Network Representation of Constructional Semantics


Alvin Cheng-Hsien Chen
(陳正賢)

National Taiwan Normal University, Taiwan

20 August, 2021, ICCG11

Outline

  • Background of the Study
  • Space Particle Construction
  • From Corpora to Network
  • Network Analysis of Constructional Semantics
  • Conclusion

Background of the Study

Co-occurrences of Linguistic Units

  • Co-occurrences have been the central focus of usage-based linguistics.

    • Collocation, Colligation, Collostruction
  • With the increasing availability of corpora data, distributional properties of linguistics units are accessible.
  • The systematic distributional patterns are closely connected to the semantics of the linguistic units, forming the basis for their semantic representation (i.e., Distributed Semantic Representation)

How to represent constructional semantics?

  • When we study lexical semantics, we look at words’ collocates and identify the emerging semantic coherence from these collocates.
  • Following the same distributed semantic hypothesis, we can analyze constructional semantics in a similar way.

Grammar as Network

  • Language consists of words and constructions.
  • Every word or construction would develop their own preferred linguistic co-texts/contexts in language use.
  • The emerging patterns in these preferred linguistic co-texts/contexts are the semantic traces of these words or constructions.

Space Particle Constructions

SPC in Mandarin

「在zai + Reference_Landmark + Space_Particle」

  • 在zai…LM 下xia ‘DOWN/BELOW’
  • 在zai…LM 上shang ‘UP/ON’
  • 在zai…LM 前quian ‘IN FRONT OF’
  • 在zai…LM 後hou ‘behind’
  • 在zai…LM 中zhong ‘IN’
  • 在zai…LM 內nei ‘IN’
  • 在zai…LM 裡li ‘IN’
  • 在zai…LM 外zhong ‘OUTSIDE OF’

Keys for constructional semantic representation?

  • Constructional meanings emerge from the distributional patterns:

    • Links between construction and word：Which words tend to co-occurr with the construction?
    • Links between word and word：For those words co-occurring with the construction, how are these words connected？
    • Links between construction and construction：Is one SPC more connected to another SPC? (e.g., zai+LM+xia vs. zai+LM+shang, zai+LM+xia vs. zai+LM+nei)

From Corpora to Network

Corpus Learning

  • Corpus data are a good source for the modeling of the linguistic co-occurrences, i.e., the three types of links (word-construction, word-word, construction-construction).
  • With these links, we can create a grammar network, which could quantitatively represent the distributional patterns of the constructions.
  • We can then utilize the network science methods to both visualize and analyze the constructional semantics quantitatively, which can be a very important step to the modeling of our knowledge of grammar network.
  1. Barabási, Albert-László. (2016). Network Science. Cambridge University Press.
  2. Diessel, Holger. (2019). The Grammar Network: How Linguistic Structure is Shaped by Language Use. Cambridge University Press.

Space Particle Network

–>

–>

Network Analysis of Constructional Semantics

Three Levels of Network Analysis

  • Macroscopic Analysis: Examine the properties of the entire graph
  • Microscopic Analysis: Examine the properties of the nodes
  • Mesoscopic Analysis: Examine the groupings of the nodes, i.e., the emergence of the Community in the graph
  1. Siew, Cynthia S. Q., Dirk U. Wulff, Nicole M. Beckage & Yoed N. Kenett. 2019. Cognitive network science: A review of research on cognition through the lens of network representations, processes, and dynamics. Complexity 2019. 1–24.

Macroscopic Analysis

Common Structures in Social Networks

  • Scale-free:

    • a pattern in which there are usually only a few nodes with high degrees and many more with low degrees.

  • Small-world:

    • a tendency in which a network usually consists of several small communities or clusters where the within-community edges are much stronger than across-community ones.
  1. Barabási, Albert-László & Re ́ka Albert. 1999. Emergence of scaling in random networks. Science 286(5439). 509–512.
  2. Watts, Duncan J. & Steven H. Strogatz. 1998. Collective dynamics of ’small-world’networks. Nature 393(6684). 440–442.

Macroscopic Analysis of SPC

  • Small-world: Semantically connected nodes form small clusters in the network, suggesting the emergence of semantic fields of the collexemes
  • Scale-free: Only a few linguistic units (central nodes) in the system are frequently connected to other linguistic units, likely to be important exemplars, or “cognitive reference points” (Diessel 2019: 34)

Microscopic Analysis

Node-level Metrics

  • Local Clustering Coefficient: a metric that indicates the extent to which the neighbors of a node are interconnected – namely, whether a node’s neighbors are also neighbors of each other.
  • Centrality：a metric that characterizes the importance of a node beyond its immediate neighbors.
  1. Also known as Local Transitivity.
  2. Common metrics for node centrality include: Betweenness, PageRank, Authority, Closeness.
  • Construction Nodes: The local clustering coefficient values indicate the degrees of semantic coherence of their collexemes.

  • Lexical Nodes (Landmark): The centrality values indicate which nodes are more central, thus more likely to be an exemplar of the sub-network.

Mesoscopic Analysis

  • In network science, we can identify small clusters as communities using the algorithms of community detection.

These communities arising from the construction network may indicate the emergence of specific semantic fields.

Conclusion

Thank you!

References：

  1. Barabási, Albert-László. 2016. Network science. Cambridge: Cambridge University Press.
  2. Diessel, Holger. 2019. The grammar network: How linguistic structure is shaped by language use. Cambridge, UK: Cambridge University Press.
  3. Chen, Alvin Cheng-Hsien. In press. Words, constructions and corpora: Network representations of constructional semantics for Mandarin space particles. Corpus Linguistics and Linguistic Theory 19(1). https://doi.org/10.1515/cllt-2020-0012. (Supplementary Materials)

20 August, 2021, ICCG11, Antwerp

alvinchen@ntnu.edu.tw  |  alvinchen.myftp.org