Last week I was in Barcelona at a Treebank workshop where one of the invited talks was about annotation of temporal expressions. I'm not going to comment much on the talk itself but the treatment of one temporal expression used as a example by the speaker struck me as strangely inadequate.
The expression in question is Christmas and the detailed representation of it included such attributes as EXTENT (1 day), PERIODICITY (1 year) and DATE (yyyy-12-25). The inadequacy lies in the specification of a concrete, discrete span of time, namely the 25th day of the twelfth month. Christmas as normally used is a much more diffuse concept. Depending on the context it can span from 1 day to perhaps the whole month of December. It even seems to be expanding as I grow older -- perhaps when I'm fifty shops will start putting on Christmas decorations in September. So my guess would be that Christmas (unlike Christmas day) refers to the 25th of December only in the minority of cases.
Such fuzzy reference is not uncommon with temporal expression. Consider the Spanish words mediodía (≈ noon) and medianoche (= midnight). I'll talk about Spanish as I'm not quite sure how it works in English. Medianoche is a bit fuzzy but not much. It would be used to refer to the zeroth minute of a given day or a short span of time around it. So if you hear that something happened at midnight, you know that it likely started after 23:30 and ended before 00:30.
The situation with mediodía is more complex. First, it seems to me to be more strechy, not necessarily tightly centered around 12:00:00 hrs. Depending on various factors, such as geographic location, the season or even the day of the week, it might span the period from 11:00:00 to about 15:00:00 hrs. Second, mediodía is not only associated with an astronomically defined moment such as 12:00:00 hrs but also with the time of the midday meal, which in northern Spain is usually taken an hour or two later.
Perhaps a less inadequate way of thinking about temporal expression might be probabilistic, vaguely analogous to the way particles are treated in quantum physics. An electron is not thought of as point-like object with a specific location but rather is represented by a probability distribution (the wavefunction). Instead of a discrete position, we think in terms of the probability of finding a particle in a certain region of space.
So with temporal reference, we can think of the probability that a given moment in time is included in the time-span covered by an expression. In such an approach medianoche would be represented as a curve sharply peaking at 00:00:00 each day. One such peak would look something like this:
The curve for mediodía on the other hand would be more spread out, and perhaps it would have two peaks instead of a single one: the first at the astronomical noon, the second coinciding with the midday hunger peak:
I leave the elaboration of the plot corresponding to Christmas as a excercise for the reader. Have fun, and a have warm, fuzzy Christmas!
