Final (I hope) labelling criteria.

Paul Carter (pgc104@york.ac.uk)
Fri, 20 Nov 1998 10:17:09 +0000 ("GMT)

Dear all,

Below is a draft summary of the decisions made at the RAs' meeting on
Monday, with a couple of additions following discussion with JKL and
consultation of YorkTalk about [r] and diphthongs. I've tried not to list
absolutely every case, but just the general issues.

Best wishes,
Paul

_ Paul Carter ____________________________________ pgc104@york.ac.uk _
Dept of Language & Linguistic Science | http://www.york.ac.uk/~pgc104/
University of York | tel: +44 (0)1904 432660
Heslington, York. YO10 5DD | fax: +44 (0)1904 432673

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
RAs' meeting - UCL, 16.11.98

SUMMARY OF LABELLING DECISIONS

1 NEW LABELS

1.1 NEW CATEGORY

As used in Cambridge, we have adopted [?] to represent a glottal stop.

1.2 NEW SUBLABEL

We have decided to label creak with the new sublabel [_L] (the letter L
for laryngeal; capital to avoid confusion with '1') so, for example, the
end of creak in an [I] vowel would be labelled [I_L].

1.3 NEW OPERATORS

1.3.1 Replacement / Realisation (-)

A segment whose phonetic realisation differs from the phoneme string in
the lexical representation is labelled using the operator [-] so, for
example, a /t/ realised as a glottal stop would be labelled [t-?].

This operator is also used to label phonemes from the lexical
representation for which there is no evidence in the waveform so an
'absent' /t/ would be labelled [t-] (ie /t/ realised as zero).

It is important for the correct operation of Mark's alignment tools that
labels such as [t-] have the same time value as the preceding label but
occur AFTER that label in the .lab file. In this way, the linear order of
the lexical representation is maintained but with the 'missing' segment
having duration 0.

1.3.2 Overlap (+)

Two segments which are completely overlapped are labelled using the
operator [+]. Again, the lexical order is maintained so a realisation of
/hI/ as a breathy [I] would be labelled [h+I]. This is how we plan to
handle rhoticised schwa: [@+r]. An inventory of overlap labels we have
used will be given to Mark.

2 PRINCIPLES

2.1 LABEL FROM THE LEFT

We plan to interpret this principle strictly in the sense that the first
label from the left will be [lab], the second will be [lab_1], etc. This
is in an effort to be as consistent as possible between labellers who may
have different preferences as to which of several possible labels is the
most appropriate in a given stream of speech.

If occasions arise where we strongly feel that this method is not
appropriate, we will discuss individual cases among the three RAs.

2.2 MAXIMAL PERIODICITY

We plan to interpret this principle strictly in the sense that the first
sign of periodicity into a vowel will be identified. This is an attempt
to remove inconsistencies, for example regarding whether aspiration is
'heavy' or not.

3 MISCELLANEOUS

3.1 VOICING ISSUES

In parallel with labels such as [b_v] (marking the cessation of voicing in
a closure), we plan to use labels such as [p_v], though care will be taken
not to confuse genuine voicing with glottal 'ringing'. (Ringing, if
desired, may be labelled with [_!]). If a clear formant structure is
present, the portion of voicing should be labelled [V_1], where [V] is the
preceding vowel.

Liquids and glides present before the onset of voicing after a voiceless
plosive (e.g. in [pl]) will be marked as a voiceless liquid / glide, so we
plan to use [p_c, p, l_0, l] as a sequence, rather than [p_c, p, p_1, l].

Preaspiration is to be labelled [V_n], where [V] is the preceding vowel
and [_n] is an appropriate number.

3.2 RELEASE OF CLOSURE

The sublabel [_c] does not always strictly follow silence; [b_c] is
possible even in a fully-voiced stop.

Multiple bursts will be labelled in the sequence [t_c, t_b, t_b_1], rather
than [t_c, t_b_1, t_b_2].

3.3 ORDERING OF LABELS

See 1.3.1 for comments on ordering of labels using the [-] operator.

Where [/] and [P_c] (where [P] is a plosive) coincide at the beginning of
an utterance, [/] will always precede [P_c] in the .lab file. Once again,
this is to avoid difficulties in alignment.

3.4 DENTAL NASALS

Where dental nasals can be identified as part of the realisation of
phrases such as 'in the', they will be labelled [D]. If no friction and
no dentality can be identified, the label will be [D-] (see 1.3.1).

3.5 STOP + STOP CLUSTERS

If there is no evidence other than formant transitions for the first stop
(i.e. no release), the burst may be labelled using the 'overlap' operator
(1.3.2). For example, a /kt/ sequence may include the label [k+t_c].

3.6 DIPHTHONGS

Label the internal boundary at the start of transitions. So, for example,
in an [aI] diphthong, the label [aI1] would come at the start of
transitions from [a] into [I]. This means that the transitions count as
part of the second portion of the diphthong.

3.7 [r]

For a [VrV] sequence, label [V] at start of steady state, [r] at end of
steady state.

ENDS