Multilinear descriptions


In this figure, the syllable /dOk/ is depicted in terms of the features defined above. This representation assumes the synchronisation function to be the same across features as can be seen in the fact that the feature boxes are all of the same length.

Figure 5: Segmental feature graph of /dOk/
\psfig {figure=figs/,width=0.7\textwidth}

In the multilinear approach, this absolute slicing of phonological representations into sequences of segments is replaced by a description in which features are represented on autonomous levels, so-called autosegmental tiers. Sequencing along tiers is thus independent of the phonemic segments and is based on the Obligatory Contour Principle (OCP) which stipulates that no two adjacent autosegments have the same value. A representation of the syllable /dOk/ which adheres to the OCP is depicted in the following figure.
Figure 6: Feature graph of /dOk/ with OCP
\psfig {figure=figs/,width=0.7\textwidth}

Association between autosegments on separate tiers is related to the notion of coarticulation or overlap of properties. This latter issue will become relevant in the next section. For now, we will take a look at how we can implement this principle using the FST technique as described in the relevant section. For the phonation tier of /dOk/ the following transducer would be required:
    <>                        ==  Null
    <[voiced]>                ==  [voiced]    <>
    <[voiced] [voiced]>       == <[voiced]>
    <[voiceless]>             ==  [voiceless] <>
    <[voiceless] [voiceless]> == <[voiceless]>.
However, this definition is not very general since it only applies to the phonation tier. The OCP can be represented more elegantly and more generally by the use of a variable $F that ranges over features.
    <>      ==  Null
    <$F>    ==  $F <>
    <$F $F> == <$F>.
This definition of the OCP is more general since it applies to any two identical features and can be regarded as a smoothing operation over features. Since we require that the OCP applies to each tier in the autosegmental representation, each resulting tier then has its own melody. In order to cater for such an autosegmental description, we must extend the syllable node as follows:
    <melody>   == OCP:<<structure>>.
We can now infer the following melodies for the phonation and manner tiers for the syllable /dOk/:
   <melody phonation> = [voiced] [voiceless]
   <melody manner>    = [plosive] [vowellike] [plosive].

Exercise 6045

Give the place, manner and phonation melodies for S_tee, S_E6, and the syllables that you defined earlier.


  Feature representations      Nonsegmental phonology