Conceptualizing Semantic Networks


Imagine a large group of people using the same language over time. Assume that the full text of their messages is available to you in natural language form. How would you come to some representation of what they are talking about? Your first thought may be to use traditional content analysis methods that categorize text, either manual procedures or computerized ones like the General Inquirer [1] or its more recent cousin LIWC, or topic modeling software based on Bayesian statistical procedures. These automated procedures, while computationally sophisticated, are relatively crude at the conceptual level. They merely assign message elements or individuals to a limited number of nominal categories.

Instead of categorizing messages, with a network perspective one can capture the relationships among words within the messages. Defining word-pair link strength as the number of times each word occurs closely in text with another, all possible word pairs have an occurrence distribution whose values range from zero on up. This ratio scale of measurement allows the use of sophisticated statistical tools from social network analysis toolkits. These enable the mapping of the structure of the word network. They identify word groups, or clusters, And quantify the structure of the network at different levels. Using these word-pair data as input to network analysis tools, you map the language landscape. On the map, instead of cities, the nodes are words. Rather than roads, there are links or edges among words.

Travelling through the word network are fleets of social objects. These communication vehicles are the concepts, ideas, or physical things that people linguistically describe. As they link words to these vehicles in the course of their everyday informal and formal communication, this propels them through the network. Sometimes these movements are unplanned. At other times, groups or organizations try to manage vehicular traffic. By means of optimal messages, they try to steer vehicles in the flow of traffic away from certain words or toward them.

Mathematically-based procedures have been developed to create optimal messages. These are constructed through systematic analysis of the paths connecting word nodes interest. The procedures identify the optimal association network across the aggregate social community. The underlying assumption is that stimulating associations across it is more effective as the shortest effective sequence of words based on particular constraints is selected for the message. This is because people process strings of words linearly over time, encoding and decoding them in sequences. Furthermore, the triggering of associations to words in context takes cognitive time. The most effective message, therefore, optimizes the association networks in the receivers' minds as they read or hear the message.

In short, the similarity of messages encoded by individuals, in other words, individuals’ semantic network similarity, constitutes a useful social network construct, in addition to cohesion based on actual communication exchange, and in addition to structural equivalence based on similarity of network position. This similarity variable is semantic equivalence. Some may think this construct means entities have the same linguistic code identifier, such as the same name for a person, organization, or object. These are not semantic network characteristics but semantic attributes of some entity. They are akin to the words in a dictionary or elements of an ontology. In contrast, one can consider the networks among semantic elements encoded by persons or other social units. Our interest is in semantic encoding similarity.

People who string words together in a similar manner are similar to one another in speech/act behaviors. If two people talk alike or write alike, because of this semantic encoding similarity they may be similar on other levels. We assume that language reflects perception and behavior. The people who talk more like each other are likely to behave more similarity to one another given similar contextual circumstances. This is probably because they perceive their environments and their choices for behavior within them more similarly.