Phonetic Transcription And Progress In Speech Synthesis

Phonetic transcription (or phonetic notation) is the visual representation of speech sounds (or phones). The most common type of phonetic transcription uses a phonetic alphabet (e.g., the International Phonetic Alphabet.

The pronunciation of words in many languages, as distinct from their written form (orthography), has undergone significant change over time. Pronunciation can also vary greatly among dialects of a language. Traditional orthography in some languages, particularly French and English, often differs from the pronunciation. For example, the words “bough” and “trough” do not rhyme in English, even though their spellings might suggest they do. In French, for example, the ‘s’ at the end of words is usually silent (“militaire” is pronounced the same as “militaires”) unless followed by a word beginning in a vowel. In the orthography of most European languages, the fact that many letters are pronounced or silent depending on contexts causes difficulties in determining the appropriate pronunciation, especially in the cases of English, Irish, and French. However, in other languages, such as Spanish and Italian, there is a more consistent-though still imperfect-relationship between orthography and pronunciation.

Therefore, phonetic transcription can provide a function that orthography cannot. It displays a one-to-one relationship between symbols and sounds, unlike the traditional Roman alphabet. Phonetic transcription allows us to step outside of orthography and examine differences in pronunciation between dialects within a given language, as well as to identify changes in pronunciation that may take place over time.

Phonetic transcription may aim to transcribe the phonology of a language, or it may wish to go further and specify the precise phonetic realisation. In all systems of transcription we may therefore distinguish between broad transcription and narrow transcription. Broad transcription indicates only the more noticeable phonetic features of an utterance, whereas narrow transcription encodes more information about the phonetic variations of the specific allophones in the utterance. The difference between broad and narrow is a continuum. One particular form of a broad transcription is a phonemic transcription, which disregards all allophonic difference, and, as the name implies, is not really a phonetic transcription at all, but a representation of phonemic structure.

]; the broad, phonemic transcription, placed between slashes, indicates merely that the word ends with phoneme /l/, but the narrow, allophonic transcription, placed between square brackets, indicates that this final /l/ is dark (velarized).

The advantage of the narrow transcription is that it can help learners to get exactly the right sound, and allows linguists to make detailed analyses of language variation. The disadvantage is that a narrow transcription is rarely representative of all speakers of a language. Most Americans and Australians would pronounce the /t/ of little as a tap [ɾ]. Many people in England would say /t/ as [ʔ] (a glottal stop) and/or the second /l/ as [w] or something similar. A further disadvantage in less technical contexts is that narrow transcription involves a larger number of symbols which may be unfamiliar to non-specialists.

The advantage of the broad transcription is that it usually allows statements to be made which apply across a more diverse language community. It is thus more appropriate for the pronunciation data in foreign language dictionaries, which may discuss phonetic details in the preface but rarely give them for each entry. A rule of thumb in many linguistics contexts is therefore to use a narrow transcription when it is necessary for the point being made, but a broad transcription whenever possible.

The International Phonetic Alphabet (IPA) is one of the most popular and well-known phonetic alphabets. It was originally created by primarily British language teachers, with later efforts from European phoneticians and linguists. It has changed from its earlier intention as a tool of foreign language pedagogy to a practical alphabet of linguists. It is currently becoming the most often seen alphabet in the field of phonetics.

Most American dictionaries for native English-speakers-American Heritage Dictionary of the English Language, Random House Dictionary of the English Language, Webster’s Third New International Dictionary-employ respelling systems based on the English alphabet, with diacritical marks over the vowels and stress marks.

Another commonly encountered alphabetic tradition is the Americanist phonetic alphabet, originally created for the transcription of Native American and European languages. There exist somewhat similar traditions used by linguists of Indic, Finno-Ugric, Caucasian, and Slavic languages. The difference between these alphabets and IPA is small, although often the specially created characters of the IPA are often abandoned in favour of already existing characters with diacritics (e.g. many characters are borrowed from Eastern European orthographies).

There are also extended versions of the IPA, for example: extIPA, VoQs, and Luciano Canepari’s.

The IPA is not the only phonetic transcription system in use. The other common Latin-based system is the Americanist phonetic notation, devised for representing American languages, but used by some US linguists as an alternative to the IPA. There are also sets of symbols specific to Slavic, Indic, Finno-Ugric, and Caucasian linguistics, as well as other regional specialties. The differences between these alphabets and IPA are relatively small, although often the special characters of the IPA are abandoned in favour of diacritics or digraphs.

Other alphabets, such as Hangul, may have their own phonetic extensions. There also exist featural phonetic transcription systems, such as Alexander Melville Bell’s Visible Speech and its derivatives.

The International Phonetic Association recommends that a phonetic transcription should be enclosed in square brackets “[ ]”. A transcription that specifically denotes only phonological contrasts may be enclosed in slashes “/ /” instead. If one is in doubt, it is best to use brackets, for by setting off a transcription with slashes one makes a theoretical claim that every symbol within is phonemically contrastive for the language being transcribed.

Phonetic transcriptions try to objectively capture the actual pronunciation of a word, whereas phonemic transcriptions are model-dependent. For example, in The Sound Pattern of English, Noam Chomsky and Morris Halle transcribed the English word night phonemically as /nixt/. In this model, the phoneme /x/ is never realized as [x], but shows its presence by “lengthening” the preceding vowel. The preceding vowel in this case is the phoneme /i/, which is pronounced [aɪ] when “long”. So phonemic /nixt/ is equivalent to phonetic but underlying this analysis is the belief that historical sounds such as the gh in night may remain in a word long after they have ceased to be pronounced, or that a phoneme may exist in a language without ever being directly expressed. (This was later rejected by both Chomsky and Halle.)

For phonetic transcriptions, there is flexibility in how closely sounds may be transcribed. A transcription that gives only a basic idea of the sounds of a language in the broadest terms is called a broad transcription; in some cases this may be equivalent to a phonemic transcription (only without any theoretical claims). A close transcription, indicating precise details of the sounds, is called a narrow transcription. These are not binary choices, but the ends of a continuum, with many possibilities in between. All are enclosed in brackets.

Here every symbol represents an unambiguous speech sound, but without going into any unnecessary detail. None of these transcriptions make any claims about the phonemic status of the sounds. Instead, they represent certain ways in which it is possible to produce the sounds that make up the word.

There are also several possibilities in how to transcribe this word phonemically, but here the differences are generally not of precision, but of analysis. The special symbol for English r is not used, for it is not meaningful to distinguish it from a rolled r. The differences in the letter e reflect claims as to what the essential difference is between the vowels of pretzel and pray; there are half a dozen ideas in the literature as to what this may be. The second transcription claims that there are two vowels in the word, even if they can’t both be heard, while the first claims there is only one.

However, phonemic transcriptions may also be broad or narrow, or perhaps it would be better to say abstract vs. concrete. They may show a fair amount of phonetic detail, usually of a phoneme’s most common allophone, but because they are abstract symbols they do not need to resemble any sound at all directly. Phonemic symbols will frequently be chosen to avoid diacritics as much as possible, under a ‘one sound one symbol’ policy, or may even be restricted to the ASCII symbols of a typical keyboard. For example, the English word church may be transcribed as in church. A close approximation of its actual pronunciation, or more abstractly as /crc/, which is easier to type. Phonemic symbols should always be explained, especially when they are as divergent from actual / Occasionally a transcription will be enclosed in pipes (“| |”). This goes beyond phonology into morphological analysis. For example, the words pets and beds could be transcribed phonetically as and (in a fairly narrow transcription), and phonemically as /pets/ and /bedz/. Because /s/ and /z/ are separate phonemes in English, they receive separate symbols in the phonemic analysis. However, you probably recognize that underneath this, they represent the same plural ending. This can be indicated with the pipe notation. If you believe the plural ending is essentially an s, as English spelling would suggest, the words can be transcribed |pets| and |beds|. If, as most linguists would probably suggest, it is essentially a z, these would be |petz| and |bedz|.

To avoid confusion with IPA symbols, it may be desirable to specify when native orthography is being used, so that, for example, the English word jet is not read as “yet”. This is done with angle brackets or chevrons: ‹jet›. It is also common to italicize such words, but the chevrons indicate specifically that they are in the original language’s orthography, and not in English transliteration.

Symbol and Sounds:

The International Phonetic Alphabet is based on the Latin alphabet, using as few non-Latin forms as possible.The Association created the IPA so that the sound values of most consonants taken from the Latin alphabet would correspond to “international usage”. Hence, the letters ‹b›, ‹d›, ‹f›, (hard) ‹ɡ›, (non-silent) ‹h›, (unaspirated) ‹k›, ‹l›, ‹m›, ‹n›, (unaspirated) ‹p›, (voiceless) ‹s›, (unaspirated) ‹t›, ‹v›, ‹w›, and ‹z› have the values used in English; and the vowels from the Latin alphabet (‹a›, ‹e›, ‹i›, ‹o›, ‹u›) correspond to the sound values of Latin: [i] is like the vowel in machine, [u] is as in rule, etc. Other letters may differ from English, but are used with these values in other European languages, such as ‹j›, ‹r›, and ‹y›.

This inventory was extended by using capital or cursive forms, diacritics, and rotation. There are also several derived or taken from the Greek alphabet, though the sound values may differ. For example, ‹ʋ› is a vowel in Greek, but an only indirectly related consonant in the IPA. Two of these (‹θ› and ‹χ›) are used unmodified in form; for others (including ‹β›, ‹ɣ›, ‹ɛ›, ‹ɸ›, and ‹ʋ›) subtly different glyph shapes have been devised, which may be encoded in Unicode separately from their “parent” letters.

The sound values of modified Latin letters can often be derived from those of the original letters. For example, letters with a rightward-facing hook at the bottom represent retroflex consonants; and small capital letters usually represent uvular consonants. Apart from the fact that certain kinds of modification to the shape of a letter generally correspond to certain kinds of modification to the sound represented, there is no way to deduce the sound represented by a symbol from the shape of the symbol (unlike, for example, in Visible Speech).

Beyond the letters themselves, there are a variety of secondary symbols which aid in transcription. Diacritic marks can be combined with IPA letters to transcribe modified phonetic values or secondary articulations. There are also special symbols for suprasegmental features such as stress and tone that are often employed.

Letter Forms:

The symbols chosen for the IPA are meant to harmonize with the Latin alphabet.For this reason, most symbols are either Latin or Greek letters, or modifications thereof. However, there are symbols that are neither: for example, the symbol denoting the glottal stop, ‹ʔ›, has the form of a “gelded” question mark, and was originally an apostrophe. In fact, there are a few symbols, such as that of the voiced pharyngeal fricative, ‹ʕ›, which, though modified to blend with the Latin alphabet, were inspired by glyphs in other writing systems (in this case, the Arabic letter `ain)

Despite its preference for letters that harmonize with the Latin alphabet, the International Phonetic Association has occasionally admitted symbols that do not have this property. For example, before 1989, the IPA symbols for click consonants were ‹ʘ›, ‹ʇ›, ‹Ê-›, and ‹Ê-›, all of which were derived either from existing symbols, or from Latin and Greek letters. However, except for ‹ʘ›, none of these symbols was widely used among Khoisanists or Bantuists, and as a result they were replaced by the more widespread symbols ‹ʘ›, ‹ǀ›, ‹ǃ›, ‹ǂ›, and ‹ǁ› at the IPA Kiel Convention in 1989.

Some of the new symbols were ordinary Roman letters typeset “turned” (= upside-down) (e.g. ÊŽ É¥ É™ É” ɹ á´š), which was easily done before mechanical typesetting machines came into use.

Usage:

Although the IPA offers over a hundred symbols for transcribing speech, it is not necessary to use all relevant symbols at the same time; it is possible to transcribe speech with various levels of precision. A precise phonetic transcription, in which sounds are described in a great deal of detail, is known as a narrow transcription. A coarser transcription which ignores some of this detail is called a broad transcription. Both are relative terms, and both are generally enclosed in square brackets. Broad phonetic transcriptions may restrict themselves to easily heard details, or only to details that are relevant to the discussion at hand, and may differ little if at all from phonemic transcriptions, but they make no theoretical claim that all the distinctions transcribed are necessarily meaningful in the language.

Phonetic transcriptions of the word international in two English dialects. The square brackets indicate that the differences between these dialects are not necessarily sufficient to distinguish different words in English.

For example, the English word little may be transcribed broadly using the IPA as [ˈlɪtəl], and this broad (imprecise) transcription is an accurate (approximately correct) description of many pronunciations.

It is customary to use simpler letters, without a lot of diacritics, in phonemic transcriptions. The choice of IPA letters may reflect the theoretical claims of the author, or merely be a convenience for typesetting. For instance, in English, either the vowel of pick or the vowel of peak may be transcribed as /i/ (for the pairs /pik, piːk/ or /pɪk, pik/), and neither is identical to the vowel of the French word pique which is also generally transcribed /i/. That is, letters between slashes do not have absolute values, something true of broader phonetic approximations as well. A narrow transcription may, however, be used to distinguish them: [pʰɪk], [pʰiːk], [pik].