Advances in Complex Systems

Random-text models have been proposed as an explanation for the power law relationship between word frequency and rank, the so-called Zipf's law. They are generally regarded as null hypotheses rather than models in the strict sense. In this context, recent theories of language emergence and evolution assume this law as a priori information with no need of explanation. Here, random texts and real texts are compared through (a) the so-called lexical spectrum and (b) the distribution of words having the same length. It is shown that real texts fill the lexical spectrum much more efficiently and regardless of the word length, suggesting that the meaningfulness of Zipf's law is high.

Beyond the local constraints imposed by grammar, words concatenated in long sequences carrying a complex message show statistical regularities that may reflect their linguistic role in the message. In this paper, we perform a systematic statistical analysis of the use of words in literary English corpora. We show that there is a quantitative relation between the role of content words in literary English and the Shannon information entropy defined over an appropriate probability distribution. Without assuming any previous knowledge about the syntactic structure of language, we are able to cluster certain groups of words according to their specific role in the text.

In this paper we present a phenomenological approach to describe a complex systemcolon; scientific research impact through Citation Mining. The novel concept of Citation Mining, a combination of citation bibliometrics and text mining, is used for the phenomenological description. Citation Mining starts with a group of core papers whose impact is to be examined, retrieves the papers that cite these core papers, and then analyzes the bibliometrics characteristics of the citing papers as well as their linguistic and thematic characteristics. The Science Citation Index is used as the source database for the core of citing papers, since its citation-based structure enables the capability to perform citation studies easily. This paper presents illustrative examples in photovoltaics (applied research) and sandpile dynamics (basic research) to show the types of output products possible. Bibliometric profiling is performed over a number of the citing papers' record fields to offer different perspectives on the citing (user) community. Text mining is performed on the aggregate citing papers, to identify aggregate citing community themes, and to identify extra-discipline and applications themes. The photovoltaics applied research papers had of the order of hundreds of citations in aggregate. All of the citing papers ranged from applied research to applications, and their main themes were fully aligned with those of the aggregate cited papers. This seems to be the typical case with applied research. The sandpile dynamics basic research papers had hundreds of citations in aggregate. Most of the citing papers were also basic research whose main themes were aligned with those of the cited paper. This is typically the case with basic research. However, about twenty percent of the citing papers were research or development in other disciplines, or development within the same discipline. In addition, there was a four year lag time between publication and citation by the development papers.

The next step in the understanding of the genome organization, after the determination of complete sequences, involves proteomics. The proteome includes the whole set of protein-protein interactions, and two recent independent studies have shown that its topology displays a number of surprising features shared by other complex networks, both natural and artificial. In order to understand the origins of this topology and its evolutionary implications, we present a simple model of proteome evolution that is able to reproduce many of the observed statistical regularities reported from the analysis of the yeast proteome. Our results suggest that the observed patterns can be explained by a process of gene duplication and diversification that would evolve proteome networks under a selection pressure, favoring robustness against failure of its individual components.

A population of agents plays a stochastic dynamic game wherein there is an underlying state process with a Markovian dynamics that also affects their costs. A learning mechanism is proposed which takes into account intertemporal effects and incorporates an explicit process of expectation formation. The agents use this scheme to update their mixed strategies incrementally. The asymptotic behavior of this scheme is captured by an associated ordinary differential equation. Both the formulation and the analysis of the scheme draw upon the theory of reinforcement learning in artificial intelligence.

A statistical inference method known as ε-machine reconstruction is introduced as a modeling procedure for turbulent transport processes in a climate model. Observational data on the atmospheric boundary layer obtained with a radar wind profiler, a radio-acoustic sounding system, and a Raman lidar system was assembled to construct this type of model for use within the unresolved (sub-grid) scales of a numerical climate model. An ensemble of 500 single-column model runs using the inferred sub-grid turbulent transport models demonstrated comparable performance to an identical ensemble of runs using the standard, eddy-diffusivity parametrizations for the turbulent transport. The primary advantages of the ε-machine models are that they are a less biased modeling framework for complex processes such as turbulent transport, and that they are more memory efficient.

Discovering relevant, but possibly hidden, variables is a key step in constructing useful and predictive theories about the natural world. This brief note explains the connections between three approaches to this problem: the recently introduced information-bottleneck method, the computational mechanics approach to inferring optimal models, and Salmon's statistical relevance basis.

Monte Carlo simulations of a Sznajd model show that if a near-consensus is formed out of four initially equally widespread opinions, the one which at intermediate times is second in the number of adherents usually loses out against the third-placed opinion.