Geneva 1997 - Semantic Communities and their Overlap

Words as Actors II:
Semantic Communities and their Overlap


NOTE: The data for this analysis was kindly provided by Jang and Barnett. See: Ha-Yong Jang and George A. Barnett 1994. Cultural differences in organizational communication: a semantic network analysis. Bulletin de Methodologie Sociologique. no. 44. pp. 31-59

Analyzing the matrices has shown, that there are stark differences between the Japanese and the American reports. Now we might want to find out more about these differences.

Look, here are the lists of the most frequently mentioned words: all, US and Japan ...

		All Companies			US			Japan
	599 182 159 149 144 140 98 98 96 89 87 86 84 82 79 78 77 77 72 71 67 66 60 59 59 59 57 56 56 56	we new business sales year products company million growth united fiscal us market billion product net financial states share global other operating continue development economic years management businesses capital customers		392 100 89 73 73 57 56 55 48 46 46 46 45 44 44 43 42 42 39 39 38 38 38 37 36 34 34 34 32 31	we business products company year financial growth sales businesses other product share operating market united global states years customers percent billion continue million earnings board industry quality services customer companies		207 94 71 71 61 60 59 54 51 49 45 44 40 40 37 36 35 33 32 32 32 32 31 29 29 28 27 26 26 26	we sales fiscal year new million business net products japan united billion growth market development income states product capital economic equipment overseas production corporate increase global efforts management operations share

We might want to know more about these words than that they are important. We might want to know in which semantic concepts they are embedded. These concepts are very similar to the social network an actor is embedded in. Words - as actors - have other words that are closer to them and some that they are nor related to.
To analyze the semantic structure these words are embedded in one can perform a quantitative KWIC (Keywords in Context) analyis. Here the analysis is done with the program TACT.

The program TACT allows you to determine the words that co-occur the most with a particular word. You can define the range of the co-occurrences the way you want. It might be a word, two or - as in this case - five. The program gives you for each word (or set of words) you analyze a list that looks as follows:

Word	FREQ	FREQ-ALL	Z-score
products product introduce introductions york zealand family innovative tractors directions host represents introduction appliance chemistry copier generation insight ideas	30 15 3 3 3 3 4 4 4 3 2 2 3 2 2 2 2 2 4	103 47 3 3 3 3 6 6 6 4 2 2 5 3 3 3 3 3 11	10.569 7.964 7.147 7.147 7.147 7.147 6.551 6.551 6.551 6.075 5.835 5.835 5.331 4.632 4.632 4.632 4.632 4.632 4.493



This is the beginning of the output for the word "new" in the American texts. It gives the frequency other words occurred within the range of 5 words and the total frequencies of those words. From these two informations it computes a Z-score that gives you an idea of the strength of the relations (if it would not do that the list would be headed by words like a, the and so on ... )

I computed these list for all 10 words on the two separated texts and selected for each word the 5 words it had the strongest relationship to. The only restriction was that the "alter" word must occur at least 3 times in the whole corpus.

Using a simple spreadsheed I made two lists (one for America and one for Japan) that were later imported in UCINET. I made some analysis and exported the matrices to KP format. Here are the semantic communities of the 10 most frequently mentioned words for the two nations. As you will see there is some difference ...

Japan:

USA: