The role of semantic transparency in the processing of Finnish compound words

Abstract

Three experiments examined whether the semantic transparency of a long Finnish compound word has any influence on how the compound word is encoded in reading. The frequency of the first constituent (as a separate word) was manipulated, while matching for the frequencies of the compound word and of the second constituent. The effect of this frequency manipulation on encoding time served as a ‘marker’ that the compound word was processed, at least in part, componentially. In Experiment 1, each high-frequency transparent compound was paired with a low-frequency transparent compound, and each high-frequency opaque compound was paired with a low-frequency opaque compound. A sentence frame was created for each pair that was identical up to the word following the target word. In Experiments 2 and 3, the matching was done between transparent and opaque word pairs. In addition, Experiment 3 had a display change manipulation in which most of the second constituent was not visible until it was fixated. Readers’ eye fixation patterns on and immediately after the target word were examined. Reliable first constituent frequency effects were observed in the fixation duration measures on the target word, but there were no effects of transparency. In addition, a comparison of the display change condition to the standard condition indicated that the constituents of the compound word were processed sequentially. It thus appears that the identification of both transparent and opaque long compound words takes place, at least in part, by accessing the constituent lexemes and does not rely on constructing the meaning from the components.

Acknowledgments

The first author was supported by Grants HD17246 and HD26765 from the National Institutes of Health and the second author acknowledges the support of Suomen Akatemia (the Academy of Finland). We thank Jonathan Grainger, Gary Libben, and an anonymous reviewer for their comments on a previous version of this article. We are also grateful to Tuomo Haikio for his help in analysing the data of Experiment 3.

Notes

¹Usually, for most transparent compound words, the meaning of the word cannot be uniquely computed from the constituents, as a carwash could be some sort of device that washes with a car; instead the meaning is usually a highly plausible combination of the constituent meanings.

²Some recent attempts to model our data quantitatively (Pollatsek, Rayner, & Reichle, 2003) suggest that the two routes are not likely to be independent, as in a standard'race' model. That is, the sizes of the frequency effects observed in the Finnish compound word experiments were virtually impossible to fit with any reasonable assumptions about processing times for the words or the constituents.

³The main reason we did not use a completely crossed design (which would have allowed transparency and frequency to both be within-item variables) is that the resulting sentences would have been much less natural. That is, it is quite difficult to construct a sentence frame such that four compound words can equally felicitously fit in. Further, in such a design, we would have had to construct four such frames for each quartet of target words.

⁴It is highly unlikely that the trend for the frequency effect being slightly larger in the duration of second and third fixation for transparent compounds is a result of the second constituent in the low-frequency opaque condition being of lower frequency than in the other conditions. This is because the trend is primarily due to relatively long fixations observed in the low-frequency transparent condition.

⁵The percentages of excluded trials were 17.0, 20.9, 27.0, and 23.5, for the transparent high-frequency first constituent, opaque high-frequency first constituent, transparent low-frequency first constituent, and opaque low-frequency first constituent condition, respectively.

⁶It is possible that the processing of opaque compounds receives benefit from a sentence context (even a neutral one).

⁷This points to another reason that it is unlikely that transparency would affect early measures. That is, if early measures do not even pick up whether a meaningful second constituent is present or not, they are quite unlikely to be sensitive to whether the composition of the meanings computes semantically.