9 Metonymy

Metonymy is a is a concept closely related to metaphor and another way to extend the use of a lexical item, and therefore, an important structure. Deignan (2005) argues that metonymy and metaphor are two sides of the same coin, and it is, in fact, difficult to draw a line sometimes. Typical examples include the part-whole- and whole-part- relationships. We can refer to smart

  1. The university had sacked Mr Jeffries. (BNC)
  2. Then I had to come back and read Shakespeare (BNC)
  3. Apple announced there new app on Monday.

As an example of metonymy let’s look at uses of Amazon. The corpus data needs to be quite recent since Amazon wasn’t called Amazon before 1995, and the web services which made it famous were started only in 2002. Many of the popular corpora that are large and balanced, like the BNC, take a lot of time and resources to be compiled. As a result, they are sometimes too old for some types of research question. While the BNC is perfectly fine for a large range of grammatical and lexical topics, it is too old to show the company name Amazon.

[no corpus]> BNC
BNC> "Amazon"

…
 1548472: hern Colombia , especially its [[[ Amazon ]]] cocaine laboratories on the b
 2187606:  , a town in the jungle of the [[[ Amazon ]]] Valley . Having read Schiller
 3307845: een of the lower waters of the [[[ Amazon ]]] River . When they were wet fr
 3317618: red miles from land the fierce [[[ Amazon ]]] river stained the dark water
 4197985: the next ten years much of the [[[ Amazon ]]] Rainforest could be wiped out
 4198128:  the systematic burning of the [[[ Amazon ]]] Rainforest . TOMORROW WILL BE
 4198254: at current rates , much of the [[[ Amazon ]]] Rainforest will have been obl
:

All hits are related to the river/region. This is an extreme example, but you always have to make sure that your corpus is representative as a data source. The results of corpus queries are vastly dependent on the makeup of the corpus. The more automatic your data retrieval, e.g. relying on available annotations and frequency lists, the more dangerous unexpected patterns or the (unexpected) absence of expected patterns might become. For instance, if you were looking for a whole class of names of which Amazon is only one, aspects like this might not be immediately be apparent. Automatizing coding in linguistics is very powerful, but with great power comes great responsibility.

You can get information on the corpus by typing info. There, you can find available attributes (is the corpus lemmatized, pos-tagged?), textual annotations (mode, genre, author/speaker), and general information. If the information in the info file is not enough, note that there is usually a publication connected to a corpus, which is the one you also have to cite if you use it as data source. For example, if I use the Corpus of Contemporary American English (COCA), I cite Davies (2008).

So, for Amazon, we need a more recent corpus. A popular choice is newspaper corpora, which can be very large and very up to date. A major disadvantage is that they are only representative of newspaper language. There are also problems with copyrights and paywalls with many corpora. We do offer some newspaper corpora, and there are some available online. For now, let’s compromise on a rather recent corpus that is also reasonably large. We have the spoken version of the new BNC 2014, which you can activate by typing BNC2014-S.

In the 2014 spoken data, we are more lucky. In fact, most of the matches appear to be about the company rather than the place. In order to get rid of the forests and rivers, we could try to look for patterns that only occur with the geographical name and don’t occur with the company name. We noticed that most occurrences are preceded by the. In order to see whether we can exclude them systematically, we first looked at all of those matches.

BNC2014-S> "the" "Amazon"

…
 763807:  of money to do it and that  s [[[ the Amazon ]]] to the Andes oh nice erm but
 766016: can see him going through like [[[ the Amazon ]]] and stuff and they get like r
 790850: the and like river dolphins in [[[ the Amazon ]]] you get you get like river do
1537836: e okay the Himalayas it  s got [[[ the Amazon ]]] it  s got everything yeah and
1694346: r Kindle right because mine is [[[ the Amazon ]]] Kindle then that  s where it
3551760: en yeah and he he travelled in [[[ the Amazon ]]] he followed the river for fif
3773906: osite side of the mountains is [[[ the Amazon ]]] in pretty much like all of La
3774293:   s but part of that is in the [[[ the Amazon ]]] yeah and he  s he  s gone dow
3774587: I mean I did n t spend long in [[[ the Amazon ]]] and it  s hard work mm it s
3774745: ver fainted was when we got to [[[ the Amazon ]]] really ? yeah it was quite sc

This filter looks rather successful, but we do get the company name when it occurs in certain attributive uses, such as Amazon Kindle or Amazon delivery. The number of matches is small enough to actually clean it up manually, but in a larger sample you would want to optimize your query more. For now, we were ok with the results. We might want to exclude attributive uses anyway eventually since they are rather different from the other nominal uses.

To exclude the results rather then restrict to them, we use the ! operator which is a logical not: [word != "the" %c] [word = "Amazon"]. Note that the bracket notation [word = "word"] is the same as using the shortcut "word". This should be your general approach before you exclude anything. Check what is in there before!

An interesting and maybe unexpected pattern that we found while browsing through the data is the fact that Amazon is used metonymically to mean the Amazon online account. This is the same that happens when your parents ask you to send them a whatsapp, or when people are looking for a Kleenex.

  1. I have stuff on my Amazon as well
  2. can you get off my Amazon?
  3. I’ll go on to my Amazon.

Characteristic of this use is the fact that we use possessive determiners my, your, their, her, his in front of them with no noun following (remember: attributive uses). We can figure out how to search for possessive determiners by looking up the right tag in the info file typing info and searching with / for “possessive.” The relevant tag is APPGE so we can search for this by using [pos = "APPGE"]. We might want to identify other possessive structures like genitive ’s and consider other structures that are characteristic of this use.

The results on Amazon in this corpus are rather limited, but the possessive + brand name structure gives us a nice place to start looking into the Whatsapp-Kleenex-Tempo type of metonymy.