Resume Phonology

 

Phoneme Inventories

 

   Examination of the IPA (International Phonetic Alphabet) reveals great diversity in the types of sounds found in languages of the world. Sounds are differentiated along various dimensions, including place of articulation, manner of articulation, laryngeal setting, airstream mechanism, and timing of articulatory gestures. In this chapter presents some of the salient cross-linguistic patterns identified in a number of cross-linguistic surveys of phoneme inventories, including Maddieson’s pioneering genetically balanced survey of  137 languages in Patterns of Sounds. the online version  of its expanded 451-language counterpart UCLA Phonological Segment Inventory Database (UPSID), several chapters of World Atlas of Language Structure, and the PHOIBLE database, which contains segment inventories  of 1,672 languages, of which those in UPSID constitute a subset. the discussion in this chapter centers on phonemes, sounds that are used contrastively to differentiate words.

 

3.1 Cross-linguistic distribution of phonemes

In discussing the typology of phonemes, it is common to impose a broad bifurcation between consonants and vowels, where consonants involve a tighter constriction in the vocal tract than vowels. Consonants differ widely in the location and degree of the constriction ranging from those produced with a slight narrowing at the lips, i.e. bilabial approximants, to those associated with a complete closure at the larynx, i.e. glottal stops. In addition, other properties such a laryngeal setting (e.g. voiced vs. voiceless vs. ejective), nasalization, secondary articulations (e.g. labialization, palatalization, pharyngealization), and relative timing of gestures.  Maddieson’s (1984) survey of phoneme inventories in 317 languages reveals a wide range in the number of phonemes found in languages of the world from a low of 11 in the East Papuan language Rotokas (six consonants and five vowels)and  in the Mura language Pirahã (eight consonants and three vowels) to a high of 141 in the Khoisan language !Xũ. Maddieson (1984) finds no tendency for a compensatory relationship between the number of vowels and the number of consonants in a language such that more vowels implies fewer consonants and vice versa.

  

3.2.      Consonant

According  to Maddieson’s the common consonant  there are 20 which conflates the dental vs alveolar. But according to reset proven by Maddienson  there are 24 language contrast dental and alveolar place of articulation. The most ‘representative’ of consonant there are five such as /z/, /ts/, /x/, /v/, dj/.

3.2.1.    Plosive

Is the common for language to contrast unaffricated oral stops. There are three places of articulation are ( billabial, denti-alveolar and velar ). Fourth place if the affricates included is ( palato-alveolar ). Most of language possess of unaspiratates. Voiceless stop and voice stop. After voiceless unaspirates and voiced stops, he common laryngeal setting for stops are voiceless unaspirates and voiced less aspirated (28.7 %), ejective (16.4 % ), and implosive ( 11.0%). Between the voiceless stops, dental and alveolar  ( 6.0%  ) which comtrast dental and alveolar,   ( 89.3 % )velar, and ( 82.9% ) and voiced stops velars ( 55.2 % ) relative to both billabial ( 62.8 % ), dental alveolar ( 61,5 % ).

3.2.2. Fricative

According to the picture was proven that there are two kind of fricative voiceless counterpart only for billabial pair ( β / θ ) and the  non-sibilant dental pair  ( δ / θ ). The modal of fricative in language is two the most common fricatives being dental/alveolar /s/ followed by / ʃ / ( 46.1 ) and f (42.6 % ) according to the  Maddieson’s survey  all of the Australian continent 15 lack fricative  with the 298 languages only additional sic case of fricative less language.

3.2.3. Nasals, liquids, and Non-liquid approximants (glides)

For the first step it discuss about Nasals. The vast majority of t he world’s nasals are voiced. Virtually all languages contrast nasal s at two (31 .9%), three (30. 0%), or four (26. 2%) places of articulation with the two most common nasal s being a dental/alveolar one (found in 95. 3% of languages) and a bilabial nasal (found in94 .3 %). The next most common nasal is a velar one (found in 53 .0% of languages) followed by a palatal or palate - alveolar one (39. 4%). And now about liquids, Most languages have one (23. 3%), two (41. 0%), or three (14 .5 %) liquids (laterals and rhotics), in languages with two liquids, it is most common to have one lateral and one rhotic (83. 1%of languages wit h two liquids) with two lateral (13. 8%) and two rhotic systems (2. 3%) being rare. In languages with three liquids, it is slightly more common to have two laterals and one rhotic (50. 0% of languages with three liquids) than one lateral and two rhotics (37. 0%), with three liquid systems consisting entirely of laterals being much sparser (13.0 %).  Most laterals are plain voiced approximants (74 .7% of laterals) with most of these occurring in the dental/alveolar region (86 6%). And other non-liquid approximants such as labial- palatals or velars are quite rare, each occurring in fewer than %of languages in the survey, although it is likely that many of the sounds described as voiced non-sibilant fricatives are, in fact, voiced approximants

 

VOWELS

Vowels belonging to different subcategories within these three height and backness groups are collapsed. For example, the high front unrounded vowels include both high and lower high vowels, i.e. /i, ɪ/, the mid front rounded vowels comprise both /e, ɛ/, and the low vowels include low vowels of different backness, height, and rounding specifications, i.e. /a, ɑ, ᴂ, ɐ, ɒ/. Both short and long vowels are included since height often co-varies with length. Secondary features such as nasalization and voice quality are not included.

The five most common vowels are (taking the cardinal vowel symbol as the prototype for each category) /a, e, o, i, u/. In contrast, relatively few languages contrast multiple degrees of height for high or low vowels. There is a considerable drop-off in frequency after /a, e, o, i, u/ to the next most common vowel /ə/, which is followed in turn by /ɨ/, /ɯ/, /y/, /ʌ/, /ø/, /ɵ/, and /ʉ/.

 

 

Phonemic Lenght

In the -language WALS sample, there is a bias for phonemic length in vowels over consonants: a total of  languages could be reliably identified as contrasting length for one or more vowel qualities morpheme-internally, while only  were described as contrasting length tautomorphemically for one or more consonants.

There is considerable variation between languages in how much short segments outnumber their long counterparts, but the clear trend is for a strong statistical bias in favor of short phonemes. The paucity of long exemplars is not merely due to the long segments constituting a subset of the short segments, since in most languages, either all or virtually all of the short sounds have phonemic long counterparts.

It should be noted that the number of languages that make length distinctions for consonants would increase considerably, however, if geminates arising across morpheme boundaries were also considered, e.g. English mundaneness, cattail. It should also be noted that length distinctions co-vary with qualitative distinctions in some languages, potentially making the source of certain vowel distinctions problematic to classify. For example, the tense high and mid vowels /i, u, e, o/ of English are phonetically longer than their lax counterparts /ɪ, ʊ, ɛ, ɔ/ (Peterson and Lehiste ).

The greater statistical discrepancy between short and long consonants (relative to short vs. long vowels) is attributed in part to distributional restrictions holding of geminates that do not apply to single consonants.

Language-internal frequency data fail to consistently line up with the sonority- sensitive continuum along the x-axis in Figure ., an inconsistency that is perhaps not surprising given the existence of exceptions even on a categorical level. The relative frequency of geminate voiceless stops compared to geminate voiceless fricatives is thus mixed: in Japanese and Hausa, long voiceless stops are more frequent than long voiceless fricatives, whereas the opposite pattern obtains in Finnish, Koasati, and Italian. Similarly, geminate sonorants are more common than geminate voiceless stops in Finnish and Hausa, whereas the opposite trend is observed in Japanese and Italian

 

3.5       Explaining the typology of phoneme inventories

There is an extensive literature devoted to explaining cross-linguistic biases in the distribution of phonemes. Most of this research proposes explanations that are rooted in considerations of speech production and/or perception, although accounts differ in whether they appeal directly to phonetic factors or indirectly through the medium of phonological features. (Adaptive) Dispersion Theory Targeting vowels as a case study, Liljencrants and Lindblom  is the first typologically informed attempt to quantify the phonetic forces claimed to condition phoneme inventories. Liljencrants and Lindblom hypothesize that phoneme inventories are preferable to the extent they possess contrasts that are maximally distinct in the perceptual domain. Their account, commonly termed Dispersion Theory (or Adaptive Dispersion Theory), is intuitively appealing since it fits with the observation that five vowel inventories characteristically consist of the wellspaced set /i, e, a, o, u/ rather than other hypothetical inventories making less use of the vowel space, e.g. /i, ɪ, e, ɛ, a/ or /i, y, u, ʊ, ʉ/.

Compares the inventories predicted by the Liljencrants and Lindblom model with the most common vowel inventories comprising from three to seven vowel qualities according to the  UPSID database (see Schwartz et al a for similar results based on the language original survey by Maddiesone). Searches were conducted for vowel inventories possessing the targeted number of vowel qualities, filtering out distinctions based on length and limiting the search to monophthongs without any secondary constrictions (e.g. frication, pharyngealization, retroflexion), laryngeal modifications (laryngealization, breathy voicing, devoicing), or nasalization.

The fit between the Liljencrants and Lindblom model and the most common three-vowel inventory is perfect. Their model also generates the most common four-vowel inventory. In the case of the five-vowel system, the mid back vowel that is most common cross-linguistically corresponds to a lower unrounded vowel in the Liljencrants and Lindblom simulation. The most common central vowel in languages of the world is schwa whereas the Liljencrants and Lindblom simulation predicts a higher central vowel, /ʉ/ for the six-vowel. System and both /ɨ/ and a high central /ʉ/ or high front /y/ in the case of the sevenvowel system. Furthermore, in predicting four high vowels /i, y or ʉ, ɨ, u/ and only two central vowels /ɛ, ɔ/ for the seven-vowel system, the Liljencrants and Lindblom model diverges sharply from the cross-linguistically dominant pattern of two high /i, u/ and four central /e, ɛ, o, ɔ / vowels in seven-vowel inventories.

Dispersion Focalization Theory Drawing on results of an analysis of vowel inventories in Maddieson’s original  language survey (Schwartz et al. a) Schwartz et al propose a revised model for predicting vowel inventories, the Dispersion Focalization Theory. They retain the original insight of Liljencrants and Lindblom’s Dispersion Theory according to which inventories containing perceptually dispersed vowels are preferred, but they introduce certain changes to their model in order to provide a better fit to attested patterns.

The first element in their model, the vowel inventory, crucially includes a notion of focalization, which incorporates a boost to the quantal vowels, i.e. vowels with two formants in close proximity (Stevens see section, including the three corner vowels /u/, /a/ (both with proximate first and second formants), and /i/ (close third and fourth formants). The second component contributing to the aggregate energy of a vowel system in Dispersion Focalization Theory captures the overall auditory dispersion of the vowels in a system. Dispersion is a function of first formant values and an integration of the second, third, and fourth formants, where formant values are expressed in Bark. In their dispersion function, Schwartz et introduce a variable that allows for an increased weighting of first formant values (the acoustic correlate of height), capturing the fact that larger vowel systems, i.e. those consisting of peripheral vowels beyond just /i, e, a, o, u/, overwhelmingly tend to fractionate the vertical rather than the horizontal space to produce more height than backness contrasts.

Becker’s  enormous survey of formant patterns for vowels in  languages has dispelled certain fallacies suggested by typological surveys based on impressionistic transcriptions. For example, he finds no support for the purported distinction in the height of the back vowel between two of the most common four-vowel systems /i, e, a, o/ and /i, e, a, u/ Rather, the back vowel in both systems tends to be intermediate in height between canonical /o/ and canonical /u/. Along similar lines, Becker observes that the distinction between systems with a single central vowel that is high, i.e. /ɨ/, vs. those in which the central vowel is mid, i.e. /ə/, is not confirmed acoustically; instead, the vowel in question is intermediate in height between the two central vowels, i.e. IPA /ɘ/.

One typological observation that has proven elusive to implement in a model incorporating dispersion and focalization is the preference for schwa over all vowels other than /i, e, a, o, u/. Schwartz et al concede that another nonperceptual factor, namely ease of articulation, is likely important in predicting the popularity of schwa. In fact, as they observe in their companion typological survey, Schwartz et al note that schwa is typically simply added as an additional non-peripheral vowel without interacting with the spacing (at least in an impressionistically salient way) of the peripheral vowels. This observation suggests that perceptual distance is not the only factor guiding the construction of vowel inventories; otherwise, one might expect to see an avoidance of mid vowels, or possibly low central vowels, in languages with schwa.

The role of articulatory ease in shaping vowel inventories also appears to be evident in languages with so-called “vertical” vowel systems, e.g. Abkhaz (Hewitt 1979, Vaux and Psiypa 1997a), Kabardian (e.g. Turchaninov and Tsagov 1940, Abitov et al. 1957, Catford 1948, Choi 1991, Colarusso 1992, Gordon and Applebaum 2006) and Marshallese (Choi 1992), in which the entire inventory of two or three vowels is central. Vaux and Samuels (2015) provide a comprehensive critique of dispersion theory, which they demonstrate is not equipped to handle the full range of typological variationn in vowel systems.

3.5.1.3 Aticulatory complexity and perceptual saturation

Lindblom and meddieson (1988) propose a model of consonant inventory construction incorporating maximization of perceptual distinctness and minimization of articulatory effort. They suggest that features can be broken down in to three groups according to their articulatory complexity. Lindblom and Maddieson test the predictions of their model by dividing the obstruent inventories for the language. Results indicate a strong cross-linguistic tendency for languages to possess the eleven basic obstruents before introducing obstruents belonging to the elaborated articulations. Similarly, complex articulations tend to come into play only after extensive exploitation of elaborated consonants, typically in consonant inventories of greater than 30 consonants. Results of Lindblom and Maddieson’s study cccomplement the work of Liljencrants and Lindblom (1972) and Schwartz et al on vowel inventories by offering support for the role of both articulatory and perceptual factors in the shaping of consonant inventories. An important issue left unresolved in Lindblom and Maddieson’s work, however, is how to quantify the distinction between basic articulations and their more complex counterparts.

3.5.1.4 Quantal theory

Steven’s Quantal Theory (19722, 1989) provided phonetic grounding for the still widely adopted articulatory-based feature set orginally proposed by Chomsky and Halle (1968) . Stevens proposes that phonological features define regions of acoustic and perceptual stability in which changes along a continous articulatory dimension result in relatively little change in the acoustic output.

In vowel systems, the quantal vowels are /i/, /u/, and /a/ since they occupy stable articulatory regions where minor shifts in tongue position result in only negligible acoustic and perceptual changes. A further virtue of the quantal vowels that is incorporated in to the Dispersion Focalization.

Quantal Theory has not been developed as estensively as Dispersion Theory in various incarnations. Evidence suggest, though, that it has some of the same shortcomings related to its failure to incurporate a notion of articulatory ease. The prevalence of schwa and the existence of vertical vowel systems are thus problematic for quantal theory.

3.5.1.5 Feature enhancement

            that features can be divided into two groups, a primary and a secondary group. The primary features include the manner features [sonorant] and [continuant] and the place feature [coronal], all of which can be implemented independently of other features. This differs from  secondary features, which may be restricted in their distribution as a function of the specification of primary features also associated with that sound. For example, only coronal consonants have the possibility of being contrasted in terms of the feature [distributed], which encodes the breadth of a consonant constriction in the front–back domain.

 

A further difference between primary and secondary features is that a change in the specification of a primary feature results in a more salient acoustic and thus auditory response than a change in a secondary feature.

The typology further supports a distinction between [continuant] and [distributed] in their salience.

 

Stevens and Keyser’s theory offers an account for why certain types of sounds are more common than others cross-linguistically. For example, sonorants are overwhelmingly voiced because the primary feature [+sonorant] ideally combines with the secondary enhancing feature [+voice].

The consonants that result from the optimal combinations of primary and secondary features, /j, w, s, f, h, n, l, m, t, p, k/, are all typologically favored.

 

3.5.1.6 Feature economy

Another common feature of phoneme inventories that was mentioned earlier in the context of vowel systems is symmetry Clements (2003, 2009) provides an explicit formalization of the principles that lead to the formation of symmetrical inventories. According to his theory of feature economy, which takes as a starting point long-standing observations about the structure of sound systems (de Groot 1931, 1948, Martinet 1955), languages prefer inventories that make maximal use of the minimum number of phonological features to expand their phoneme inventories.

 

Clements tests the cross-linguistic validity of feature economy through case studies of certain combinations of sounds based on the 451-language UPSID database (Maddieson and Precoda (1990)

In particular, he tests two predictions made by the theory of feature economy. The first of these, Mutual Attraction, predicts that sounds will occur more frequently if all of their features are present in other sounds in the same language. For example, a voiced labial fricative is predicted to be more common in inventories that already contain another labial sound, another fricative, and another voiced sound, since adding a voiced labial

fricative boosts the economy index of the language by exploiting features that are independently employed in the language.

A companion prediction of Clements’s theory of Feature Economy is that sounds will be less likely to occur if one or more of its features are not distinctively used elsewhere in the language. This effect of Avoidance of Isolated Sounds works against a plosive inventory like the one in Chickasaw (Table 3.5 ), which contains a single voiced stop, the only segment for which voicing is contrastive in Chickasaw. For example, /b/ is less likely to occur in a language without both /d/ and /g/ than in a language with at least one of the two. This means that Chickasaw is typologically unusual in having only a single voiced stop.

Voiced fricatives require a delicate articulatory balancing act for aerodynamic reasons. It is difficult to simultaneously sustain voicing in the face of the pressure build-up behind a fricative constriction while also generating sufficient airflow through the constriction to make the fricative turbulence audible. Clements (2009) hypothesizes that intralanguage frequency plays a decisive role in determining markedness, such that sounds that are less frequent in a language are more marked than others. Two other factors to which Clements (2009) appeals in his theory are Robustness and Enhancement. Robustness entails the existence of a hierarchy of features ordered in terms of their phonetic salience.

            The work by Stevens (1972) and Stevens and Keyser (1989) discussed in 3.5.1.4 potentially serves as the backbone for an explicit metric of Robustness, though Clements suggests that the mapping between perceptual salience and typological frequency is not always transparent. For example, clicks would appear to be perceptually salient (though they are difficult to temporally order relative to adjacent sounds) since they involve a rapid increase in energy at their release, but nevertheless they are crosslinguistically

rare. Under Lindblom and Maddieson’s account (section 3.5.1.3) incorporating articulatory ease in addition to perceptual salience, clicks are typologically rare due to their articulatory difficulty.

Clements 2003 addresses the issue of whether the economy that he captures with reference to phonological features could actually reflect a phonetic preference for gestural economy. In other words, it could be the case a priori that featural economy is really articulatory economy that could be modeled more directly with reference to gestures rather than indirectly via phonological features encoding the articulatory gestures.

To tease apart the two possibilities, Clements compares the predictions of the Browman and Goldstein (1989) model of articulatory phonology in which gestures are captured via features referencing properties such as

the primary articulator and the location and degree of the constriction. In the Browman and Goldstein model, labiodental and labial articulations are distinguished since only the former involves the upper teeth.

Clements tests the predictions of the two theories of economy by assessing the likelihood of /f/ occurring in a language with /p/ and /s/ (both extremely common sounds) versus one in which either /p/ or /s/

is missing. As the feature-based theory of economy predicts, /f/ is in fact more common in languages with at least one bilabial and one other fricative.

 

 

 

Frequency  of sounds within  languages

It is  instructive to assess the frequency of sounds within languages to determine the extent to which sounds that are typologically more common are also relatively common in languages that have other cross-linguistically rarer types of sounds.it is also a reasonable hypothesisthat the frequency of phonemes within a language mirrors their cross-linguistic frequency.

In order to quantify the relative commonness of sounds within languages, frequency of occurrence was examined for a set of 34 languages whose genetic diversity is roughly commensurate with that of the WALS sample.There are two Slavic languages included in the survey, Czech and Russian, but Czech is only used in the tabulation of vowels and Russian only in the figures for consonants.

Figure plots  the  frequency of occurrence of the  25 consonants  most frequently attested cross-linguistically as compared to the intra- language frequency (computed as the ratio of the observed number of tokensrelative to the number of expected tokens were each sound to occur with equal frequency) for the surveyed languages. For the small set of languages (Basque, Kayardild, Malayalam, Martuthunira, and Tiwi) contrasting dental and alveolar sounds, frequency values reflect the place associated with the higher relative frequency of the two since the typological frequency data conflates the dental and alveolar categories for languages not contrasting the two (which is most languages of the world). Similarly, for the language contrasting dental/alveolar trills and taps (Basque), the intralanguage frequency data corresponds to the frequency of the more frequent of the tap or trill, since sources providing the cross-linguistic frequency data on /r/ are often inexplicit about whether the rhotic is a trill or tap.o the intra- language frequency.

Historical sound  changes may also play a  role in  boosting (or  reducing) language-internal frequency. For example, glottal stop in Samoan is descended from proto-Polynesian *k which has the second highest mean frequency among the sampled languages.The frequency of a sound may also be inflated due to historical mergers. For example, intervocalic /l/ from Latin merged with /r/ in Romanian, a sound change that contributes to /r/ being by far the most common consonant in Romanian occurring at a level more than three times greater than chance (Renwick 2011).

in some languages were open to reanalysis as phonemic /x/. Similar possibilities for reanalysis hold for the phonetically similar /w/ and /v/. According the figure plots the number of occurrences of the 13 most common vowels from the 451-language UPSID survey against their frequency (relative to other vowels) in the 29-language frequency sample. To be consistent with the UPSID values, vowels in the frequency sample are separated into three height categories (high, mid, and low) and, in the case of non-low vowels, three backness categories (front, central, and back) and two rounding categories (rounded, unrounded).

Vowels belonging to different subcategories within these three height and back- ness groups are collapsed. For example, the high front unrounded vowels include both high and lower high vowels, i.e. /i, ɪ/, the mid front rounded vowels comprise both /e, ɛ/, and the low vowels include low vowels of different backness, height, and  rounding  specifications, i.e. /a,  ɑ, ᴂ, ɐ, ɒ/. 

The frequent vowels within languages, /ø, y, ɨ, ɯ/, are also among the five least frequently attested vowels (of the top 11). The two frequency metrics diverge, however, in certain respects. Most striking is the clear separation in frequency between the five cardinal vowels /a, e, i, o, u/ and other vowels in the UPSID survey contrasted with the more gradual cline in language-internal frequency proceeding from more com- mon vowels to rarer ones.Furthermore, schwa occurs with greater frequency within languages than three of the cardinal vowels /e, o, u/, even though schwa is considerably less common across languages.

The frequency distributions within languages As discussed , sound changes typically alter frequency patterns. For example, a merger of two phonemes inflates the frequency of one while either eliminating the other (in the case of an unconditioned merger) or reducing its frequency (in the case of a conditioned  merger). (Potentially the  frequency of both  phonemes  could be reduced if the output of the merger were a phoneme that differs from either of the merged ones.) Under the assumption that sound change is characteristically driven by phonetic and functional considerations,  one would predict that languages would display an overall drift (with local deviations) toward an increase in both the number of phonetically preferred phonemes and their frequency relative to other phonetically less advantaged phonemes.

Marten suggests that speakers are sensitive to considerations of phonetic naturalness even at the lexical level when choosing words to borrow and coining new  words.  Martin  hypothesizes that  words  with  phonetically  advantaged phonemes are preferentially introduced into languages, thereby increasing the frequency of those  preferred  phonemes  relative to  others.  He  explores this hypothesis through a study of Romance historical phonology and models the diachronic development of frequency distributions through a series of computer simulations employing a neural network speech processing model.

The distribution of phonemes is quite consistent at the two stages of English, though there are some differences that Martin (2017:7) discusses. One difference is that voiceless obstruents are more common in modern English, which is due to the loss in modern English of a once productive rule of intervocalic voicing.

Martin (2007,2009) provides an account of the distribution of phonemes using a spreading activation model of speech encoding (Dell 1986). In this type of model, nodes encoding various levels of linguistic representation ranging from high-level semantics down to low-level phonological features are hierarchically interlinked via weighteActivation spreads between nodes as informa- tion is accessed. For example, the word zebra would activate the lexical node associated with the word, those associated with other lexemes belonging to the same semantic field (e.g. lion, giraffe, cheetah, etc.), those associated with the CV and CCV syllable structures of zebra, those associated with the morphological category noun, with the phonemes /zibɹɑ/, with the features comprising those phonemes, etc. A particular lexical item is selected when its activation level reaches a certain threshold, where nodes associated with more frequently occur- ring properties have higher resting activation levels. d connected nodes.

In Martin’s account, words that enter the lexicon tend to contain commonly occurring phonemes. A key factor that contributes to the likelihood of a lexical item gaining traction in the community of speakers is its phonetic attributes. Martin suggests two potential mechanisms by which an asymmetric statistical skewing in favor of /b/ over /d/ could emerge diachronically assuming a starting point without this bias and no systematic sound change that would have elimin- ated /p/ in certain contexts. One possibility is that /d/ could have phonetically less voicing than /b/ for articulatory reasons just discussed, which could lead to the misperception of /d/ as voiceless /t/ (but not /b/ as /p/) over time on a lexeme- specific basis, thereby asymmetrically lowering the frequency of /d/. Alternatively, it is conceivable that speakers preferred to retain, borrow, or coin words that began with /b/  over those that  began with /d/,  a bias that  would lead to a synchronic skewing in favor of /b/.

This suggests that the bias in favor of /b/ in French stems from other sources beyond inheritance, including borrowings and the creation of new words from existing ones through  word-derivation processes such as compounding  and suffix addition or loss, e.g. bêche ‘spade’ from bêcher ‘to dig’, dureté ‘hardness’ from dur ‘hard’. Borrowings in fact show a split in their behavior depending on the source of the borrowed word. Words that were re-borrowed from Latin, predominantly consisting of religious or scientific terms, were skewed in favor of initial /d/ ( 77 borrowings from Latin beginning with /d/ versus 31 starting with /b/), a bias that Martin suggests might be an artifact of the documented statistical bias in favor of /d/ (by roughly three to one) that existed in the Latin vocabulary.

Drawing on this finding from French, Martin incorporates a notion of articu- latory  ease into  his  spreading  activation  model  by assigning higher  resting activation levels to nodes associated with articulatory gestures that are easier to implement. Through a series of computer simulations of his model, Martin shows that the addition of a sufficiently weighted factor of articulatory ease is able to allow even a comparatively rare but  phonetically preferred phoneme to gain eventual statistical prevalence, as in the case of /b/ in the progression from Latin to French.

 

3.7       Phoneme inventories : a summary

The modal number of consonant across languages languages is 21 and the modal number of vowel is five, though the number of vowel rranges from three to 46 and the number of consonants from sixto 95. There are consint of consonant and vowel more common cross linguistically within languages. Evidence suggest that some combination of the competing factors of minimizing articulatory effoert the while perceptual differentiaayion is pivotal in predicting the structure of phoneme inventories.