Introduction
Many linguistic researchers are interested in semasiological variation, that is, how the meaning of words and expressions may be observed to vary over time or space. One word might have one dominant meaning or use at one point in time, and other meanings may supplant them. This is of obvious interest to etymology. How do new meanings come about? Why do others decline? Do old meanings die away or retain a specialist use?
Most of the research we have discussed on this blog is, by contrast, concerned with onomasiological variation, or variation in the choice of words or expressions to express the same meaning. In a linguistic choice experiment, the field of meaning is held to be constant, or approximately so, and we are concerned primarily with language production:
- Given that a speaker (or writer, but we take speech as primary) wishes to express some thought, T, what is the probability that they will use expression E1 out of the alternate forms {E1, E2,…} to express it?
This probability is meaningful in the language production process: it measures the actual use out of the options available to the speaker, at the point of utterance.
Conversely, semasiological researchers are concerned with a different type of probability:
- Given that a speaker used an expression E, what is the probability that their meaning was T1 out of the set of {T1, T2,…}?
For the hearer, this measure can also be thought of as a kind of exposure rate: what proportion of times should a hearer (reader) interpret E as expressing T1? This probability is likely to be meaningful to a language receiver, as it helps them disambiguate what is said. However, it is not that meaningful a statistic for the language producer (except perhaps in anticipating ambiguity).
From the speaker’s point of view we can think of onomasiological variation as variation in choice, and semasiological variation as variation in relative proportion of use.
The uses of very
Historically there have been three principal uses of very in British English that might be easily distinguished grammatically:
- ‘very N’: the very person – very as a particulariser (arguably a determiner?). Historically the dominant form, but in contemporary English very much a minority form.
- ‘very ADJ’: the very tall person – very (usually considered an intensifying adverb) modifying an adjective. Historically the first alternative use of very to arise, and now the dominant form.
- ‘very ADV’: very slightly moving – very modifying a second adverb, and a frequent case.
Note that this is not a complete description (e.g. ‘very very‘ may be a special case of the last), but these three broad uses of very can be observed to have changed in relative proportion over time. Another way of putting this is that the likely use that a hearer is exposed to in different time periods has changed. (Thanks to Sylvia Adamson for this example.)
Even within a short period of time such as the three decades covered by the DCPSE corpus, it is possible to find change in the way very is used. The following data is extracted from DCPSE using FTFs for precision (i.e. we do not rely on POS tagging but require very to modify the head noun, adjective or adverb).
| very N | … ADJ | … ADV | TOTAL | |
| LLC (1960s) | 21 | 438 | 1,173 | 1,632 |
| ICE-GB (1990s) | 17 | 254 | 957 | 1,228 |
| TOTAL | 38 | 692 | 2,130 | 2,860 |
Share of uses of very over two time periods in DCPSE. See also this Excel spreadsheet.
We can now plot the share of each of these forms over time, with 95% Wilson score intervals to express uncertainty about each observation.
We can conclude that, in this data and over these time periods, ‘very ADJ’ significantly increases and ‘very ADV’ decreases in their share. What does this mean? Note that the third form, ‘very N’, is low frequency and almost constant.
- We do not know what is driving this change. The fact that ‘very ADV’ decreases in share of the dataset may mean that ‘very ADJ’ increases as a proportion merely as a corollary. Alternatively it may be increasing independently.
This graph expresses exposure rates. It does not tell us about what is happening to the speaker’s choice. We do not know from this graph whether the opportunity to express very before an adjective is also increasing between the LLC and ICE-GB subcorpora, or if the opportunity to express very before an adverb is decreasing.
To find out, we need to perform a different analysis against a suitable choice baseline for each case.
The choice of very
If we consider the ‘archaic’ use ‘very N’, we might create a simple alternation set ‘{very, exact, precise} N’ (one might wish to also allow the insertion of same (exact same thing, very same thing, etc.). Plotting ‘very N’ against this alternation set obtains the following graph.
The result is not significant, and indeed similar results are obtained for ‘very N’ in this data against other choice baselines (all NPs, all NPs with a premodifying adverb, etc.). It seems that, in this data at least, ‘very N’ has reached a plateau.
A similar non-significant result is found for the adjective modifying form. When comparing ‘very ADJ’ against the ‘ADV ADJ’ baseline (i.e. all adjectives with a premodifying adverb), we obtain a non-significant result. It seems that given the choice, speakers are not increasingly opting for this form, despite the observed increase in share of use.
However if we evaluate ‘very ADV’ against a choice baseline, we do find a significant fall in our data. This is shown visually in the following graph.
The lower line in the graph shows the changing proportion of uses of adverb-premodifying very given that the speaker has chosen to use an adverb in a premodifying position. The upper line shows the changing proportion if we further restrict the baseline so that this premodifying adverb is an intensifying (and not a comparative or superlative) adverb. This eliminates wh-adverbs and superlatives such as most, for example, which necessarily express a different thought.
In either case, ‘very ADV’ falls in our data.
What do these confidence intervals mean?
Confidence intervals represent the range of values where we expect to find (at say a 95% level of certainty) an observation were we to repeat the experiment, sampling the data in the exact same way. However this does not always imply that the confidence intervals are useful, or mean what we might think! Significance tests are based on confidence intervals, so this question also relates to the interpretation of significance.
Speaker choices against a set of true alternates are free to vary, and therefore p can be any value from 0 to 1. This means that the mathematical model underpinning the confidence interval is reliable when dealing with alternates.
Semasiological observations are also free to vary. That is, it is perfectly conceivable that, say, ‘very N’ could be 0 (no cases are ‘very N’) or 1 (all cases are ‘very N’, as they were historically) in the data. So these intervals are robust. But we still have to be careful about interpreting observed variation, even if it is significant.
When we discussed choice experiments elsewhere we noted that if a baseline is expanded to include numbers of non-alternating cases, confidence intervals become conservative. This is not an issue here. However, something else happens, which we saw in our example. The effect of variation on other forms (other adverbs, say) can obscure or compete with the change we want to investigate.
Note that any two subsets of data (subcorpora) may vary in many different ways, as different speakers, topics, text sampling conditions etc. are sampled across what we have generously referred to as “time”. Reducing a baseline to a set of alternates (see A methodological progression) allows us to minimise the effect of these other factors, but it cannot eliminate them altogether. So wherever the baseline expands to include non-alternating forms, it becomes less likely that the observed change is solely due to the stated independent variable (here, “time”).
If we turn to questions of use, these intervals are problematic for similar reasons. The baseline for all three sets of confidence intervals in the first figure was the total number of cases of very. As we noted, this is the baseline for the semasiological relative proportion, but the conditions in which each instance was sampled will likely vary according to different factors.
This means that sound claims based on observed semasiological change should be carefully worded.
- Accounting for variation as being due to “time” is questionable because there are multiple reasons for a particular choice that we have not filtered out, and these may correlate with other sampling variables. The larger and more varied the corpus the better, but as we found when examining VP density in DCPSE (Wallis 2012), different types of text samples can lead to very different results.
- An apparent change in the relative proportion of one use within a set of uses (such as the increasing proportion of very before an adjective) does not necessarily represent a variation in the tendency for speakers to prefer it, because the opportunity for that use may also have changed.
This means that (as far as possible) a semasiological argument should be backed up with an onomasiological analysis of speaker choices for each distinct meaning. These analyses are tangential and complementary.
Finally, in this example, we looked at uses of very, where the complementary choice can be expressed relatively easily as a lexical-grammatical one. However there are other cases of semasiological change which are not so easy to pose in terms of choice. Levin (2013) examined changes in the meaning of the word think, where alternates would represent other ways of expressing the same meaning.
Citation (extended published version)
Wallis, S.A. 2021. Choice versus Meaning. Chapter 4 in Wallis, S.A. Statistics in Corpus Linguistics Research. New York: Routledge. 77-86.
References
Levin, M. 2013. The progressive in modern American English. In Aarts, B., J. Close, G. Leech and S.A. Wallis (eds). The Verb Phrase in English: Investigating recent language change with corpora. Cambridge: CUP. » Table of contents and ordering info
Wallis, S.A. 2012. That vexed problem of choice. London: Survey of English Usage, UCL. » Post
corp.ling.stats 