adverb(so)代动词pro-locative(there)代处所词/代方位词determiners限定词predeterminers前置限定词central determiners中置限定词post determiners后置限定词ordinal number序数词cardinal number基数词morpheme词素morphology形态学free morpheme自由词素bound morpheme黏着词素root词根affix词缀stem词干root morpheme词根语素prefix前缀infix中缀suffix后缀bound root morpheme黏着词根词素inflectional affix屈折词缀derivational affix派生词缀inflectional morphemes屈折语素derivational morphemes派生语素word-formation构词compound复合词endocentric compound向心复合词exocentric compound离心复合词nominal endocentric compound名词性向心复合词adjective endocentric compound形容词性向心复合词verbal compound动词性复合词synthetic compound综合性复合词derivation派生词morpheme语素phoneme音位morphonology形态语音学morphophomemics形态音位学morphemic structure语素结构phonological structure音素结构monosyllabic单音节polysyllabic多音节phonological conditioned音位的限制morphological conditioned形态的限制coinage/invention新创词语blending混成法abbreviation缩写法acronym首字母缩写法back-formation逆序造次/逆构词法analogical creation类比构词法borrowing借词法loanword借词loanblend混合借词loanshift转移借词loan translation翻译借词loss脱落addition添加metathesis换位assimilation同化contact assimilation接触性同化contiguous assimilation临近性同化theory of least effort省力理论non- contiguous assimilation非临近性同化distant assimilation远距离同化morpho-syntactic change形态-句法变化morphological change形态变化syntactical change句法变化finite element有定成分semantic change语义变化multisemous多种意义broadening词义扩大narrowing词义缩小meaning shift词义转移class shift词性变换folk etymology俗词源orthographic change拼写的变化conversion变换/变码domain范围/领域meaning shift意义转移split infinitives分裂不定式(She was told to regularly classes)calque仿造词语clipping截断法metanalysis再分化finiteness定式proximate(this)近指代词obviative(that)远指代词non-productivity/unproductive非多产性semiotics符号学paradigmatic relations聚合关系associative relations联想关系syntagmatic relations组合关系sequential relations序列关系logogram语标register语域passive vocabulary消极词汇lexis/vocabulary词汇表第四章句法number数gender性case格nominative主格vocative呼格accusative兵格genitive属格dative与格ablative离格tense 时aspect体perfective完成体imperfective未完成体concord/agreement一致关系/协同关系government支配关系the governor支配者the governed被支配者signified能指signifier所指syntagmatic relationship组合关系paradigmatic relationship聚合关系associative relationship联想关系animate noun有生名词the two axes两根坐标坐标轴immediate constituent analysis(ICanalysis for short)直接成分分析法linear structure线性结构hierarchical structure层级结构construction结构体constituent成分substituability替换性labeled tree diagram标签树形图endocentric/headed construction向心结构/中心结构exocentric construction离心结构subordinate construction主从结构coordinate construction并列结构recapitulation再现the declarative陈述句the interrogative疑问句dative movement与格移位morph-phonemic rule形态音位规则constituent morphemes成分规则affix hopping词缀越位nominalization名物化object-deletion宾语删除subject-deletion主语删除categories语类lexicon词库temporal subject表时间的主语syntactic limitation句法限制standard theory标准理论trace theory语迹理论the same index带同标志government管辖binding约束a rule system规则系统a principle system原则系统constituent command(C-command forshort)成分统制plain English普通英语anaphor照应语pronominal指代语r-expression(referential-expression)指称语INFL(inflection)形态变化reciprocals(each other)相互代词accessible subject可及主语local domain局部语域binding domain约束语域logophoricity主人公视角CS(computational system)计算系统Merger合并move移动theme主位rheme述位empty subject空主语objective order客观顺序subjective order主观顺序actual sentence division实义句子切分法functional sentence perspective 功能句子观communicative dynamism (CD)交际动力bipartition二分法tripartite classification三分法representative function表达功能expressive function表情功能appellative/vocative function称呼功能conative function意欲功能poetic function诗学功能ideational function概念功能interpersonal function人际功能textual function语篇功能transitivity及物性actor动作者mood system语气系统the finite verbal operator限定部分residue剩余部分indicative直陈语气imperative祈使语气mental-process(a process of sensing)心理过程（感觉过程）relational process(a process of being)关系过程（属性过程）verbal process(a process of saying)言语过程（讲话过程）existential process生存过程第四章句法number数gender性case格nominative主格vocative呼格accusative兵格genitive属格dative与格ablative离格tense 时aspect体perfective完成体imperfective未完成体concord/agreement一致关系/协同关系government支配关系the governor支配者the governed被支配者signified能指signifier所指syntagmatic relationship组合关系paradigmatic relationship聚合关系associative relationship联想关系animate noun有生名词the two axes两根坐标坐标轴immediate constituent analysis(ICanalysis for short)直接成分分析法linear structure线性结构hierarchical structure层级结构construction结构体constituent成分substituability替换性labeled tree diagram标签树形图endocentric/headed construction向心结构/中心结构exocentric construction离心结构subordinate construction主从结构coordinate construction并列结构recapitulation再现the declarative陈述句the interrogative疑问句dative movement与格移位morph-phonemic rule形态音位规则constituent morphemes成分规则affix hopping词缀越位nominalization名物化object-deletion宾语删除subject-deletion主语删除categories语类lexicon词库temporal subject表时间的主语syntactic limitation句法限制standard theory标准理论trace theory语迹理论the same index带同标志government管辖binding约束a rule system规则系统a principle system原则系统constituent command(C-command for short)成分统制plain English普通英语anaphor照应语pronominal指代语r-expression(referential-expression)指称语INFL(inflection)形态变化reciprocals(each other)相互代词accessible subject可及主语local domain局部语域binding domain约束语域logophoricity主人公视角CS(computational system)计算系统=derivational procedure推导系统Merger合并move移动theme主位rheme述位empty subject空主语objective order客观顺序subjective order主观顺序actual sentence division实义句子切分法functional sentence perspective 功能句子观communicative dynamism (CD)交际动力bipartition二分法tripartite classification三分法representative function表达功能expressive function表情功能appellative/vocative function称呼功能conative function意欲功能poetic function诗学功能ideational function概念功能interpersonal function人际功能textual function语篇功能transitivity及物性actor动作者mood system语气系统the finite verbal operator限定部分residue剩余部分indicative直陈语气imperative祈使语气mental-process(a process of sensing)心理过程（感觉过程）relational process(a process of being)关系过程（属性过程）verbal process(a process of saying)言语过程（讲话过程）existential process生存过程empiricism经验主义（洛克，白板说）rationalism 理性主义（笛卡尔）mentalism心灵主义new empiricism新经验主义(Bloomfield)priori先天综合判断（康德Kant）Cartesian linguistics笛卡尔语言学派Syntactic structure (SS)早期转换句法时期Standard theory (ST)标准理论时期Extended Standard theory (EST)扩展的标准理论Revised Standard theory(REST)扩展的休正标准理论The theory of government and binding(GB theory)管辖和约束理论时期（管约论）Minimalist program (MP)最简方案时期Structural description结构描写式Performance system应用系统Modular theory模块理论Spell-out拼写Language faculty语言机制/官能Mental organ心智器官Knowledge of language 语言知识Meaning potential 意义潜势Context culture 文化语境Field语场Tenor语旨Mode语式pivot words轴心词mental construct心理构念theoretical cognitive psychology理论认知心理学psychological faculty心理官能autosyn/autogram/autoknow语法自主(arbitrariness任意性,systemacity系统性, self-containedness自足性)typological functionalism类型学功能主义extreme functionalism极端的功能主义external functionalism外部功能主义integrative functionalism一体化功能注主义exceptional case marking例外格标记specifier标定成分fall-category maximal projection全语类的最大投射two-segment category两节语类complement domain补足语区域minimal domain最小区域internal domain内部区域checking domain检验区域sisterhood姐妹关系minimizing chain link最小语链联结representational system表达系统strict cyclic principle严格的层级条件structure-preserving principle结构保存原则C-commanding condition成分统领条件articulatory-perceptual system发音-听音系统conceptual-intentional system概念-意旨系统interface conditions中介条件full-interpretation完全解释原则procrastination逻辑形式操作优先原则greed句法操作自利原则the shortest linkage principle最短联接原则the shortest movement principle最短移位原则primarycomplement/modifier(referential NP)一级补语位/修饰语位(定指名词短语)secondary complement(non- referentialNP) 二级补语位(非定指名词短语)empty category principle空范畴原则aspect checking特征验证aspect feature基本体貌特征ASPP is functional projection .ASPP是功能投射.crossing branch交叉分支across the board extraction抽取跨界移动principles-and-parameters framework原则与参数语法head parameter中心语参数logical form(LF)逻辑形式phonetic form(PF)语音形式spell-out拼读phonological component音韵部分overt component显性部分covert component隐性部分core computation核心运算asymmetric c-command不对称成分统制linear correspondence axiom线形对应定理adjunction加接determiner限定词concatenate联结linearization线性化functional parameterization hypothesis功能参数设定假设right-branching右向分支X’(V,N,A,P)词项X’’=XP=Xmax是X的二阶投射结构Y’’=指示语specifierZ’’=补述语complementIP=屈折短语inflection phraseXP=general phrase structureC HL人类语言的运算系统=computational system for humanlanguageLCA线性对应定理=linearcorrespondence axiomXmin=X0=最小投射。