3.3.3. Discussion

In sum, we observe significant differences in distributional properties between word categories. Word categories differ with respect to the frequency range they populate, the average utterance position, and lexical diversity. From the perspective of learning, this implies that the properties which distinguish word categories interact with the order in which they are learned while the order in which they are learned appears to be a consequence of the regularity with which they are represented across samples.

We sugges<sup>t</sup> that the aggregation effects in token distributions across lexical classes is correlated to the degree in which category types are regularly distributed across language samples, reflecting the extent to which their communicative function is mediated by the contextual frames they appear within. Our analysis shows that lexical classes differ both in the average utterance position and in the rate at which lexical diversity increases as a function of utterance position and that the increase rate is inversely related to the average utterance position of the class.

In the next part of the analysis, we explore the extent to which the variety of abstract grammatical constructions in which words are embedded can serve to capture the differences in distributional structure and recurrence patterns that we observe across lexical categories.

#### *3.4. Distribution of Grammatical Context*

#### 3.4.1. How Do Different Parts of Speech Carry Out Their Communicative Function?

Words often occur in multiple grammatical contexts. The word *claim*, for instance, appears 5989 times in the spoken section of the Corpus of Contemporary American English [46]: 2719 times as a verb and 3270 times as a noun, 3016 times as noun singular, 1994 times as base form verb (1), and 1276 times as an infinitive (2), so that the three instances of *claim* in the three following examples serve distinct communicative functions which are not equally probable.


The particular uses of *claim* that speakers intend to communicate will thus be determined by the lexical and grammatical context in which it is used. If one were to count the word *claim* as one type across all the contexts it occurs within without taking into consideration its lexical status, one would run the risk of aggregating over the multiple communicative functions it serves, and this problem will clearly increase as a word's frequency increases.

To explore the extent to which lexical subcategories receive support from these kinds of contextual frames, we next analyzed the distributional properties of the frames that words are embedded within by lexical category.
