Comprehending Texts
  1. Two approaches to comprehension
    1. Top down approach: emphasizes role of higher level knowledge in comprehension
      1. Draws on idea of a Schema-a representation of knowledge organized around s structure that is shared across examples of the schema
      2. Illustration. One well known illustration of this approach is the demonstration of Bransford and Johnson. One hears or reads a text that begins "The procedure is quite simple"..the text turns out to be difficult to comprehend, and subject recall it poorly. But when the text is preceded by a title-"Washing cloths"-everything is more comprehensible.
      3. The general point of such demonstrations: In normal understanding, much knowledge is necessary beyond what is explicit in the text.
      4. It is useful to note that these demonstration rely on texts that are written with vague referring expressions. Most texts, especially written texts are much more explicit in their referring expression.

        5.  
    2. Bottom-up approach: Emphasizes that the linguistic content of the text is the starting point in comprehension. The information contained in words is quickly combined with the reader's knowledge in an interactive manner
      1. Words-meanings activated automatically; the context detects the right one in a second phase
      2. Parsing-words are assembled into constituents-phrases-on the basis of syntactic strategies (and perhaps other means)
      3. Propositional assembly: The concepts in a sentence are interrelated in atomic units called propositions.
      4. Microstructure vs macrostructure. Micro is the representations of local propositions as the words and sentences are read. Macro is the derived "gist" of the text. In Kintsch's model, one gets from the micro to the macro through specific operations that correspond to rule. The effect of these operations is to delete e from the representation many Microstructure representations (propositions).

        5.  
  2. The C-I (Construction-Integration) model of Kintsch
    1. The model begins at the propositional level, representing propositions extracted from sentences in processing cycles.
    2. Working memory is a factor: Only some words can be kept active in working memory at a given time; what determines what will be in memory during comprehension?
      1. Illustration of sentence boundary effects in memory. Readers can remember more from within the sentence they are processing than from the previous one, for a given number of words.
      2. Carryover-one way a given proposition can be better retained in the Microstructure representation is to have it "carry over" across processing cycles. This happens for concepts that recur across propositions.
      3. Concepts that recur across propositions benefit from argument overlap. That is, the same argument (concept) appears in more than one proposition. This allows more processing of the concept (involved in more cycles) and increases the chances it will survive into the macrostructure representation.
    3. Coherence. A text gains its coherence through argument overlap. Notice this implies that the factor that protects against forgetting is the same factor that promoted coherence. Hence, our sense that texts that are coherent are more memorable than texts that are not.
    4. An illustration of propositional representation for a simple text (Gernsbacher and Shoyer)
      1. I went to the coast last weekend with Sally. We had checked the tide schedules and planned to arrive at low tide because I just love beach combing. Right off, I found three whole sand dollars. Sally found an egg.
      2. This text in a run of the C-I model would be processed in 6 cycles, one per sentence, except for sentence 2 which would have 3 cycles. The amount in any one cycles is taken to reflect limitations in processing resources (working memory). Roughly, one cycle for a clause. The following is a list of propositions that could be assembled. Notice the application of embedding (one proposition inside another) and argument overlap.
        1. Went (S&I, coast)
        2. Time (a, last weekend)
        3. Checked (we, schedules)
        4. Schedules (tide)
        5. And (c, f)
        6. Planned (we,f)
        7. Arrive (we WHEN:low tide)
        8. Because (e.h)
        9. Love (I, beach combing)
        10. When (K, right off)
        11. Found (I, sand dollars)
        12. L. whole (sand dollars)
        13. Three (l)
        14. Found (sally egg)
      3. Although one might determine a somewhat different set of propositions in this example, the point is to notice how many different concepts or propositions underlie this short text. Also notice how the recurrence of concepts across propositions works to promote coherence. And allow some propositions to become part of the macrostructure.
    5. How the model works in two phases-- construction-the general activation of all information associated with a word-relevant and irrelevant-followed by a phase of integration in which the ongoing text representation and external knowledge combine to select the relevant information.
    6. How syntax can affect the strength of activation: Replace "an egg' in the Sally story with "this EGG". Notice you now expect the egg to take on more importance in the story: it will be involved in more propositions. In the model, one can increase the strength of its activation to reflect this and other aspects of how the proposition is expressed.