One question is does the resulting scale reflect your intuition as to the relative differences between the letters - does 'banana' seem 3 times as long as 'cheese'? Does 'radiator' seem 8 times as long as 'take'? How important this is may depend on your intended use for the measure. It may be overemphasising the syllable factor through using it a multiplicative factor. If you feel this may not be ideal for your purpose perhaps you could consider that the two variables (number of letters and number of syllables) are not independent. You therefore have some information about the likely letter length if you know the number of syllables. You could get this information from existing psycholinguistic databases, e.g.. average number of letters in a 2 syllable word, in a 3 syllable word, etc. Then you could use this to your advantage by scaling the number of syllables on the basis of how many more/less letters the word has than you'd expect on the basis of the number of syllables. This could mean you'd calculate-

(actual_number_of_letters/expected_number_of_letters)*number_of_syllables

For instance, if on average 2 syllable words have 4 letters but the word chosen has 5 then you'd do (5/4)*2 = 2.5. A 2 syllable word with 4 letters (as expected) would get a value of 2 and a 2 syllable word with only 3 letters would get a value of 1.5 (from (3/4)*2). This may provide a more reasonable scale, if a word had a different number of syllables it would still fall on a similar scale which may make it more comparable, e.g. if a 3 syllable word should have 5 letters but had 4 this would score 2.4 whereas a 2 syllable word of 5 letters would score 2.5. Overall bi-syllabic words would score around 2 and monosyllabic around 1, 3 syllables around 3, etc. This could provide a more natural scale for the words and provide a more reasonable relative weighting of the importance of syllables and letters. There are probably other ways to do this as well. Just a thought if you're still looking for possible alternatives. Hope that helps.

If you're trying to capture both the orthographic and phonological representations in a single measure, something like number of letters + number of phonemes seems the most intuitive to me. But I suppose it depends on which aspects of the word you think are most likely to affect your dependent variable.

Hi Julia,

I was thinking more along the line of capturing syllabic and single letter distinctions within a composite measure... Think of two words:  ego and rat,  both would be scored "6" by your metric (3 phonemes + 3 letters). What I would like to capture is the word length difference conveyed by the fact that ego has 2 syllables.  

Hi Jamie,

like Julia said above, we might need to know something about your purpose before we can advise. Does your task involve presenting written stimuli? Or is there another reason why you think orthographic measures of length might be especially important for your study?

I do feel a bit unsettled by the idea of combining an orthographic and phonological measure of length. Its will be so hard to work out what might be underpinning any effect. Could you perhaps use two alternate (non-composite) measures and choose the one that captures the most variability (if your interest isn't really length)? So for example, syllable number and phoneme number? Or alternatively, if length is the thing of interest, could you compare the effectiveness of alternate measures directly in a regression-type situation?

You;d have to watch with the second option, though, because the more variability a measure has, the better it will predict - so number of phonemes or letters both have a huge advantage over number of syllables here.

Carolyn

As others have pointed out, the question of what method is appropriate is hard to determine because there are many possible ways to combine features related to word length and it seems that the purpose of creating the measurement is needed before the utility or reasonableness of the method selected.

And as you've pointed out, certain methods of combining elements produces equivalent scores despite differences you consider important.

For example, the equivalence of "rat" and "ego" when the method is to add phonemes and letters. This is not just a problem with the use of phonemes and letters, any use of addition to combine elements is more likely to result in increased likelihoods of equivalent scores. Put alternatively, use of addition produces a restricted range of scores compared to multiplication.

This is an inherent feature of the mathematical operation, though. But this has been a known issue in measurement for centuries, and various mathematical transformations are used at different times. Using a base 10 logarithmic transformation, or a natural log, or use of z-scores, or otherwise, again the appropriateness is difficult to evaluate without knowing the purpose.

Your multiplicative method is less likely to result in shared scores that hide what may be important distinctions than an additive scale, but it still has that potential. For example, "ego" has a score of 6 (3 letters * 2 syllables) but so does "strong" (6 letters * 1 syllable).

The likelihood of this occurring would vary by language - German has a rather high correlation between number of letters and syllables compared to English (i.e., if a letter is present in the written form in German, such as the word final "e", it is not silent). But both German and English have a higher correlation between number of letters and number of syllables than French, where the number of silent letters may be very high. The word "habitaient" has 10 letters and 3 syllables (roughly "ab-tay-an"; 3-person past tense of "lived"/"resided"/"stayed" at a location). Maybe 2 depending on your dialect ("abt-an" or "ab-tan").

But there are other factors that many consider critical. "Ego" has 2 syllables, so does "method". But despite the fact that "method" has double the number of letters as "ego", "ego" takes longer to pronounce. However, under your system, "ego" has a score of 6 while "method" has a score of 12. Similarly, "double" has more letters and syllables than "twice", but "twice" takes longer to pronounce. Here, "double" has a score of 12 and "twice" a score of 5.

Even if you are only interested in creating a measurement for visual word representations, vocal pronunciation times are correlated with reading times for words. Different syllables take longer to read/say/comprehend, and the time to read/say/comprehend any letter depends on what syllable, and in what word, that letter is in.

However, if you are truly interested in avoiding a restricted range of scores, using reading or pronunciation times in milliseconds as one of your factors.