«Andrés Gregor Zelman The University of Amsterdam 2002 ii Mediated Communication and the Evolving Science System: Mapping the Network Architecture of ...»
81 The first notable thing about this distribution is the fact that the majority of keywords occur with a high frequency in the first time period (Amsterdam, Analysis, Data, Knowledge, Networks Task and Theoretical), and all keywords tapered in their general usage over time; indeed, words designated as key in the final time period were all negatively de-noted, suggesting a decreasing importance of these terms in the electronic medium towards the end of the project. The keywords Evolutionary, Meaning, and System each occurred with a high frequency in the middle time periods of the dataset, and then decreased significantly in emphasis.
Table 5.4: Electronic x Each shows the results of comparing the word lists from each time period with each other, not the full document set.
The aim of this technique was to extract not only those words which are designated as keywords, but to isolate those that are specifically shared between texts, thereby representing periods of transition.
By comparing the texts with each other instead of the full document set, one gains a better sense of the transmission of information (measured as keyword distribution) over the time periods. It is interesting, for example, that words one would relate directly with the functioning of the SOEIS research project (Data, Meeting, Project, and Task) all decreased in emphasis over the transition phases. Perhaps more significant is that those words which arguably represent processual aspects of the SOEIS appeared to increase in importance over the dataset (Meaning and Understanding). It is also notable that Identity, and Integration decreased in the second transition phase and then rose again in the final phase, whereas the keywords Evolutionary, and History had a high frequency and were positively de-noted in the second phase and then decreased significantly in the final. However, in the context of the SOEIS research project these latter words are not as central to the project.
When compared with the results of the print analysis, it is interesting that those words from the electronic dataset associated with the functioning of the research project (Data, Meeting, Project, and Task) all decreased in emphasis over the dataset, whereas function words in the print dataset (Empirical, Methods, Partners and Task) were designated negatively in the middle phase and increased in frequency in the final phase. This result confirms our expectations. The differences in identified keywords reinforce what was learned by analyzing the print dataset – that medium and designated function are uniquely related.
82 Finally, a collocate analysis was performed on the electronic dataset set using a handful of keywords repeatedly isolated in the network analyses outlined above (Evolutionary, Meaning, Project, and Task). Evolutionary and Meaning were selected because they emphasize the conceptual aspects of the SOEIS, whereas Project, and Task have been specifically selected for their direct relevance to the operation of the SOEIS project. In order to determine the occurrence of keyword collocates the electronic dataset was examined for neighbourhood collocates and their tightly associated keywords. As with the last chapter, the network analysis entailed the examination of the original texts (E1, E2, E3, and E4) for frequently occurring neighbourhood collocates of the selected keywords.
For the collocate analysis the occurrence of each query word was first plotted across the electronic document set.5 The results of the electronic collocate distribution analysis are presented below in Table 5.5: Distribution of Electronic Keywords for the Collocate Analysis.
The distributions of the keywords selected from the Electronic dataset differed significantly from the distributions outlined in the collocate analysis of the print dataset. Most notable is that the frequency and standardized frequency (per 1000 words) of the keywords Evolutionary, Meaning, Project, and Task remain aligned, unlike the distributions found in the print dataset where the keywords selected for the collocate analysis were shown to have to have differing values depending on the frame of analysis. For example, Evolutionary occurs quite frequently in the second document set, and this is the same when measured for its occurrence every 1000 words. This level of correlation was not observed in the print dataset. This difference between the print and electronic datasets may be due to the cut and paste environment identified with EU research projects in the last chapter. The print communications dataset was found to accumulate over the time periods (as an archived dataset that was continually updated and resubmitted); by contrast, the keyword distribution in the email communications appears to be more stable over the time periods.
A further result was found here comparing the distribution of the keywords Project and Task in both the print and electronic datasets. In the case of the keyword Project, its frequency is high in the first time period and decreases over the course of the SOEIS project – this is the case for both print and electronic datasets. By contrast, the electronic keyword Task appeared frequently in the first dataset and decreased over time, whereas in the print dataset Task increased in relative use. As with the previous analysis, the distributions of the electronic keywords gave an immediate sense of the cognitive priorities of each time period. These words were further contextualized by isolating their central neighbourhood collocates.
5 Visual displays revealing the collocate word distributions across the individual documents for each time period were generated and can be viewed in Appendix B.7 through B.10 83 The difference between associate and neighbourhood collocates is important.
Associate collocate keywords are those words which one would logically group together based on association to a common theme; no computer assisted textual analysis program is able to organize collocates in this manner. By contrast, finding neighbourhood collocates (words which occur near other words) is very easily accomplished using textual analysis computing tools. Interestingly, the collocate analyses performed in this study have been able to hone in on associative words by performing not only a singular neighbourhood collocate analysis but by grouping keywords with individual collocates. In this way we gained a better perspective on the cognitive biases revealed by the words and their collocates, rather than just identifying individual neighbourhood collocates.
The key collocates of each query word were then isolated and a number of secondary collocate associations were selected for comparison; following the logic of the previous analyses, the additional words selected for the secondary collocate analysis were selected on the basis of their relevance to the SOEIS project. The secondary collocates were run through the electronic dataset with the original query word to determine its tightest word associations. The selected collocates for each keyword, and it successive co-collocates are shown below in Figure 5.1: ‘Evolutionary’ Collocate Analysis, Figure 5.2: ‘Meaning’ Collocate Analysis; Figure 5.3: ‘Project’ Collocate Analysis, and Figure 5.4: ‘Task’ Collocate Analysis.
The query performed on the keyword Evolutionary produced a list of interesting collocates from which two additional query terms were selected: Systems and Neural.
When sought in tandem, the words Evolutionary and Systems were shown to have CoEvolution, Fuzzy and Neural as their collocates. These words appear to be associate collocates, as they are so close in theme to the keyword. Similarly, when Evolutionary and Neural were queried together, the resulting collocates included: Programming, Networks, Evolution, Co-, Genetic, Artificial, and Systems. Again the associations isolated in this analysis provide us with the necessary arguments to claim that cognitive networks of the SOEIS electronic communication have been identified and that these networks resemble associate collocates – that is, words clustered together by association. Arguably, one might expect that the keyword Firms would arise in the context of the keyword Evolutionary, but this is not the case for the electronic dataset.
84 If Evolutionary were used as a query keyword for a collocate analysis of the print dataset, then it is likely that the word Firms would arise in its neighbourhood given the centrality of both words to the second time period of the print dataset; the relationship between these words was observed in the keyword network analysis of the SOEIS print communications.
When Meaning was used as the query word several unusual collocates were identified. In particular, for the first time we observe the appearance of personal names, a dynamic that is intensified when both meaning and words are queried together. The latter resulted in a collocate list comprised only of first and last personal names. More surprisingly, when the words Meaning and Information were queried together no collocates were found. Rather unlike the keyword Evolutionary which had many interesting associate collocates, Meaning has no associate collocates when combined with Information, and when combined with Words, only personal names as neighbourhood collocates were identified. The word Meaning either has little functional role here in the SOEIS or its co-occurrence with personal names is due to the use of the term Meaning near the end of individual email communications.6 With the keyword Project as the query word many SOEIS project related collocates arose. Of these both Meeting and Meetings were selected for further analysis, as was 6 Email communications often end with a personal name. Note that all email signatures were removed in the email filtering process; personal names that were part of the message were not removed.
85 the term Management. Project with Meeting and Meetings yielded several locations where project meetings were held (though not all), the word Date as well as the term Management and one personal name. When Project and Management were queried together for shared collocates, only the word Meeting arose. This collection of words suggests the planning of meetings and co-ordination of events. By contrast, the results of the print collocate analysis yielded words such as Research and Dissemination which rather suggested the ‘doing’ of research.
The final word queried for collocates was Task. From the results, Project, Task and Theoretical were selected for the combined searches. When Task and Project were queried together the words Member and Responsible arose; with Meeting and Task the result was the word Bielefeld. Finally, the query of Task and Theoretical yielded absolutely no collocates. By comparison, the print collocate analysis of the word Task yielded words like Implications and Deliverables. The electronic communications do not appear to discuss the act of fulfilling the SOEIS tasks, but rather focus on the meetings, members and the project itself.
Unlike the print analysis where it was discovered that most words occur in the final document set, thereby indicating an ‘aggregating text’, the electronic text appears more evenly distributed. This is reinforced by the distribution between keywords as highlighted above. Thus, while the print document sets appeared to serve an archival function, the electronic appear to serve a supplementary function whereby issues are discussed that are not seen as central to the performance of project tasks, but to the evolution of the cognitive realm of those communicating. The architecture of the electronic communications has been enriched with the network analysis, suggesting different functional roles of print and electronic communications in the context of the SOEIS. The electronic distribution reveals that the types of keywords repeating across the time series prove to be more informally oriented, to contend with managerial issues of the research project, and to appeal to a larger audience than the print distribution.7 7 It is notable that certain words simply do not occur in the print dataset whereas they are prominent in the electronic; the word ‘dear’, for example, was found among the top 50 electronic keywords.
Informal words were used more frequently in the electronic medium – email communication does differ from print in this respect, as argued by Collot & Belmore (1996).
86 System The third stage of analysis involved the assessment and comparison of the document sets to determine the nature of transformation in the SOEIS electronic communications system. Critical transitions or path dependencies were not located in the print database, but are measured for here in the electronic to discover if they differ significantly, and if so then to describe why. In addition, the electronic dataset is examined for the specificity and transmission of word use, and there results are compared with the results of the print system analysis.
The system transformation analysis involved the identification of critical transitions over the four electronic documents. The words in each respective time period were compared; first the linear transition periods between E1 and E2, E2 and E3, and E3 and E4 were examined for continuity, and then compared with non-linear relationships between E1 and E3, E1 and E4, and E2 and E4. The comparison will show whether the linear pathway along which the electronic writing evolved was critical for the information’s development. As argued in Chapter IV: Analysis of Print Communication critical transitions or revisions in the dataset can be identified by comparing the bits of information exchanged linearly between the respective time periods with the bits of information shared non-linearly. However, since words occur in some of the time periods and not others, two distinct levels of analysis were performed.8 The comparison was first performed upon the complete word lists of each respective time period (Electronic-All), and then upon only the words shared across all time periods (Electronic-Shared). The results of the comparison performed on the complete word lists for each time period are shown below in Figure 5.5: Four Time Periods Compared Linearly (7672 words) and Figure 5.6: Four Time Periods Compared Non-Linearly (7672 words).