Gernsbacher's book (1990) explores experimentally her hypothesis that language comprehension involves the construction of structures. As such, it is a good, book-length treatment of her research over the last ten years. The experiments shed valuable light upon the nature of language processing. However, the underlying theory accounting for these results is less convincing, through vagueness, and needs to be made more explicit.
1. Gernsbacher's book Language Comprehension as Structure Building (1990) presents a theory called the "Structure Building Framework". The theory states that humans comprehend language by constructing "structures". As new information is presented, the human builds, by a process known as "shifting", new structures. These structures, by a process called "mapping", are then merged with the previously built structures. Shifting and mapping are controlled by structure "enhancement" and "suppression". In the theory, the first mentioned word/sentence/whatever has a priority over later mentioned words/sentences/whatever. The theory is taken to be realised in a connectionist architecture. To test the theory, a series of experiments is presented. These experiments, most of which measure the reaction time to spot a given word, are claimed to support the hypothesis.
2. I found the book to be heavy-going at first, largely due to the lack of a detailed theory up-front that would motivate the numerous experiments. It would have been nice if Gernsbacher gave her work more context, that is, a description of how her work relates to other research on psycholinguistics.
3. When the experiments were presented, the repetitive format hindered eager reading. For example, p.13 contains the phrase "To encourage our subjects to attend to all aspects of the sentences ( ...), we followed each experimental sentence with a two-alternative question." This phrase is then repeated, without the parenthetical, on p. 15. Likewise, the results are presented as crude bar graphs, showing the mean reaction time, in nearly all of the experiments.
4. In some of the experiments, the differences seemed to be close, and hence it would have been nice to have, in addition to the mean, other statistical information. Standard deviations and standard errors would have given a more convincing account of the results.
5. The book picked up by the third chapter, in part because of the greater amount of supporting text and in part because of the greater reference to other people's work. The best chapter was the final one, which gave a detailed, critical account of her theory. This account should have been in the introduction, thereby giving the advantage of first mention.
6. My background is in computational linguistics, and not psycholinguistics. I therefore do not feel competent to criticise in detail the experimental design. However, it is evident that Gernsbacher is trying to cover a lot of material in her book. Consequently, she does not fully explore the results of each of her experiments. On p.14, she investigates first mention and its relation to the object and subject position of a sentence. That is, would first mention be evident despite the first mention being in the subject or the object position. She did find this, but also found (though did not comment upon) a slower reaction time for first mentioned names in the object position. This rapid skipping from experiment to experiment is evident throughout the book.
7. On p. 16, Gernsbacher again reports that first-mentioned words are recognised more quickly than other mentioned words. However, she does not explain why having a postposed adverbial phrase, but with the target word being at the front of the sentence, is slower than for the same sentence, but without the postposed phrase. Indeed, having a preposed phrase before the target word is quicker than having a postposed phrase. This is unexplained by Gernsbacher.
8. On p. 63, when talking about coherence helping in mapping, the author says that reaction times for recognising a target word are quicker when using a coherent text than when using a text that is less coherent. She says that this supports the idea that coherence helps in the mapping process. However, playing the devil's advocate, I could say that no structure building takes place at all for coherent texts. Structures are not mapped to other structures. This would be consistent with the finding that coherent texts are quicker to read, but does not support her structure mapping view. That is, the experiments might show interesting results, but I am unconvinced that those are due to the reasons Gernsbacher gives.
9. Gernsbacher is shy of linguistics. This seems odd for a book on language comprehension. She does mention linguistics (p.72 and p. 241), but seems to believe that theories such as (the now long-outdated) transformational grammars (TG) are taken to be psychologically real. Because TG is evidently inadequate, she concludes that linguistics as a whole has nothing to offer. This is a straw man argument, and would be more convincing if she looked at, for example, Government and Binding Theory (Chomsky 1981). Consequently, her theory says nothing about linguistic processing, which is a pity.
10. Throughout, the author is concerned with performance, and not competence. There is no mention of knowledge of language at all.
11. After my initial trepidations, I soon got to grips with the book and found it to be both informative and interesting. I can recommend it as a book-length treatment of Gernsbacher's approach to language comprehension. The same cannot be said of it as a general introduction to psycholinguistics, given the lack of background material. At least now I know of the importance of that first word, even if I don't know exactly why it is important.
Chomsky, N. (1981) Lectures on Government and Binding. Dordrecht: Foris.
Gernsbacher, M.A. (1990) Language Comprehension as Structure Building. Hillsdale NJ: Lawrence Erlbaum.
Gernsbacher, M.A. (1992) Precis of: Language Comprehension as Structure Building. PSYCOLOQUY 3(69) language-comprehension.1.gernsbacher.