Good or Bad, Round or Spiky?
A Russian researcher named Valeri Belianine, who has posted to the Forensic Linguistics Web list, has been developing a new subfield called “phonosemantics” (see here). This is based partly on the wholly legitimate and intriguing topic of “sound-symbolism.”
Although linguistic sounds are themselves essentially arbitrary, in a small minority of cases, certain sounds do seem to have rather general semantic associations, either within a given language or even crosslinguistically. For instance, in English, many words including the sequence \fl\ (flee, flick, fly, etc.) refer to rapid movement, although it is difficult to argue that \fl\ is an English morpheme meaning “to move rapidly.” (One obvious objection to this analysis would be that if it is, we must treat, for example, \i:\ in flee as another morpheme, but what meaning could we ascribe to it?) Crosslinguistically, almost all listeners, whatever their first languages, agree that an object called eekeekee will be spiky, whereas one called oomoomoo will be rounded.
Belianine’s specific approach is based on the work of Charles Osgood (circa 1960). Belianine has applied this to Russian, and he now sees further (if arguably dubious) applications: “We may well hide our emotions, but still we can evaluate almost everything. What about the sounds of the English language? . . . This method may be helpful in finding a proper name for your company, and building your future.” One is invited to participate in an experiment using Likert-scale judgments, involving various phonemes and a range of opposed pairs of evaluative terms, starting with “good” and “bad.”
Unfortunately, there are several major problems with this project. First, some of the linguistic terminology used is informal and imprecise. In addition, certain phoneme clusters (such as \gz\ in exact) are included merely because they are typically spelled with a single letter. Judges are supposed to be reacting to sounds here; any effects relating to awareness of the spelling are surely at best a marginal issue.
More seriously, unannounced assumptions are made about the accent used by the (linguistically untrained) judge. Many nonlinguists who know a little linguistics might think that this would not be a major issue, because they would assume that inter-accent differences involve only phonetic realization, i.e., phonemic transcriptions are neutral between accents. In fact, this is often false. For example, London people on the one hand and most Liverpool people on the other have different phonemes in words such as bath (roughly \bäth\ versus \bath\). But even where it is true in a particular case, this is not enough for Belianine. Even naïve judges are reacting here to sounds rather than to phonemes, and hence to specific realizations (of which they are typically well aware in cases like this, because the realizations are not in complementary distribution). Thus, even if differences are only of this structurally trivial type, they will render responses from judges with noticeably different accents totally incommensurable.
For instance, in words such as the name Bob, there is no issue of phoneme selection; but in realizational phonetic terms, most Americans and Canadians nevertheless have a mid-length, unrounded, low central-back vowel, whereas most English and Australian people have a short, half-rounded, mid low-back vowel. If sounds and especially vowels are the things that matter here, people on either side of the pond are not even reacting to the same type of thing (vowel)! (English people hear Americans’ Bob as Barb, as is famously illustrated in Powell and Pressburger’s World War II movie, A Canterbury Tale.)
There is also a major methodological issue here. Apparently people’s answers are completely different if they are in different “moods,” but this is not factored into the analysis.
Some of these problems may reflect a lack of expertise and some may involve deliberate popularization. But one wonders where the project will go—other than commercially, that is!
I posted critical comments to the Web site. At first, there were some arguments from one site visitor who supported probabilistic treatment of accent differences. The specific method proposed would not be satisfactory, as it involves the often false assumption that there will be one dominant phonemic and phonetic form in each key word, with other forms constituting a small peripheral minority. For instance, for the name Bob, one would have to select as “basic” either the typical American/Canadian vowel or the typical English/Australian vowel, treating the other variant (and all further variants) as peripheral exceptions. This makes no dialectological or sociolinguistic sense, and how far either choice actually reflected the true patterning even for a given unsystematic sample of judges would obviously depend entirely on where and how the sample was drawn.
My offer to assist with dialectological information was ignored. After this, things went quiet.