Network Science—an emerging field drawing from Computer Science, Physics, Sociology, Mathematics and a number of other disciplines—uses networks to model complex systems. Although network science has primarily been used to model social, biological, and technological systems, this approach has recently been used to model complex cognitive systems, providing novel insights into several domains of cognition. The papers collected in the present Research Topic examine the insights that network science has brought to the study of language and language processing. These studies cover many levels of processing (e.g., phonology, semantics, syntax), and different language processes (e.g., perception, production, acquisition).

University of Kansas

University of Basel

Views

Article Views

Article Downloads

Frontiers in Psychology

Psychology of Language

Frontiers in Communication

Insights to the study of Language from Network Science

The GOLD model (Graph Of Language Distribution) is a network model constructed based on co-occurrence in a large corpus of natural language that may be used to explore what information may be present in a graph-structured model of language, and what information may be extracted through theoretically-driven algorithms as well as standard graph analysis methods. The present study will employ GOLD to examine two types of relationship between words: semantic similarity and associative relatedness. Semantic similarity refers to the degree of overlap in meaning between words, while associative relatedness refers to the degree to which two words occur in the same schematic context. It is expected that a graph structured model of language constructed based on co-occurrence should easily capture associative relatedness, because this type of relationship is thought to be present directly in lexical co-occurrence. However, it is hypothesized that semantic similarity may be extracted from the intersection of the set of first-order connections, because two words that are semantically similar may occupy similar thematic or syntactic roles across contexts and thus would co-occur lexically with the same set of nodes. Two versions the GOLD model that differed in terms of the co-occurence window, bigGOLD at the paragraph level and smallGOLD at the adjacent word level, were directly compared to the performance of a well-established distributional model, Latent Semantic Analysis (LSA). The superior performance of the GOLD models (big and small) suggest that a single acquisition and storage mechanism, namely co-occurrence, can account for associative and conceptual relationships between words and is more psychologically plausible than models using singular value decomposition (SVD).

The GOLD model (Graph Of Language Distribution) is a network model constructed based on co-occurrence in a large corpus of natural language that may be used to explore what information may be present in a graph-structured model of language, and what information may be extracted through theoretically-driven algorithms as well as standard graph analysis methods. The present study will employ GOLD to examine two types of relationship between words: semantic similarity and associative relatedness. Semantic similarity refers to the degree of overlap in meaning between words, while associative relatedness refers to the degree to which two words occur in the same schematic context. It is expected that a graph structured model of language constructed based on co-occurrence should easily capture associative relatedness, because this type of relationship is thought to be present directly in lexical co-occurrence. However, it is hypothesized that semantic similarity may be extracted from the intersection of the set of first-order connections, because two words that are semantically similar may occupy similar thematic or syntactic roles across contexts and thus would co-occur lexically with the same set of nodes. Two versions the GOLD model that differed in terms of the co-occurence window, bigGOLD at the paragraph level and smallGOLD at the adjacent word level, were directly compared to the performance of a well-established distributional model, Latent Semantic Analysis (LSA). The superior performance of the GOLD models (big and small) suggest that a single acquisition and storage mechanism, namely co-occurrence, can account for associative and conceptual relationships between words and is more psychologically plausible than models using singular value decomposition (SVD).

Using a high-dimensional graph of semantic space to model relationships among words

Semantic knowledge has been investigated using both online and offline methods. One common online method is category recall, in which members of a semantic category like “animals” are retrieved in a given period of time. The order, timing, and number of retrievals are used as assays of semantic memory processes. One common offline method is corpus analysis, in which the structure of semantic knowledge is extracted from texts using co-occurrence or encyclopedic methods. Online measures of semantic processing, as well as offline measures of semantic structure, have yielded data resembling inverse power law distributions. The aim of the present study is to investigate whether these patterns in data might be related. A semantic network model of animal knowledge is formulated on the basis of Wikipedia pages and their overlap in word probability distributions. The network is scale-free, in that node degree is related to node frequency as an inverse power law. A random walk over this network is shown to simulate a number of results from a category recall experiment, including power law-like distributions of inter-response intervals. Results are discussed in terms of theories of semantic structure and processing.

Semantic knowledge has been investigated using both online and offline methods. One common online method is category recall, in which members of a semantic category like &#x201c;animals&#x201d; are retrieved in a given period of time. The order, timing, and number of retrievals are used as assays of semantic memory processes. One common offline method is corpus analysis, in which the structure of semantic knowledge is extracted from texts using co-occurrence or encyclopedic methods. Online measures of semantic processing, as well as offline measures of semantic structure, have yielded data resembling inverse power law distributions. The aim of the present study is to investigate whether these patterns in data might be related. A semantic network model of animal knowledge is formulated on the basis of Wikipedia pages and their overlap in word probability distributions. The network is scale-free, in that node degree is related to node frequency as an inverse power law. A random walk over this network is shown to simulate a number of results from a category recall experiment, including power law-like distributions of inter-response intervals. Results are discussed in terms of theories of semantic structure and processing.

Walking across Wikipedia: a scale-free network model of semantic memory retrieval

Purpose: Cochlear implants (CIs) enable children with severe and profound hearing impairments to perceive the sensation of sound sufficiently to permit oral language acquisition. So far, studies have focused mainly on technological improvements and general outcomes of implantation for speech perception and spoken language development. This study quantitatively explored the organization of the semantic networks of children with CIs in comparison to those of age-matched normal hearing (NH) peers.Method: Twenty seven children with CIs and twenty seven age- and IQ-matched NH children ages 7–10 were tested on a timed animal verbal fluency task (Name as many animals as you can). The responses were analyzed using correlation and network methodologies. The structure of the animal category semantic network for both groups were extracted and compared.Results: Children with CIs appeared to have a less-developed semantic network structure compared to age-matched NH peers. The average shortest path length (ASPL) and the network diameter measures were larger for the NH group compared to the CIs group. This difference was consistent for the analysis of networks derived from animal names generated by each group [sample-matched correlation networks (SMCN)] and for the networks derived from the common animal names generated by both groups [word-matched correlation networks (WMCN)].Conclusions: The main difference between the semantic networks of children with CIs and NH lies in the network structure. The semantic network of children with CIs is under-developed compared to the semantic network of the age-matched NH children. We discuss the practical and clinical implications of our findings.

Purpose: Cochlear implants (CIs) enable children with severe and profound hearing impairments to perceive the sensation of sound sufficiently to permit oral language acquisition. So far, studies have focused mainly on technological improvements and general outcomes of implantation for speech perception and spoken language development. This study quantitatively explored the organization of the semantic networks of children with CIs in comparison to those of age-matched normal hearing (NH) peers.Method: Twenty seven children with CIs and twenty seven age- and IQ-matched NH children ages 7–10 were tested on a timed animal verbal fluency task (Name as many animals as you can). The responses were analyzed using correlation and network methodologies. The structure of the animal category semantic network for both groups were extracted and compared.Results: Children with CIs appeared to have a less-developed semantic network structure compared to age-matched NH peers. The average shortest path length (ASPL) and the network diameter measures were larger for the NH group compared to the CIs group. This difference was consistent for the analysis of networks derived from animal names generated by each group [sample-matched correlation networks (SMCN)] and for the networks derived from the common animal names generated by both groups [word-matched correlation networks (WMCN)].Conclusions: The main difference between the semantic networks of children with CIs and NH lies in the network structure. The semantic network of children with CIs is under-developed compared to the semantic network of the age-matched NH children. We discuss the practical and clinical implications of our findings.

Semantic organization in children with cochlear implants: computational analysis of verbal fluency

Community structure, which refers to the presence of densely connected groups within a larger network, is a common feature of several real-world networks from a variety of domains such as the human brain, social networks of hunter-gatherers and business organizations, and the World Wide Web (Porter et al., <xref ref-type="bibr" rid="B57">2009</xref>). Using a community detection technique known as the Louvain optimization method, 17 communities were extracted from the giant component of the phonological network described in Vitevitch (<xref ref-type="bibr" rid="B70">2008</xref>). Additional analyses comparing the lexical and phonological characteristics of words in these communities against words in randomly generated communities revealed several novel discoveries. Larger communities tend to consist of short, frequent words of high degree and low age of acquisition ratings, and smaller communities tend to consist of longer, less frequent words of low degree and high age of acquisition ratings. Real communities also contained fewer different phonological segments compared to random communities, although the number of occurrences of phonological segments found in real communities was much higher than that of the same phonological segments in random communities. Interestingly, the observation that relatively few biphones occur very frequently and a large number of biphones occur rarely within communities mirrors the pattern of the overall frequency of words in a language (Zipf, <xref ref-type="bibr" rid="B82">1935</xref>). The present findings have important implications for understanding the dynamics of activation spread among words in the phonological network that are relevant to lexical processing, as well as understanding the mechanisms that underlie language acquisition and the evolution of language.

Community structure, which refers to the presence of densely connected groups within a larger network, is a common feature of several real-world networks from a variety of domains such as the human brain, social networks of hunter-gatherers and business organizations, and the World Wide Web (<a href="fpsyg-04-00553-HTML.html#B57">Porter et al., 2009</a>). Using a community detection technique known as the Louvain optimization method, 17 communities were extracted from the giant component of the phonological network described in <a href="fpsyg-04-00553-HTML.html#B70">Vitevitch (2008)</a>. Additional analyses comparing the lexical and phonological characteristics of words in these communities against words in randomly generated communities revealed several novel discoveries. Larger communities tend to consist of short, frequent words of high degree and low age of acquisition ratings, and smaller communities tend to consist of longer, less frequent words of low degree and high age of acquisition ratings. Real communities also contained fewer different phonological segments compared to random communities, although the number of occurrences of phonological segments found in real communities was much higher than that of the same phonological segments in random communities. Interestingly, the observation that relatively few biphones occur very frequently and a large number of biphones occur rarely within communities mirrors the pattern of the overall frequency of words in a language (<a href="fpsyg-04-00553-HTML.html#B82">Zipf, 1935</a>). The present findings have important implications for understanding the dynamics of activation spread among words in the phonological network that are relevant to lexical processing, as well as understanding the mechanisms that underlie language acquisition and the evolution of language.

Community structure in the phonological network

Network science describes how entities in complex systems interact, and argues that the structure of the network influences processing. Clustering coefficient, C – one measure of network structure – refers to the extent to which neighbors of a node are also neighbors of each other. Previous simulations suggest that networks with low C dissipate information (or disease) to a large portion of the network, whereas in networks with high C information (or disease) tends to be constrained to a smaller portion of the network (Newman, <xref ref-type="bibr" rid="B33">2003</xref>). In the present simulation we examined how C influenced the spread of activation to a specific node, simulating retrieval of a specific lexical item in a phonological network. The results of the network simulation showed that words with lower C had higher activation values (indicating faster or more accurate retrieval from the lexicon) than words with higher C. These results suggest that a simple mechanism for lexical retrieval can account for the observations made in Chan and Vitevitch (<xref ref-type="bibr" rid="B10">2009</xref>), and have implications for diffusion dynamics in other fields.

Network science describes how entities in complex systems interact, and argues that the structure of the network influences processing. Clustering coefficient, C – one measure of network structure – refers to the extent to which neighbors of a node are also neighbors of each other. Previous simulations suggest that networks with low C dissipate information (or disease) to a large portion of the network, whereas in networks with high C information (or disease) tends to be constrained to a smaller portion of the network (<a href="#B33">Newman, 2003</a>). In the present simulation we examined how C influenced the spread of activation to a specific node, simulating retrieval of a specific lexical item in a phonological network. The results of the network simulation showed that words with lower C had higher activation values (indicating faster or more accurate retrieval from the lexicon) than words with higher C. These results suggest that a simple mechanism for lexical retrieval can account for the observations made in <a href="#B10">Chan and Vitevitch (2009)</a>, and have implications for diffusion dynamics in other fields.