Phonological Events and Event Descriptions

 

In Carson-Berndsen (1991) the event concept used in the ASL project was introduced. For the phonological component, a simplex event is either a conventional feature having a property such as voice or a ``higher'', more unconventional feature having a property such as obstruent or syllable coda. The simplex events are only simplex in the phonological component; at the level of the acoustic-phonetic analysis, these events are as a rule complex. A complex event is defined in the phonological component on the basis of the temporal relations of overlap, precedence and inclusion. It was seen there that, based on information in Fant (1973), a complex event plosive could be constructed compositionally from the simplex events occlusion, transient and noise (Carson-Berndsen, 1991:12).

Bird & Klein (1990) show how it is possible to define phonological segments less rigidly in terms of overlapping events. A phonological segment is defined in this paper as phonological segment = < E, > where E is a set of phonological events of the form i:p (interval:property) and is the set of overlap relations between these events in terms of ordered pairs of elements from the set E. For the definition of the phonological segments of German, an event set (i.e. a set of properties of events) is chosen which based on the classification of the International Phonetic Alphabet. Phonologically relevant tiers are introduced to which the events are uniquely assigned; that is to say, the same property of an event must not occur on more than one tier. The relevant tiers together with a selection of possible event properties are:

phonation : voiced, voiceless

manner : plosive, fricative, nasal, lateral, affricate, vowellike, diphthong

place : labial, apical, palato-alveolar, velar, palatal, uvular, glottal, front, back, central

length : long, short

height : high, low, mid

roundness : round, unround

openness : open, close

It is important to note here that these tiers are phonologically relevant and do not therefore correspond directly to event types identified by the acoustic-phonetic analysis. The complex event plosive may be constructed on the basis of overlap of acoustically relevant simplex events such as occlusion and noise. However, much empirical work has still to be done on the compositional construction of complex events from the simplex events provided by the acoustic-phonetic analysis. The events given in the phonotactic description and in the inventory of phonological segments are taken to be basic categories of the phonological level and it may be necessary to enhance the description with different complex events at the allophonic level. However, since the phonotactic representation is underspecified and allows underspecified data, a further enhancement of the knowledge base at the allophonic level can only be advantageous for optimisation purposes.

Below the event-based inventory of phonological segments of German is given where the temporal dimension is incorporated by associating not only a property but also an interval to each phonological event in line with Bird & Klein (1990). A phonological event takes the form i:p and a phonological segment is defined on the basis of sets of phonological events and the overlap relations between them. The format given above, namely phonological segment = <E, > is used for the inventory A selection of phonological segments is given here.

i=, , , , , , ,, , , , }, , }, {,,I, ,,I, ,,I, ,,I, ,,I, ,,I, ,,I, ,,I, ,,I, ,,I, ,,I, ,,I, ,,I, ,,I, ,,I, ,,I, ,,I, ,,I, ,,I, ,,I, ,,I

@ =, , ,, , , , , ,, , }, {,,I, ,,I, ,,I, ,,I, ,,I, ,,I, ,,I, ,,I, ,,I, ,,I, ,,I, ,,I, ,,I, ,,I, ,,I

m=, , , , , }, {,,I, ,,I, ,,I

p=, , , , , }, {,,I, ,,I, ,,I

ts=, , , , , }, {,,I, ,,I, ,,I

Further relations between events are necessary in order to provide a complete phonological description. In addition to the overlap relation, the relations precedence,inclusion and immediate precedence are required. The relations have the following properties:

Note that inclusion does not refer to true simultaneity. It may be the case that A includes B and B includes A but this is not a property of the relation.

On the basis of these relations between events, Bird & Klein (1990:43) define autosegmental representations using the notions of melody and phonological event structure.

A melody is an ordered pair m = < E, >

(i)

E is a set of events,

(ii)

is an irreflexive and an asymmetric relation over E and

(iii)

E contains unique maximum and minimum elements with respect to , denoted max(m) and min(m) respectively.

A phonological event structure is an ordered triple R = < M, , > where M =m,.....,m where each m has a distinct type and two sets of ordered pairs specifying the precedence and overlap constraints existing between elements of distinct melodies from M.

Bird & Klein (1990) use these notions to describe single utterances (words or syllables) as opposed to a set of syllables or words. This means that for each word or utterance only one phonological event structure is required. However, the domain of phonotactic description is not a single utterance but a description of all potential syllables or phonological words of a language. Therefore, for the purposes of phonotactic description, a disjunction of phonological event structures is required which in turn requires a disjunction of melodies of a single type. In order to avoid defining a single phonological event structure for each possible (i.e. actual and potential) syllables of a language, a means of generalising over such structures is necessary. It is here that underspecified or partial information plays a role together with other generalisation strategies using the phonological knowledge from the feature-based phonotactic network for German found in Carson-Berndsen et al. (1989).

The following example indicates the complexity of the problem. Assuming the following canonical form for a German syllable:

C V C

For this structure , melodies would have to be defined for each tier in the phonological representation (phonation, manner, place, length, height, roundness, openness). The tiers roundness and openness require only two disjunctive melodies each. In the case of the phonation tier, only two properties of the voicing events can occur. However, there are five disjunctive melodies as seen below where i:p represents the interval and property of the event (voiced or voiceless):

1. PHONATION: voicelessPHONATION: voiced>>, PHONATION: voiceless
2. PHONATION: voicelesssc phonation: voiced>>, PHONATION: voicelessPHONATION: voiced>>, PHONATION: voiceless
3. PHONATION: voicedPHONATION: voiceless>>, PHONATION: voiced PHONATION: voiceless>>}, {
4. PHONATION: voicedPHONATION: voiceless
5. PHONATION: voiced

Due to the number of event properties which can occur on the tiers and to the fact that they are relevant for both consonants and vowels, the definition of melodies for manner and place are much more complicated. In addition to the definition of melodies, however, the phonological event structure requires the specification of the overlap and precedence constraints existing between elements of distinct melodies.

As mentioned in section 3 above, up to now, phonotactic description has relied on the phonological segment as the basis for the statement of well-formedness constraints on phonological representation. Therefore, the information which can be gathered from such sources must be adapted to cater for event descriptions of the kind discussed in this section. In Carson-Berndsen et al. (1989) a complete feature-based phonotactic network for German syllables was presented. It is the phonological knowledge contained in this representation which is used for the phonotactic description in section 5. The event-based network structure which allows for additional generalisations can be compiled into disjunctive phonological event structures describing all the possible (actual and potential) syllables of the language.

Before the complete event-based phonotactic description is presented in the next section, a diagram in terms of phonological segments is given in fig. 1 overleaf in order to show the permissable combinations. It is these combinations which are described by the event-based autosegmental phonotactic representation in section 5. Phonological classes, as in the phonotactic description below, are grouped together within square brackets and are underspecified in the feature or event-based phonotactic network. Since the phonotactic description is used as a knowledge component for a syllable parser of a speech recognition system, it also describes morphological appendices which are usually considered to be extrasyllabic. This is due to the fact that the syllable parser should not be expected to disregard information provided by the acoustic-phonetic analysis which is useful for morphological parsing. Such extrasyllabic information may be marked by the syllable parser if required.

German phonotactic network