1 / 61

Spoken Language Processing

Spoken Language Processing. Spring,2007 Youngah Do. Ch.2. Spoken Language Structure. Contents. 2.2 PHONETICS AND PHONOLOGY 2.2.1 Phonemes 2.2.2 The Allophone: Sound and Context 2.2.3 Speech Rate and Coarticulation 2.3 SYLLABLES AND WORDS 2.3.1 Syllable 2.3.2 Words

vincentfrank
Télécharger la présentation

Spoken Language Processing

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Spoken Language Processing Spring,2007 Youngah Do

  2. Ch.2. Spoken Language Structure

  3. Contents 2.2 PHONETICS AND PHONOLOGY 2.2.1 Phonemes 2.2.2 The Allophone: Sound and Context 2.2.3 Speech Rate and Coarticulation 2.3 SYLLABLES AND WORDS 2.3.1 Syllable 2.3.2 Words 2.4 SYNTAX AND SEMANTICS 2.4.1 Syntactic Constituents 2.4.2 Semantic Roles 2.4.3 Lexical Semantics 2.4.4 Logical Form

  4. 2.2 PHONETICS AND PHONOLOGY • Phonetics is the study of speech sounds and their production, classification, and transcription. • Phonology is the study of the distribution and patterning of speech sounds in a language and of the tacit rules governing pronunciation.

  5. 2.2 PHONETICS AND PHONOLOGY2.2.1 Phonemes • Relation between a sign and the object (Saussere) –arbitrariness • signified Signifier • Arbitrary relation between the pronunciation and the meaning cup

  6. 2.2 PHONETICS AND PHONOLOGY2.2.1 Phonemes • Phoneme is used to denote any of the minimal units of speech sound in a language that can serve to distinguish one word from another. • Underlying, abstract and lexical representation • Phone is used to denote a phoneme’s acoustic realization. • Surface, phonetic and allophonic representation For example, English phoneme /t/ have two very different acoustic realization in the words satand meter

  7. 2.2 PHONETICS AND PHONOLOGY2.2.1 Phonemes • Table 2.4 shows a complete list of phonemes used in American English.

  8. 2.2 PHONETICS AND PHONOLOGY2.2.1 Phonemes

  9. 2.2 PHONETICS AND PHONOLOGY2.2.1 Phonemes2.2.1.1 Vowels ① • The primary energy entering the pharyngeal and oral cavities in vowel production vibrates at the fundamental frequency. • The major resonances of the oral and pharyngeal cavities for vowels are called F1 and F2. • F1 and F2 are determined by tongue placement and oral tract shape in vowels. • In general, F1 correspond to the back or pharyngeal portion of the cavity, while F2 is determined more by the size and shape of the oral portion, forward of the major tongue extrusion.

  10. 2.2 PHONETICS AND PHONOLOGY 2.2.1 Phonemes 2.2.1.1 Vowels ② • Table 2.5 and Figure 2.17 show typical values of F1 and F2 of American English vowels.

  11. 2.2 PHONETICS AND PHONOLOGY 2.2.1 Phonemes 2.2.1.1 Vowels ③ • The characteristic F1 and F2 values for vowels are sometimes called formant targets, which are ideal locations for perception two main factors that prevent speakers from attain ideal target ① speed of speech or other limitation on performance ② fast working of articulation in production of diphthong

  12. Typical diphthong of American English

  13. 2.2 PHONETICS AND PHONOLOGY 2.2.1 Phonemes 2.2.1.1 Vowels ④ • The major articulator for English vowels is the middle to rear portion of the tongue. The linguistically important dimensions of movement are generally the ranges [front <->back] and [high<->low]. • Figure 2.18 shows a schematic characterization of English vowels in terms of relative tongue positions. • Two kinds of vowels : • Those in which tongue height is represented as a point • Those in which it is represented as a vector

  14. 2.2 PHONETICS AND PHONOLOGY 2.2.1 Phonemes 2.2.1.1 Vowels ④

  15. 2.2 PHONETICS AND PHONOLOGY 2.2.1 Phonemes 2.2.1.1 Vowels ⑤ • The most important secondary vowel mechanism for English and many other languages is lip rounding. • Linguistic study of phonetic abstractions, called phonology, has largely converged on the five binary features : • Controversy on features that cannot be represented by binary mechanism and nonexistence of [front] feature • Generally agreed phonological features : +/- high, +/- low, +/- front, +/- back, and +/- round, plus the phonetically ambiguous but phonologically useful feature tense • Table 2.7. shows the feature decomposition of basic English vowels.

  16. 2.2 PHONETICS AND PHONOLOGY

  17. 2.2 PHONETICS AND PHONOLOGY 2.2.1 Phonemes 2.2.1.2 Consonants ① • Consonants are characterized by significant constriction or obstruction in the pharyngeal and/or oral cavities. • Many consonants occurs in pairs. • /s/: [+Voiced] vs. /z/:[-Voiced] • Manners of articulation refers to the articulation mechanism of a consonant.

  18. 2.2 PHONETICS AND PHONOLOGY • Table 2.8 shows the major distinctions in manner of articulation.

  19. 2.2 PHONETICS AND PHONOLOGY 2.2.1 Phonemes 2.2.1.2 Consonants ② • The English phones include • Consonants that typically have voicing without complete obstruction and narrowing of the vocal tract : semivowels (approximants) • Approximants: the tongue approaches the top of the oral cavity, but does not completely contact so as to obstruct the air flow. • /l, r/ : liquid group • /y, w/: glide group (cf. Pre-vocalic glides that share the syllable-initial position with another consonant are sometimes called on-glides.)

  20. 2.2 PHONETICS AND PHONOLOGY • Phonemes that have continuous voicing : sonorant • Liquid, glide and vowels e.g.)Liquid : vowel-like, sometimes syllabic or entirely as vowels in certain positions.

  21. 2.2 PHONETICS AND PHONOLOGY 2.2.1 Phonemes 2.2.1.2 Consonants ③ • Table 2.19 shows some semivowels.

  22. 2.2 PHONETICS AND PHONOLOGY • Obstruent stop, or plosive consonant : a consonant that involves complete blockage of the oral cavity • Nasal sound : not only has significant constriction but has continuous voicing

  23. 2.2 PHONETICS AND PHONOLOGY 2.2.1 Phonemes 2.2.1.2 Consonants ④ • Table 2.9 shows the complete consonant inventory of English.

  24. 2.2 PHONETICS AND PHONOLOGY 2.2.1 Phonemes 2.2.1.2 Consonants ⑤ • Pairs of stops : /b // p/, /d// t/, and /g/ /k/ - [+voiced] / [-voiced] (there can be voicing for the early sections of stop) • Voiceless plosive consonants exhibit a characteristic aperiodic burst of energy at the point of closure as shown in Figure 2.20 just prior to /i/.

  25. 2.2 PHONETICS AND PHONOLOGY

  26. 2.2 PHONETICS AND PHONOLOGY • Sounds that are complex combinations of manners of articulation • Affricates : stop + fricative • In English, affricates are [+/-voiced ]pairs.: /j/ and /ch/

  27. 2.2 PHONETICS AND PHONOLOGY 2.2.1 Phonemes 2.2.1.2 Consonants ⑥ • /m/, /n/, and /ng/ : voiced nasal consonants • The difference lies in the location of the major constriction along the top of the oral cavity. • As shown in Figure 2.1, the combination of articulator and place of articulation gives each consonant its characteristic sound:

  28. 2.2 PHONETICS AND PHONOLOGY 2.2.1 Phonemes 2.2.1.2 Consonants ⑦ • Labial consonants have their major constriction at the lips. • Dental or labio-dental have their major constriction by teeth. • Alveolar consonants bring the front part of the tongue into contact or approximation to the alveolar ridge. • Palatal consonants have approximation or constriction on or near the roof of the mouth. • Velar consonants bring the articulator, up to the rearmost top area of the oral cavity. • Table 2.20 shows the descriptive inventory of English consonants, arranged by manner (row), place (columns), and voiceless/voiced (pairs in cells).

  29. 2.2 PHONETICS AND PHONOLOGY

  30. 2.2 PHONETICS AND PHONOLOGY 2.2.1 Phonemes 2.2.1.3 Phonetic Typology • In Japanese, the existence of minimally-contrasting pairs is taken as conclusive evidence that length is phonemically distinctive. • kado (corner) vs. kaado (card) • In Spanish, trill is distinctive • pero (but) vs. perro (dog) • What is linguistically distinctive in any one language is generally present as a less meaningful signaling dimension in other languages.

  31. 2.2 PHONETICS AND PHONOLOGY • The primary dimension lacking in English that is exploited by a large subset of the word’s languages is pitch variation. • An example of systematic use of pitch contrasts. • Though English does not make systematic use of pitch in its inventory of word contrasts, pitch is systematically varied in English to signal a speaker’s emotions, intentions, and attitudes.

  32. 2.2 PHONETICS AND PHONOLOGY 2.2.2 The Allophone: Sound and Context ① • The process by which neighboring sounds influence one another is called coarticulation. • When the variations resulting from coarticulatory processes can be consciously perceived, the modified phonemes are called allophones. • Allophonic differences are always categorical.

  33. 2.2 PHONETICS AND PHONOLOGY 2.2.2 The Allophone: Sound and Context ② • Experiments on allophonic variations • [l] in like: light or clear /l/ • [l] in kill : dark /l/ • [p] in pin : aspiration. • [p] in /sp/ :loses its aspiration. • /t/ or /d/: may or may not have flapping. • Bitter, batter, murder, quarter

  34. 2.2 PHONETICS AND PHONOLOGY 2.2.3 Speech Rate and Coarticulation ① Other variations in speech are gradient. Individual speakers may vary their rates according to the content and setting of their speech, and there may be great inter-speaker differences as well. • At the faster rates, formant targets are less likely to be fully achieved. In addition, individual allophones may merge.

  35. 2.2 PHONETICS AND PHONOLOGY 2.2.3 Speech Rate and Coarticulation ② • There are also coarticulatory influences in the spectral appearance of speech sound which can only be understood at the level of spectral analysis. • Figure 2.23. shows the different relative spreads of F1 and F2 following the initial stop consonants.

  36. 2.2 PHONETICS AND PHONOLOGY 2.2.3 Speech Rate and Coarticulation ③ • The coarticulatory effect anticipation is demonstrated in Figure 2.24.; i.e., in the latter part of the vowel the articulators are moving to prepare for the upcoming consonant articulation.

  37. 2.3 SYLLABLES AND WORDS 2.3.1 Syllables ① • Syllables is an intermediate unit for parsing a word. • To completely parse a word into syllables requires making judgments of consonant affiliation (with the syllable peak vowels). -articulatory or perceptual criteria ? -how they can be rigorously applied? unsolved questions.

  38. 2.3 SYLLABLES AND WORDS 2.3.1 Syllables ② • Analysis on syllable • Centers : peaks in sonority • peaks have affiliated shoulders of strictly non-increasing sonority • As long as the sonority conditions are met, the exact affiliation of a given consonant that could theoretically affiliate on either side can be ambiguis.

  39. 2.3 SYLLABLES AND WORDS • Syllable structure of strengths /s t r eh nx th s/. • Nucleus : vowel peak, surrounded by the other sounds in characteristic positions. • Onset : initial consonants if any, and the rime is the nucleus with trailing consonants. • Coda : consonants in the rime following the nucleus.

  40. 2.3 SYLLABLES AND WORDS 2.3.2 Words • In spoken language, there is a segmentation problem • Only a few true pauses may be present • E.g. never give all the heart, for love nevergiveallltheheart forlove • Certain facts about word structure and combinatorial possibilities • groups of related words that seem intuitively similar along some dimension of for or meaning.  (paradigmatic relations) • behavior and distribution when combined for communicative purposes in fully functioning utterances.  (syntagmatic relations)

  41. 2.3 SYLLABLES AND WORDS 2.3.2 Words 2.3.2.1 Lexical Part-of-Speech ① • Lexical part-of-speech (POS) posits a restricted inventory of word-type categories, which capture generalizations of word forms and distributions. • A typical set of POS categories would include noun, verb, adjective, adverb, interjection, conjunction, determiner, preposition, and pronoun. • Created new words had best fall within the predictable paradigmatic and syntagmatic patterns of use summarized by the existing POS generalizations. Open-class words are sometimes called content words, for their referential properties.

  42. 2.3 SYLLABLES AND WORDS 2.3.2 Words 2.3.2.1 Lexical Part-of-Speech ② These open POS categories These closed POS(functionwords) categories

  43. 2.3 SYLLABLES AND WORDS 2.3.2 Words 2.3.2.1 Lexical Part-of-Speech ③ • The set of POS categories can be extended indefinitely. • Penn Treebank project in Table 2.14. incorporate morphological attributes of words into the POS label system.

  44. 2.3 SYLLABLES AND WORDS • many algorithms to automatically tag input sentences into a set of tags. · Rule-based method · Hidden Markov models (Ch.8.) · Machine-learning method.

  45. 2.3 SYLLABLES AND WORDS 2.3.2 Words 2.3.2.2 Morphology ① • English mainly uses prefixes and suffixes to express inflection and derivational morphology. • Inflectional morphology : variations in word form that reflect the contextual situation of a word in phrase or sentence syntax, and that rarely have direct effect on interpretation of the fundamental meaning expressed by the word. In English, common paradigm types include the verbal set of affixes: -s,-ed,-ing; the noun set: -s; and the adjectival –er, -est.

  46. 2.3 SYLLABLES AND WORDS 2.3.2 Words 2.3.2.2 Morphology ② • Derivational morphology: a given root word may serve as the source for wholly new words, often with POS changes as illustrated in Table 2.15. For example, the terms racial and racist, though presumably based on a single root word race, have different POS possibilities (adjective vs. noun-adjective) and meanings. • Generally, word formation operates in layers, according to a kind of word syntax • (deriv-prefix)*root(root)* (deriv-suffix)*(infl-suffix).

  47. 2.3 SYLLABLES AND WORDS 2.3.2 Words 2.3.2.3 Word Classes • With improved computational resources, grouping words according to their actual behavior can be possible. • Word classification according to the similarity of usage with repect to their word neighbors. • E.g. {mother wife father son husband brother daughter sister boss uncle} • Word classification which can identify semantic relatedness across traditional POS categories. • {problems problem solution solve analyzed solved solving}

  48. 2.4 SYNTAX AND SEMANTICS 2.4.1 Syntactic Constituents • Syntax • the patterns of formation sentences and phrases from words • By the concept of syntactic constituents , a sentence can be divided into its grammatical subparts as constrained by common grammatical patterns. • the rules for the formation of grammatical sentences. • Semantics • meaning, including the ways meaning is structured • changes in meaning and form over time.

  49. 2.4 SYNTAX AND SEMANTICS 2.4.1 Syntactic Constituents 2.4.1.1. Phrase Schemata ① • The goal is to create a simple, uniform template that is independent of POS category • arbitrary category, say category X ( which could be a noun N or a verb V). The generalized rule for a phrase XP XP =>(modifiers) X-head (post-modifiers) X is the head, since it dominates the configuration and names the phrase. XP, the culminating phrase node, is called a maximal projection of category X. the whole structure an X-template.

  50. 2.4 SYNTAX AND SEMANTICS 2.4.1 Syntactic Constituents 2.4.1.1. Phrase Schemata ② • NP (det) (modifier) head-noun (post-modifier)

More Related