Using Topic Modeling as a Semantic Technology: Examining Research Article Claims to Identify the Role of Non-Human Actants in the Pursuit of Scientific Inventions

Featured Application

The study demonstrates the possibility of adopting a Natural Language Processing-based analytics approach—topic modeling—as a semantic technology that could enhance the application of actor-network theory in the identification of key non-human actants affecting the successful pursuit of scientific inventions. The research design provides an example of an actionable framework enabling the automation of the processing, categorization, and analysis of textual data, uncovering latent meanings, and enhancing the comprehensibility of the key insights.

Abstract

Actor-network theory (ANT) represents a research paradigm that emerged within science and technology studies by explicitly focusing on the contingency of scientific inventions and the role of non-human actants in the invention course of action. The article adopts an ANT perspective to focus on the invention of Sub-Wavelength Grating (SWG) photonic metamaterials by the members of a research group in the National Research Council (NRC) of Canada. The results are based on unstructured interviews with the key inventor and two domain experts as well as on textual analysis (topic modeling) of the contributions and novelty claims in the corpus of research articles by the NRC group crafting the concept and potential applications of SWGs in the photonics domain. Topic modeling is a type of statistical modeling that uses unsupervised machine learning to identify clusters or groups of similar words within a body of text. It uses semantic structures in texts to understand unstructured data without predefined tags or training data. Adopting topic modeling as a semantic technology allowed the identification of two of the key non-human factors or actants: (a) photonics design and simulations and (b) the fabrication techniques and facilities used to produce the physical prototypes of the photonics devices incorporating the invented SWG waveguiding effect. Using topic modeling as a semantic technology in ANT-inspired research studies focusing on non-human actants provides significant opportunities for future research.

1. Introduction

Actor-network theory (ANT) represents a research paradigm that has already been applied to science and technology studies and innovation management research by focusing on the contingency of the transition from invention to innovation [1]. The uniqueness of ANT’s perspective is rooted in the adoption of a fundamental principle of symmetry between human and non-human agents, which allows for the inherent contingency of real-life innovations [2,3,4]. The principle of symmetry refers to the idea that all entities within a network, whether human or non-human, should be analyzed and described using the same premises, meaning that all actors are considered equally important and active in shaping the network’s dynamics.

Taking an ANT approach to the study of scientific inventions having the potential to be adopted in a real-life context may provide new insights into how human and non-human actors (researchers, businesses, non-human actants such as patents, design tools, or fabrication equipment) negotiate the development and the adoption of an emerging invention [5]. The focus on the influential role of non-human actants is particularly relevant because of its unique explanatory potential [6]. In ANT, the terms ‘actor’ and ‘actant’ are often used interchangeably but also have some nuanced differences. Both terms refer to entities that can act within forming or emerging networks. At the same time, ‘actor’ is used in a more general sense to denote any participant with agency, while ‘actant’ is a more specific term borrowed from semiotics to describe any entity, regardless of its nature, that could be granted activity by others in contributing to actions within a forming network. This distinction goes beyond ANT’s usual focus on the symmetry between human and non-human actors in terms of their potential impact on novelty and has enabled further ANT research focusing on refining the symmetry principle by exploring (a) the unique role of personal factors (e.g., human persistence, endurance, ingenuity) and their potential to weaken or reshape the supposed symmetry between human and non-human actors by taking the initiative and driving the course of action in a way that could impact the innovation outcomes; (b) the extent to which the quality of the coupling, entanglement or synergy between human actors and non-human actants can affect an emerging invention [7,8,9]. The objective of the present research study is to expand the applicability of these later insights into the study of scientific inventions and their potential transition to real-life innovations.

The article focuses on a case where there is a clear separation between the invention and adoption phases of an invention. This is the invention of Sub-Wavelength Grating (SWG) photonic metamaterials by members of the Silicon Photonics Research Group in the Advanced Electronics and Photonics Research Centre at the National Research Council (NRC) of Canada. The results of this study were part of the Master Thesis of Samantha Sieklicki (“Using Actor-Network Theory to Examine the Invention-to-Innovation Transition of SWG Photonic Meta-Materials”, Master of Applied Science Thesis, Engineering, Technology Innovation Management, 2024, Carleton University, Ottawa, ON, Canada: DOI: https://doi.org/10.22215/etd/2024-15963). The invention of the SWG metamaterials was orchestrated by one of the group members, who is a Principal Research Officer at NRC (in this study, we will be referring to him as the PRO). The term ‘orchestration’ could be misleading since it suggests some degree of linearity and predictability in the pursuit of the invention. It is, however, highly adequate in highlighting the unique personal ability of the PRO to engage, align, and synergize human, technological, and research resources across multiple countries and research institutions. The details of the invention were communicated and demonstrated by an evolving team of co-authors in multiple publications and the filing of two patents in the period between 2006 (the first publication claiming the use of SWGs as a new waveguiding mechanism [10]) and 2018 when the PRO received the NRC 2018 Excellence in Research Award, initiated and coordinated the publication of a review article in Nature [11]. This is the foremost international scientific journal in the world. The PRO was recognized as the lead inventor of a new type of optical waveguide mechanism based on the application of SWGs (US patent 7,376,308 B2, 2008 and US patent 8,503,839 B2, 2013). A photonic metamaterial is engineered by a unique one-step lithography approach, which allows the synthesis of a new meta-material. This is a type of material that does not exist in nature but has properties that are highly beneficial for the design and fabrication of advanced micro-photonic and nanophotonic devices. The PRO and his colleagues demonstrated the implementation of the new optical waveguide principle in multiple devices such as fiber-chip couplers, crossings, multiplexers, ultra-fast optical switches, athermal waveguides, sensors, polarization rotators, transceiver hybrids, and ultra-broadband interference couplers. The impact of this invention has been widely recognized, and subwavelength structures are now considered key building blocks for the next generation of integrated photonic circuits.

The preliminary details of the invention of the SWG photonic metamaterials suggest the adoption of an ANT-driven methodology shaped around the following research questions:

• Who were the key actors that affected the pursuit of the SWG photonic metamaterial invention, and how did the research group form?
• Can topic modeling of the contributions and novelty claims articulated in the published research articles help identify some of the key non-human actants in the invention of the SWG photonic metamaterials?
• How did the entanglement between human and non-human actors affect the course of action leading to the invention?

The adoption of the ANT perspective in the present study has several distinctive features. First, it focuses on the traditional description of the contingency and the unpredictability of the invention process but enhances its scope by highlighting the role of the personal factor—the perseverance and ingenuity of the key inventor—in driving the course of action and affecting the innovation outcome. This enhancement refines ANT’s traditional understanding of an unconditional symmetry between human and non-human actors [7,8,9], which is highly relevant to the context of real-life scientific inventions and needs to be studied in more depth.

Second, our study demonstrates the possibility of adopting a Natural Language Processing (NLP) approach—topic modeling—as a semantic technology that could help identify key actants in the pursuit of scientific inventions. The potential of topic modeling as a semantic technology is enhanced by the specific nature of the corpus of text documents on which it is applied. This is the contribution and novelty claims in the published research articles documenting the gradual shaping of the invention.

Third, the study distinguishes between the traditional explanatory (descriptive) way of applying ANT to examine how a specific invention has actually happened and a more proactive (anticipatory) way of applying ANT that could help in the identification of future actors and actants that may affect the ultimate adoption of the invention [5]. The intention is to explore how the explanatory application of the ANT framework could inform the development of insights about the potential future adoption of the invention.

The article is structured as follows. It provides a literature review on ANT in the context of scientific inventions and innovation management, the topic modeling text analytics approach, and its potential to enhance ANT. The method section describes the research approach adopted in answering the initial research questions, the details, and the potential of the topic modeling approach as a semantic technology that could help enhance the application of ANT. The Results section addresses the research questions by examining the authorship of the research articles documenting the SWG photonic metamaterial invention, the analysis of the observations from the interviews with key respondents, and the interpretation of the topic modeling results in light of the specific research questions. In the conclusion, we summarize the key contributions and distinctive features of the research study.

2. Materials and Methods

2.1. Literature Review

2.1.1. Actor-Network Theory (ANT)

ANT is a research paradigm calling for a mixed-method approach to data collection and analysis that includes as many types of accessible data sources as possible and can combine interviews, published textual data, observations, and quantitative statistics. Sieklicki and Tanev [5] identified two major ways of applying ANT to studying innovations: descriptive (or explanatory) vs anticipatory. Thus, researchers should select a specific ANT approach (explanatory vs anticipatory) depending on the context of their specific research case. The descriptive or explanatory approach is the traditional way of using ANT. It is a way of looking back at a case of an emerged novelty or practice with the intention of uncovering a logic of change that goes beyond the role of human actors [4]. In the anticipatory case, the researcher moves away from the role of an observer describing the course of past action to an examiner of the potential actors and active participants in an ongoing course of action that could lead to potentially anticipated future developments [12]. The anticipatory way of using ANT could be particularly useful in studying the transition from invention to innovation in cases where the adoption of an invention and its impact on existing practices is still ongoing [1]. However, its application is significantly more challenging because of its forward-looking anticipatory mode of application.

In the dominant way of using ANT, actors are always shaped by their surrounding connections, and the network is constantly being reshaped by the actors who act on it [4]. ANT actors can be humans, non-human objects, or complex non-human actants such as culture and technology. The actor is a subject/object hybrid with the agency to act but also controlled by the relationships with other actors [13,14]. Limiting actors to humans destroys the symmetry between humans and non-human agents by failing to consider how humans are transformed by the agency of their physical, legal, and cultural surroundings. Selecting actors based on whether they are a source of agency (hence, the term actant) preserves the focus on action, transformation, and novelty, which fits very well in the context of emerging innovations [15].

Latour [4] emphasizes that ANT departs from the traditional understanding of the processes of formation of groups that always precede them [15]. Groups must be constantly maintained, renovated, or rebuilt; otherwise, they cease to exist. ANT’s approach to the dynamics and unpredictability of group formation and agency could be described by its commitment to address several types of uncertainties related to: (a) the need to explore actors’ agencies within a collective; (b) the need to understand the nature of actors’ actions (an actor is what is made to act by others; an actor ‘is not the source of an action but the moving target of a vast array of entities swarming toward it’ [4], (p. 46); (c) the need to inquire about the agency of objects (non-human entities can no longer be considered neutral intermediaries but as silent mediators which modify the relationship among other agents); (d) agencies should not be presented as matters of fact but as matters of concern focusing on making their mode of fabrication and stabilizing mechanisms visible.

In ANT, ‘mediation’ refers to the cooperative work between different actors and actants. It implies that non-humans can participate in developing mediations and producing different assemblages, actions, and effects. Different artifacts and technologies can develop various forms of mediation that can transform, translate, distort, or modify the meanings or agents and elements [4] (p. 39). ‘Translation’, on the other hand, is a dynamic process of negotiation and transformation that binds human and non-human actors (actants) into networks. Translations happen between mediators that generate traceable associations, resulting in emerging networks [4] (p. 108).

The entanglement or synergy between human and non-human actors is of special interest and has been specifically addressed in the ANT literature. For example, Tchalakov suggested introducing the concept of “coupling” to describe the synergy in the inter-action processes occurring in scientific experiments and laboratory life. He considered the emerging relations between researchers and the actant objects they are studying as “heterogeneous couples” [9]. In the context of ANT, coupling can be defined as a process by which scientists gradually emerge as “spokesmen” for the nonhuman agents they are studying, their messengers to the rest of the audience. In essence, heterogeneous couples can be seen as the “constituent elements” of a scientific laboratory. They could also be seen as elementary “micro-communities”, which should often be considered beyond the relationships between the scientists and the specific nonhuman agents they are examining. The main question here is about what supports and stabilizes such couples because some of the actors may not (or may not be able to) speak, thus concealing essential layers of what is happening in the inner life of the micro-community. Getting deep into the nature of a heterogeneous couple often means breaking standing relations with other humans and a disintegration of previously stabilized social entities. One of the key messages of Tchalakov’s heterogeneous coupling theory is that the quality and intensity of the coupling between human and non-human actors affect the translation outcomes.

A later development of ANT that could be highly relevant for the present study is the attempt to account for the role of the personal factor in breaking the symmetry between human and non-human actors [7,9]. The concept of ‘asymmetry’ that was introduced by Kapriev and Tchalakov [8] does not assume the pre-existence of a privileged and ontologized subject of action. The reason for introducing such asymmetry is the fact that in every practical situation, there is someone or something that is the first to initiate a change, and by being first, this someone or something is in a stronger position to influence the specific situational dynamics of any resulting translation. Latour himself insists that ‘ANT is not […] establishment of some absurd “symmetry between humans and non-humans”’. He explains that ‘to be symmetric, for us, simply means not to impose a priori some spurious asymmetry among human intentional action and a material world of causal relations’ [4] (p. 76). The important point made by Kapriev and Tchalakov [8] (see also [7]) is that the suggested asymmetry is equally applicable to both human and non-human agents. Another dimension of the asymmetry arises from the importance of the personal factor in distinguishing between different actors without ascribing an unconditional priority to the human actor as a Cartesian subject. This aspect of asymmetry is based on the personal uniqueness and specificity of the activities, which may affect the way a specific change is actualized.

ANT has traditionally focused on examining real-life practices, such as people’s everyday actions, activities, and behaviors [16]. In this sense, it encourages a direct involvement of the researcher that may remind of action research methodological setups. The methodological orientation to practices underpins ANT’s sensitivity to heterogeneity and multiplicity since practices are inherently heterogeneous and multi-layered. “Studying social life via the method of closely tracing practices involves acknowledging the disparate elements that routinely enter into the complex and active processes by which people give meaning to the world, their activities, and their lives” [16] (p. 113). However, in many cases, a researcher does not have the luxury of direct access to past practices and must rely on written textual records instead. The reliance on textual records may be seen as one-dimensional and, to a certain extent, limited since it offers a filtered version of a living reality centered upon the subjectivity of the human actor. At the same time, texts can be seen as windows on practices that could provide a valuable picture of what has been found as tangible, stable, and persistent. Thus, in ANT, textual documents such as scientific lab drawings, patents, and research articles are intermediators [17], documenting, explaining, inspiring, and driving change [18].

In summary, some of the key aspects of ANT that help to better investigate the emergence of science and technology inventions could be summarized as follows [15]:

• Human and non-human associations are enacted in practice;
• ‘Following the actors’ helps to better understand the social as a movement of re-association and reassembling of human and non-human elements;
• ANT relies on as many types of accessible data sources as possible and can combine interviews, published textual data, observations, etc.;
• Material conditions matter, and non-human elements can be agents that come into existence in association with other human and non-human agents;
• Non-humans mediate human and non-human interactions;
• Actants are not equal, and their agential uniqueness and specificity may affect the way changes or translation happen;
• Texts are (inter-)mediators that have the ability to transform, translate, and modify but also stabilize the meaning or the actants and elements they are referring to.

2.1.2. Topic Modeling

Topic modeling is a text analytics approach based on algorithms that enable the discovery of latent themes in large collections of text documents [19]. These algorithms identify the key terms associated with the different topics and organize the collection of documents according to their associated topics. The topical organization of the documents allows for the application of various text-mining tasks such as document classification, summarization, trend detection, and corpus exploration [19]. It is important to point out that topic modeling does not require any prior annotations or labeling of the documents. The topics emerge from the analysis of the texts, including the original collection of documents.

The premise of topic modeling is that the documents could be considered as mixtures of latent topics, where each topic is characterized by a distribution over words. Several algorithms have been put forth to build viable topic models. Two of the most representative examples are Latent Semantic Indexing and Latent Dirichlet Allocation. The latter model, often referred to as LDA, has become a standard tool for topic identification [19]. An LDA-based topic model is a probabilistic generative model that has been used widely in recent years with a specific focus on text mining and information retrieval. It has received significant attention and gained widespread interest among researchers in many research fields, including management studies [20].

The purpose of LDA is to determine the representation of a fixed number of topics in a corpus of documents and, given this number of topics, define the topic distribution within each of the documents in the corpus. The way LDA learns how topics and documents are represented could be intuitively summarized as follows [21] (Figure 1).

First, the researcher selects the number of topics to be discovered. This step is similar to the first step in cluster analysis, when the researcher needs to specify the number of clusters to be considered. Once the number of topics is selected, the LDA algorithm goes through each of the words in each of the documents and randomly assigns each word to one of the topics. The result of this step is a first representation of the topics in terms of words (specific distribution of the words in each topic) and a representation of the documents in terms of topics. It is important to point out that the assignment of words to topics is carried out in a purely random manner. In this sense, this initial representation is not accurate and needs to be further improved. To improve the initial (inaccurate) representation, the LDA algorithm performs multiple iterations by re-assigning each of the words to the topic that is most highly associated with it: (a) for every single document, it calculates the percentage of words within it that were assigned to a particular topic; (b) for each word in the document, it calculates the percentage of times that particular word has been assigned to a particular topic across all documents; (c) for every word, multiplies the values obtained in (a) and (b) to obtain the probability (degree) of its association to a particular topic; (d) for every word, compares the values obtained in (c) to determine the set of words which are most highly associated with every single topic (this is how the word constitution of each topic is defined); uses a similar logic to determine the set of documents which are most highly associated with every single topic. The key outcomes of the LDA approach are the combination of words and the set of documents that are most highly associated with each of the topics (Figure 1). These results need to be further analyzed to articulate the theme associated with each of the topics. This is performed by the researcher, who needs to go deeper into the interpretation of the topic by examining the most highly associated documents and identifying specific themes emerging in association with each of the topics.

2.1.3. Topic Modeling and ANT

There are a few cases of publications that have combined topic modeling and ANT, particularly in analyzing networks of actors and emerging topics in various contexts. For example, Celikyilmaz et al. [22] suggested an actor-topic modeling approach to examine literary narratives. The proposed Actor-Topic Model (ACTM) was developed to analyze social relations within literary texts by combining topic modeling with some of the key principles of network analysis. This approach identifies actors (characters) in a narrative and maps their interactions based on the topics emerging from their dialogues. Following a topic modeling logic, it treats dialogues as distributions over topics and links actors based on the topical similarity of their conversations. By doing so, it constructs a network of relationships between characters without requiring any preliminary annotated data. The ACTM method exemplifies how topic modeling can uncover hidden relationships and patterns within networks, aligning with ANT’s focus on tracing connections between human and non-human actants [22].

Hellsten and Leydesdorff [23] provide another example of the integration of ANT and textual analysis by using topic modeling to analyze actor-topic networks in online debates on the former Twitter (current X) social platform. They examine co-emerging topics and actors to visualize and understand the dynamics of online discussions. By mapping how topics are distributed among different actors and how these actors interact within the network, this method aligns with ANT’s emphasis on the interplay between material (e.g., tweets) and semiotic (e.g., topics) elements in forming networks.

Studies such as the ones by Celikyilmaz et al. [22] and Hellsten and Leydesdorff [23] demonstrate how topic modeling can complement ANT by providing a computational means to identify and analyze the relationships between actors and the topics they engage with, thus enriching the understanding of emerging network dynamics.

2.2. Research Method

2.2.1. Research Design

This study adopts the ANT perspective to explore the outcomes of a research project driven by a Principal Research Officer (PRO) from the Advanced Electronics and Photonics Research Institute at the National Research Council (NRC) of Canada. The project consisted of the conceptual development and experimental validation of a new optical waveguiding principle through the engineering of a new type of photonic metamaterial (a material that cannot be found in nature but only developed by engineering the material composition and the geometry of specific nanophotonic structures). The research project has a well-defined duration from 2006 (first publication) to 2018. The year 2018 marks the publication of an invited review article in the Nature [11] journal, which, together with two previously granted patents and an NRC Excellence Research Award for the PRO, provides evidence for a publicly recorded recognition of the invention and the achievement of the inventors. Thus, the year 2018 marks the formal existence of an invention and the potential beginning of its adoption phase. Delimiting the research study period to 2018 is part of the methodological setup, which requires a specific end of the study period. However, ending the study period in 2018 should not lead to the misleading impression that research on SWG photonic metamaterials has been completed. A recent review paper by the NRC research group [24] demonstrates that the field has grown significantly stronger since 2018. In addition, the group has recently discovered a new type of metamaterial waveguide known as the Huygens waveguide [25].

The study is based on data that were collected from four different sources: (a) multiple formal and informal conversations with the PRO; (b) formal interviews and follow up conversations with two domain experts who are deeply familiar with the photonics technology world and the research and development context of the Silicon Photonics Research Group in the Advanced Electronics and Photonics Research Centre at the NRC of Canada (the two domain experts were substantially involved in the local photonics community and the photonics industry, including the opportunity to develop direct personal impressions from the research activities and management of the group, the research qualities and achievements of the PRO and other group members); (c) public information at the NTC website and other relevant online sources that have provided valuable insights about the case under study; 22 articles published in the period between 2006 and 2018 that included the PRO as a co-author and discussed different aspects or applications of SWG photonic metamaterials.

The formal interviews and the informal conversations with the respondents were driven by the logic of the key research questions. The interviews were held in person and took approximately 2 h each. The two interviews with the two domain experts were also recorded with the Zoom online meeting platform (https://www.zoom.com/en/products/virtual-meetings/), providing the opportunity to automatically transcribe the conversations and use the transcripts as a basis for further elaborations by the two researchers. The discussions resulted in the identification of key insights relevant to the research questions and the formulation of follow-up questions as part of informal Q&A sessions with the key respondents. The informal sessions were shorter (10–15 min) and were mostly performed by phone. Some of the questions were also replied to by e-mail.

The use of the 22 articles as a data source deserves special attention. First, these were all the articles that were published by the PRO and his co-authors on the subject of SWG photonic metamaterials. The 22 articles were identified by examining the publicly available complete list of publications of the PRO and crosschecking with the references provided in the invited review article in Nature [11]. Second, as was already pointed out, in ANT, scientific publications play a crucial role as actants that can influence the direction of research, document progress, shape scientific debates, and impact funding decisions. They contribute to the stabilization of networks by documenting and legitimizing scientific novelty and contribution claims. At the same time, examining the temporal development of such claims helps in following unexpected changes and new developments. Publications help transform complex scientific ideas into accessible formats that can be understood and utilized by various stakeholders, including competing research groups, policymakers, businesses, practitioners, and the general public [26].

The analysis of the articles focused on two key aspects: authorship and novelty/contribution claims. Both the authorship and the novelty/contribution claims are easily identifiable parts within the body of the articles. The novelty/contribution claims can be found in the Abstracts and the Introductions of the articles, but most importantly, in the Conclusions, which is the usual place for the authors to summarize what they have completed and explicitly articulate the key unique findings, the novelty of the approach, and the contributions to the research domain and the potential practical applications of the new findings. The two co-authors adopted a parallel examination process, resulting in the identification of 104 novelty/contribution claims and shaping a corpus of text documents to be used in the topic modeling analysis (Figure 2).

2.2.2. Using Topic Modeling as a Semantic Technology to Examine Novelty and Contribution Claims

The analysis of the content of the research articles was based on performing a topic modeling analysis [19,20,21] on the corpus of 104 extracted paragraphs that were identified as novelty claims and research contributions. Each claim paragraph was considered an individual text document. We have used the text analytics capabilities of Orange Data Mining software (version 1.31.1, https://orangedatamining.com/) and adopted the most popular topic modeling algorithm, LDA [19] (Figure 3). The advantages of using the Orange Data Mining tool are that it is a well-documented, well-maintained, and reliable open-source tool.

Figure 3 illustrates the logic of the topic modeling process. The uploaded corpus of text documents is pre-processed by the Preprocess Text widget, which splits each text into smaller units (tokens), filters them, runs normalization such as stemming and lemmatization, and removes stopwords. These are common, frequently occurring words that are removed from the text before the analysis because they are considered to have little semantic meaning and can obscure the underlying topics within a document. Removing such stopwords is an important (and sometimes time-consuming) step that allows the model to focus on more informative words that help identify more distinctive topics.

The LDA algorithm (selected in the Topic Modeling widget) considers every pre-processed text document (novelty/contribution claim) as a mixture of topics and every topic as a mixture of words. Words can be shared between topics, and the topics can be shared among documents. The LDA algorithm identifies combinations of words that are semantically interrelated and tend to appear together across different documents. The combinations of words help the identification of the specific themes that are latently present in the corpus. In addition, LDA organizes the corpus by clustering the documents corresponding to each topic. The claim paragraphs clustered in each topic are ranked in terms of the degree of their association with it. A closer examination of the topical organization of the paragraphs enables the interpretation of the overall theme and the labeling of the topics [21].

Figure 4 shows a snapshot of the Data Table shown in Figure 3. The first three columns are the same as the ones shown in Figure 2. The last five columns (Topic 1, Topic 2, …, Topic 5) show the degree of association of each claim on the left with each of the five topics. The claims are ranked in terms of the degree of their association with Topic 1. Selecting the claims that have a degree of association higher than 0.6 provides the list of claims that are used for the interpretation of Topic 1. The analysis follows the same logic to identify the claims that are most highly associated with the rest of the topics. The research design provides an example of an actionable semantic technology framework enabling the automation of the categorization and processing of textual data to uncover emerging latent themes and enhance its overall comprehensibility.

3. Results

3.1. Examining the Authorship of Scientific Publications

We have adopted the ANT research philosophy to examine the authorship of the 22 articles identified in the methodology section. The results of this analysis contribute to answering the first research question: Who were the key actors that affected the pursuit of the SWG photonic metamaterial invention, and how did the research group form? 

Table 1. A visual illustration of the contribution of different authors to the 22 articles published between 2006 and 2018. The table shows the countries of origin of the authors as well as their participation as co-authors in different years of publication.

N	Author	Country	2006	2007	2010	2011	2012	2014	2015	2016	2017	2018
1	Cheben	Canada	C	C	C	C	C	C	C	C	C	C
2	Janz	Canada	C	C	C	C	C		C
3	Xu	Canada	C	C	C		C	C	C	C
4	Densmore	Canada	C		C		C
5	Schmid	Canada		C	C	C	C	C	C	C	C	C
6	Lapointe	Canada		C	C	C	C		C
7	Post	Canada		C
8	Bock	Canada			C	C	C		C
9	Delage and Hall	Canada			C		C
10	Lamontagne	Canada			C				C
11	Glesk	UK				UK
12	Halir, Molina-Fernandes, and Ortega-Monux	Spain					SP	SP	SP	SP	SP	SP
13	Wanguemert-Perez	Spain					SP	SP	SP	SP		SP
14	Aers and Maese-Novo	Canada and Spain					C and SP
15	Alonso-Ramos	Spain						SP	SP	SP	SP
16	Perez-Galacho	Spain						SP			SP
17	Xiong and Ye	Canada						C
18	Wang	Canada							C	C	C
19	Dado	Slovakia							SL		SL	SL
20	Benedikovic and Vachon	Slovakia and Canada							SL		SL
21	Painchaud and Picard	Canada							C
22	Khokhar et al.	4 UK and 2 Spain								UK and SP	UK and SP	UK and SP
23	Duran-Valdeiglesias et al.	3 Spain and 1 France								SP and FR
24	Baudot et al.	7 France, 4 Can, 1 Spain, and 1 USA									FR, C, SP and US
25	Atwater et al.	4 UK, 3 Czechia, 1 Can, 1 Slovakia, 1 USA										UK, CZ, C, SL and US

provides a visual representation of the involvement of different authors during the study period between 2006 and 2018. The first author (P. Cheben) is the PRO and is part of all 22 publications. The letters in the cells corresponding to each year of publication indicate the country of origin of the co-authors, which corresponds to the country indicated in the second column of the table. A closer examination of Table 1 leads to several interesting findings that demonstrate the need for the adoption of the ANT research perspective. These findings could be summarized as follows.

• It is impossible to identify a stable group of co-authors. The composition of the group is continuously changing. There are multiple contributors but no stable group of authors (the key formal actors in the shaping of the invention). The only “stable” actors who are consistently present are the PRO and one of his colleagues (Schmid), who succeeded in engaging the contributions of researchers from different countries at different times. The variety of countries of authors’ origin that can be seen in Table 1 illustrates the diversity of contributions and actors at different times. The interviews with the two domain experts who had direct and intimate exposure to some of the inventors and the overall context of the invention have indicated that one of the key group formation factors in the pursuit of the invention was the person of the PRO who was able to build on his current and pre-existing personal and institutional relationships, both locally and across different countries, to engage other researchers who can substantially contribute to developing further PRO’s initial idea.
• One should also acknowledge the constructive role of the emerging synergies between specific actors. For example, the PRO mentioned that without incidentally meeting one of the domain experts interviewed as part of this study, who introduced him to the OptiFDTD simulator, the SWG coupler idea may not have been efficiently validated. The initial collaboration between the PRO and the domain expert resulted in a joint publication [27]. Another interesting case is the last one (Row 25). It refers to the 2018 publication in Nature, which includes the co-authorship of one of the competitive groups in the USA [11]. The PRO has invited the leaders of a competing research group from another country to make a more valuable invited review article. This is just an illustration of the unique personal ability of the PRO to cooperate and engage others in pursuing a common valuable goal. This finding supports two points made by Kapriev and Tchalakov [8] and Tchalakov [9], respectively: first, the unique personal characteristics and ingenuity of an actor could be highly influential in shaping the course of action; second, the quality of the entanglement or emerging synergies between specific actors could be highly beneficial.
• A deeper analysis of the involvement of specific co-authors could provide additional insights into the diversity and unexpectedness of other key contributions to the invention. For example, according to the PRO, day-to-day technical discussions with Janz (row 2) and Xu (row 3 in Table 1) were critical to bringing the invention into practice. Janz and Xu were part of the core team that initiated the research project and co-authored most of the articles published between 2006 and 2016 but did not contribute to later publications. On the other hand, if PhD student Bock (row 8 in Table 1) had not been involved in the period between 2010 and 2012, the research project would not have developed the way it did. At that time, it was only the PRO and his colleague Schmid who were driving the SWG research activities, struggling to invest their efforts across several other research projects (the SWG photonic metamaterials project has never received dedicated research funding). The specific involvement of Schmid (see row 5 of Table 1), a Senior Research Officer at NRC, should be explicitly emphasized. Interestingly, Schmid was not one of the co-authors of the first article describing the novel SWG waveguiding mechanism [9]. However, his substantial involvement in the research activities and in the publication of all future articles shows the emergence of a stable partnership with the PRO that enabled the constructive engagement and supervision of young Spanish researchers who contributed substantially to the second half of the invention period (see rows 12–16 of Table 1 which show a continuous and substantial involvement of Spanish co-authors from 2012 on). More importantly, Schmid’s extraordinary fabrication skills were critical for the experimental validation of the invention. His fundamental technical ingenuity complemented the photonic design and modeling expertise of the PRO and his colleagues, resulting in the fabrication of the very first SWG waveguide structure. We can identify the emergence of a constructive synergy or entanglement within the initial team of NRC researchers that became a sustainable factor driving the successful validation of the scientific invention. Most interestingly, this synergy amplifies the impact of two of the most important non-human actants: (a) world-class micro- and nano-fabrication capabilities and (b) micro-photonics design and simulation expertise.
• The involvement and coordination of the contributions of research collaborators and PhD students from Spain (“the Spaniards”) were critically important because, according to the PRO, without their involvement, the project would never have reached its culmination in 2018. More specifically, the collaboration with Molina-Fernandez’s group at the University of Málaga (row 12 in Table 1) was very important in the second half of the invention period (2012–2018), which focused on designing and demonstrating new SWG-engineered devices. It is not by accident that the Spaniards became the second-largest group of co-authors after the Canadians. The PRO completed his PhD in Physics/Optics in Spain (Complutense University of Madrid) and spent years of professional research work between 1992 and 1997 at the National Institute for Aerospace Technology of the Ministry of Defense near Madrid. He maintained valuable working relationships with his former colleagues, which allowed him to seek cooperation with and support younger researchers who were eager to make a difference in the newly emerging research domain.
• There are two gaps in the years of publication, 2008/2009 and 2013. There are two ways of interpreting these publication gaps. First, as it appears, the SWG photonic nanomaterial research was not part of any initially planned and well-funded research project. In fact, it has never been funded. In this sense, its progress was contingent on the availability of resources, the negotiation with and engagement of new potential contributors who have found it valuable to engage and contribute. Second, the two gaps in the publication years correspond to the filing of two patents that were directly related to the invention. The first patent was filed at the end of 2007 and granted at the beginning of 2010. It could explain the lack of publications in 2008 and 2009. The second patent was filed at the beginning of 2012 and granted in the second half of 2013. The two patents refer to two different applications of the SWG waveguiding effect.
• Patents are completely different documents compared to research articles since they, after being granted, address issues specifically related to the potential future adoption of an invention. They play the role of an emerging actant directly affecting any potential future adoption. This is another finding contributing to the answer to the first research question (the Who question). The filing of the patents by the PRO and his co-authors demonstrates an anticipation of the future adoption of the invention of SWG photonic metamaterials. The analysis of the timing of the patent filing shows that the application of ANT in its traditional explanatory mode of inquiry could inform its application in an anticipatory mode, which aims at identifying potential future adoption actors [5]. Interestingly, an announcement on the NRC website indicates that the subwavelength coupler, which is based on the SWG metamaterial invention, has already been adopted “for volume optoelectronic chip manufacturing by industry players such as IBM (U.S.A.), Global Foundries (U.S.A.), and CEA-Leti (France)”. The same announcement acknowledges the recent naming of the PRO as an International Fellow by the Royal Academy of Engineering, referring to him as the “founder of metamaterial integrated photonics”. (https://nrc.canada.ca/en/stories/nrc-photonics-researcher-named-international-fellow-royal-academy-engineering, accessed on 16 March 2025).

We can see that the analysis of the authorship, timing, and the nature of the published documents offers some direct input into answering the first research question. The scope of insights goes beyond the identification of the proactivity of the human actors (the authors) by emphasizing two specific additional aspects. On the one hand, this is the role of patents as actants in the invention course of action but also as actants affecting and enabling the future adoption of the invention. This finding demonstrates the validity of one of the distinctive features of this study, i.e., the opportunity to use the traditional explanatory mode of application of ANT in a way that could inform its anticipatory potential [5]. On the other hand, our observations support the initial assumptions that the unique personal characteristics of a human agent (perseverance, dedication, ingenuity) could result in a productive entanglement or synergy between specific human or non-human actors and play a key role in the pursuit of an invention [7,8,9]. This finding goes against the traditional ANT understanding of a generalized symmetry between the capacity of human and non-human actors to affect change. There is no question about the need for the adoption of such a symmetry principle, but, as we have seen, the personal factor can have a special role in driving and making inventions happen, a role that could weaken or ‘twist’ the initially assumed symmetry in a way that could provide ANT with even more explanatory power.

3.2. Topic Modeling of Novelty and Contribution Claims

We have explored the possibility of different numbers of topics that could describe the latent themes emerging from the corpus of 104 claims derived from the 22 research articles. Each topic comes with 10 of the most frequent words that characterize the text paragraphs of the research claims associated with it. Below, we show the results (the most frequent words and the number of claims associated at least 60% with the most frequent words for a given topic) of three different topic models, including five, four, and three topics (Figure 4 shows that 0.6 is a natural threshold for the selection of the most associated claims associated with Topic 1). The topics are labeled in a way that could distinguish between different topic models and different topics within a specific topic model. For example, Topic 5T2 refers to the second topic in the five-topic model. The text in the brackets after each topic label suggests a brief, intuitive interpretation of its content based on the combination of the most frequent words and the content of the most highly associated claims.

3.2.1. A Topic Model Including Five Topics

• Topic 5T1 (photonics design/simulation tools and techniques): photonic, FDTD calculations, calculations, FDTD, operation, metamaterial, simulations, advantage, microphotonic (18 claims).
• Topic 5T2 (photonic metamaterial properties): microphotonic, novel, compact, metamaterials, procedure, transmission, material, photonic, simulation, implement (16 claims).
• Topic 5T3 (SWG metamaterial applications): SWGs, operation, photonic, compact, superior, unique, metamaterial, advantage, applications, impact (18 claims).
• Topic 5T4 (novel photonic devices and structures): photonic, operation, simulation, SWGs, Mach, Zehnder, Mach Zehnder, novel, structure, all-optical (22 claims).
• Topic 5T5 (fabrication equipment, techniques, and capabilities): structure, remarkably, UV lithography, UV, simulations, lithography, composite, material, compact, ultraviolet UV lithography (22 claims).

3.2.2. A Topic Model Including Four Topics

• Topic 4T1 (photonics design/simulation tools and techniques; fabrication equipment, techniques, and capabilities): photonic, simulations, FDTD calculations, calculations, FDTD, operation, lithography, metamaterial, UV, UV lithography (21 claims).
• Topic 4T2 (photonic metamaterial properties): compact, structure, novel, microphotonic, metamaterials, material, procedure, compact structure, transmission, applications (23 claims).
• Topic 4T3 (unique advantages of device-based on photonic metamaterials): operation, photonic, remarkably, structure, metamaterial, unique, superior, compact, advantage, impact (23 claims).
• Topic 4T4 (SWG metamaterial applications): photonic, SWGs, operation, Zehnder, Mach, Mach Zehnder, simulation, material, novel, all-optical (29 claims).

3.2.3. A Topic Model Including Three Topics

• Topic 3T1 (photonics design/simulation tools and techniques; fabrication equipment, techniques, and capabilities): simulations, simulation, photonic, lithography, UV, UV lithography, metamaterial, unprecedented, FDTD calculations, calculations (32 claims).
• Topic 3T2: (novel photonic metamaterial applications) photonic, compact, novel, structure, material, microphotonic, Zehnder, Mach Zehnder, Mach, applications (33 claims).
• Topic 3T3: (unique properties of photonic devices based on SWG metamaterials) operation, SWGs, photonic, remarkably, lithography, unique, superior, compact, metamaterial, structure (31 claims).

The examination of the results of the three topic models helps demonstrate the relative stability of the distinctive emerging themes across the different topic models (varying the number of topics is a natural way to test the stability of emerging topics). What is most interesting is that the three topic models explicitly identify two major non-human actants that have substantially influenced the invention: (a) photonics design/simulation tools and techniques (simulations, FDTD calculations, FDTD calculations), and (b) the fabrication equipment, techniques and capabilities (lithography, UV lithography, composite material) used to produce the physical prototypes of the photonics devices incorporating the invented SWG waveguiding effect. The five-topic model includes these two actants as separate topics. The four and three-topic models merge them together (this is normal since the most frequent words in the corpus get distributed across a smaller number of topics). They are, however, distinctive enough to be considered equally important and fundamentally necessary factors that have enabled the emergence of the invention. Photonics design/simulations and experimental fabrication techniques are key elements of the experimental configurations used to demonstrate the concept, the application, and the operation of the initial invention. Photonics design and simulations provide a basis for running virtual experiments, while the fabrication equipment is key for prototyping and proof of concept. Simulations are the starting point that enables the production and validation of real physical prototypes. That is why the first publications focus on simulations while the later ones include the fabrication and testing of physical prototypes.

The analysis below will focus on the results of the three-topic model. The focus on the three-topic model was justified by both qualitative and quantitative reasons. Qualitatively, we have used our domain knowledge to assess the distinctive clustering of the words in the different topics. The examination of the word composition of the three topic models suggests that the emerging themes (including the ones referring to the two non-human actants) are quite intuitive. From a quantitative perspective, we have chosen the topic model with the lowest Perplexity value. This is an intrinsic evaluation score of a topic model that is used to evaluate how well it predicts a sequence of words [28]. The Perplexity value is provided automatically by the Topic Modelling widget of the Orange Data Mining tool and serves as an indicator of the topic model’s predictive performance on novel or unseen data. “It mirrors the model’s ability to generalize. A lower perplexity score indicates the model’s confidence and accuracy in predictions, while a higher perplexity score suggests uncertainty and inaccuracy in its predictions [28], (p. 6)”.

3.3. Examining the Novelty/Contribution Claims Associated with the Three-Topic Model

Topic 3T1 refers to two major non-human actants in the invention process, photonics design/simulation tools and fabrication methods and techniques. This is an impressive result since it suggests that topic modeling can play the role of semantic technology that could help in structuring authors’ novelty claims in a way that identifies two of the major non-human actants from an ANT point of view. This finding contributes to answering both the first (the Who?) research question and the second research question: Can topic modeling of the contributions and novelty claims articulated in the published research articles help in identifying some of the key non-human actants in the invention of the SWG photonic metamaterials? It should be emphasized that the photonics design/simulation tools and the micro-fabrication methods/techniques are not the only non-human actants in this case. As an example of another non-human actant, one could mention the networking infrastructure of the photonics community in Ottawa, Ontario, which has significantly enhanced the advanced adoption of the design and simulation tools by the PRO and the NRC Silicon Photonics Research group. This is a point that was made by both domain experts who were interviewed as part of this study. It shows that there is a potential for combining the traditional ANT research approach with text analytics tools as semantic technologies to develop valuable insights relevant to the study of scientific inventions. The existence of such potential provides an opportunity to emphasize that topic modeling should be considered not as a mere replacement of traditional human-driven coding approaches but as a complementary addition or alternative that could be highly valuable in cases where there is a significant number of textual documents to analyze.

Topic 3T2 refers to novel applications based on the implementation of SWG metamaterials. Topic 3T3 refers to unique and remarkable features of the principle of operation of devices using SWG metamaterials. These specific interpretations of the three topics were based on the straightforward examination of their most frequent words and their most highly associated claims.

Table 2. Topic 3T1 textual examples referring to photonics simulation and fabrication methods: simulations, simulation, photonic, lithography, UV, UV lithography, metamaterial, unprecedented, FDTD calculations, calculations.

Degree of Association to Topic 1	Claims Associated with Topic 1: Photonics Design/Simulation Tools and Techniques; Fabrication Equipment, Techniques and Capabilities
0.982	An obvious practical advantage of our fabrication method is that AR facets can be produced by using standard lithographic and etching techniques at the wafer level.
0.982	We reported results of numerical simulations of reflection and transmission spectral filters based on Bragg gratings in subwavelength grating metamaterial waveguides. We demonstrated that filters with spectral bandwidths as small as a few tens of picometers can be implemented in silicon waveguides while keeping minimum structural dimensions compatible with deep-UV lithography (>100 nm).
0.969	We have shown both numerically and experimentally that substrate leakage losses in SWG waveguides are related directly to the effective index (or, equivalently, group index), and that this relation is invariant on the specifics of SWG geometry. This finding allows the designer to treat SWG structures abstracting from the BOX thickness constraint, and readily estimate the leakage losses from the calculated effective index using the results.
0.965	We demonstrate the potential of SWG refractive index engineering through the realization of practical functional components. … The coupler principle is based on a gradual modification of the waveguide core refractive index and the corresponding mode size transformation by changing the volume fractions of the Si and SU-8 materials that form the composite waveguide core.
0.958	We have shown, for the first time, that the efficient and widely used CMT formalism can be advantageously applied to Bragg gratings in SWG metamaterial waveguides, providing the coupling coefficients are determined by rigorous 3D tools such as FMM.
0.957	In this Letter, for the first time, to the best of our knowledge, we report on the experimental demonstration of an ultra-directional fiber-chip grating coupler seamlessly fabricated by using 193 nm deep-ultraviolet (deep-UV) optical lithography, a viable tool for large-volume production.
0.904	By exploiting the principle of SWG refractive index engineering in planar waveguides, the coupler was fabricated using a single-etch step process and backside metal deposition.
0.830	We used 3D finite-difference time-domain simulations to minimize loss, crosstalk and polarization dependence.
0.826	We demonstrate the proposed coupling principle on examples of various two-dimensional coupling structures using 2D Finite Difference Time Domain (FDTD) calculations
0.825	We demonstrate the proposed principle by two-dimensional Finite Difference Time Domain (FDTD) calculations of various SWG structures designed for the silicon-on-insulator (SOI) platform
0.664	We demonstrate the use of subwavelength gratings etched into the facets of silicon-on-insulator ridge waveguides as a means of reducing facet reflectivity by the gradient-index effect.

,

Table 3. Topic 3T2 textual examples referring to novel photonics device applications: photonic, compact, novel, structure, material, microphotonic, Zehnder, Mach Zehnder, Mach, applications.

Degree of Association to Topic Two	Claims Associated with Topic 2: Novel Photonic Metamaterial Applications
0.976	We report on the experimental demonstration of ultrafast all optical switching and wavelength down-conversion based on a novel nonlinear Mach-Zehnder interferometer with subwavelength grating and wire waveguides. Unlike other periodic waveguides such as line-defects in a 2D photonic crystal lattice, a subwavelength grating waveguide confines the light as a conventional index-guided structure and does not exhibit optically resonant behaviour. Since the device had no dedicated port to input optical signal to control switching a new approach was also implemented for all optical switching control.
0.976	Our subwavelength grating waveguide is unique in that the structure supports a true lossless mode. By modifying the pitch, width and duty cycle of the subwavelength grating, the effective index of the medium can be engineered locally. This technique provides a means to tailor the effective index, mode profile and dispersion of the subwavelength grating waveguide.
0.975	We demonstrated refractive index engineering in a microphotonic waveguide using SWGs, including implementations in practical components at telecom wavelengths. Our technique circumvents an important limitation in integrated optics, that is, the fixed value of the refractive indices of the constituent materials.
0.973	We introduced a new concept of refractive index engineering in optical waveguides. The waveguide is longitudinally patterned with a subwavelength grating (SWG), consisting of segments of a high-refractive-index core material interlaced with a lower refractive-index cladding material. Since the refractive-index contrast can be changed by simply controlling the grating period, waveguides with different optical parameters can be realized on the same chip.
0.973	In this paper, a new waveguide principle based on the formation of a subwavelength grating (SWG) in a waveguide core is proposed and demonstrated experimentally. In contrast to waveguides based on line-defects in 2D photonic crystal lattices, the light is confined in a SWG waveguide core covered with a cladding material of a lower refractive index, as in conventional index-guided structures. The core is a composite medium formed by periodically interlacing silicon segments with a material of a lower refractive index at the subwavelength scale.
0.964	Fundamental to the excellent performance of our coupler is also an original design of the coupler tip near the chip edge, facing the optical fiber. This design procedure, reported here for the first time, is critical to obtain high coupling efficiency, as are the effective medium synthesis procedure and the quantitative understanding of the effect of coupling to substrate, both presented here for the first time.
0.958	An important novelty in this design is compensation of the index mismatch and loss at the junction between two coupler sections with different grating geometries, i.e., the segmented section and the section with the gaps partially filled with narrow silicon segments.
0.958	In this paper to the best of our knowledge we demonstrate for the first time a practical realization of a novel ultra-fast all-optical interferometric sampling/switching device based on photonic nanowire and subwavelength grating waveguides placed in the Mach-Zehnder geometry capable of all-optical picosecond sampling and switching.
0.958	We experimentally demonstrated and analysed a novel ultrafast all-optical nonlinear Mach-Zehnder interferometric switching device, the MZIS which is based on the combination of subwavelength grating waveguide and wire waveguide. The device is capable of performing all-optical picosecond switching.
0.958	In this review we discuss how bringing metamaterials into optical waveguide technologies and on-chip architectures provides new degrees of freedom to control the flow of light in integrated photonic devices.
0.958	In summary, we have proposed and investigated a novel method to achieve a compact and fabrication tolerant polarization splitter and rotator (PSR) using a subwavelength grating (SWG) structure. The PSR comprises a silicon wire waveguide coupled to a SWG waveguide. The fabrication tolerance is improved by engineering the equivalent refractive index of the SWG waveguide.
0.948	This compact SWG structure is among the smallest yet efficient couplers reported, and further reduction in length appears feasible.
0.948	We propose a novel method to implement a compact and fabrication-tolerant polarization splitter and rotator (PSR) on the silicon-on-insulator platform. The PSR consists of a silicon wire waveguide coupled to a subwavelength grating (SWG) waveguide in an asymmetrical directional coupler.
0.904	By exploiting the dispersive properties of sub-wavelength gratings for the first time, we have designed a directional coupler that exhibits a fivefold bandwidth enhancement compared to conventional directional couplers. Our approach yields a compact device and does not deteriorate the coupler’s phase response.
0.903	We report on the experimental demonstration and analysis of a new waveguide principle using subwavelength gratings. Unlike other periodic waveguides such as line-defects in a 2D photonic crystal lattice, a subwavelength grating waveguide confines the light as a conventional index-guided structure and does not exhibit optically resonant behaviour.
0.901	In conclusion, we have demonstrated that a remarkably efficient AR effect is exhibited by monolithic GRIN SWG structures on SOI ridge waveguide facets, fabricated using standard patterning and etching techniques at the wafer scale.
0.897	To our knowledge, this is the highest efficiency with minimal wavelength and polarization dependence yet reported for a microphotonic coupler. Furthermore, the SWG coupler exhibits a high tolerance to the feature size variations that may arise from limited accuracy of lithography and etching.

and

Table 4. Topic 3T3 textual examples referring to the unique principles of operation of SWG photonic metamaterials: operation, SWGs, photonic, remarkably, lithography, unique, superior, compact, metamaterial, structure.

Degree of Association to Topic 3	Claims Associated with Topic 3: Unique Properties of Photonic Devices Based on SWG Metamaterials
0.983	SWG sensing waveguides exhibit two advantages that are of great practical interest. First, it is clear that these sensitivities can be achieved for a wide range of silicon thicknesses (220–300 nm). This affords a significant design flexibility, especially compared to TM sensing waveguides that operate with silicon thicknesses of at least 250 nm. Second, propagation losses in SWG waveguides have been experimentally shown to be of the order of only 2 dB/cm. Thus, these waveguides may achieve substantially longer interaction length than slot waveguides that exhibit losses of 8–10 dB/cm.
0.983	We developed a new technique for implementing the all-optical switching control of the MZIS. … Note that one MZIS arm is a wire waveguide and the second is a SWG waveguide.
0.978	Here, we leverage the inherent anisotropy and dispersion of a sub-wavelength structured photonic metamaterial to demonstrate ultra-broadband integrated beam splitting. Our device, which is three times more compact than its conventional counterpart, can achieve high performance operation over an unprecedented 500 nm design bandwidth exceeding all optical communication bands combined, and making it one of the most broadband silicon photonics components reported to date.
0.964	Compared to SOI couplers with shallow etch grating region, the present approach has two important practical benefits: (i) both grating coupler and interconnecting waveguide are defined in a single-etch step down to the depth of the BOX layer and (ii) the SWG structure allows to control the coupling strength and apodize the grating profile.
0.958	We show for the first time that SWG waveguides can be utilized to engineer the dimensional dependence of the waveguide mode effective index. We use the unique property of SWGs to implement a polarization splitter and rotator with a substantially enhanced tolerance to waveguide size variations resulting from fabrication.
0.958	The benefit of using the SWG to engineer the mode profile for sensing include the mode delocalization effect and providing the analyte to reach the modal field in the gaps of the SWG structure. More than 25% of the field power is concentrated in the sensing part of the SWG waveguide (superstrate and gaps), compared to less than 7% in a slab waveguide (only the superstrate is sensing). This explains the sixfold enhancement in the bulk and surface sensitivities, as indicated by our results.
0.948	In this paper we present design solutions that eliminate these sources of loss and demonstrate that SWG couplers can outperform inverse taper couplers in terms of both coupling loss and polarization dependent loss.
0.948	Here we present the use of SWGs for facet reflectivity reduction by the gradient-index (GRIN) effect. We use SWGs with triangular teeth, etched into the facet from the top.
0.946	In this work, we demonstrate for the first time the realization of an ultra-directional L-shaped grating coupler, seamlessly fabricated by using 193-nm deep-ultraviolet (deep-UV) lithography. We also include a subwavelength index engineered waveguide-to-grating transition that provides an eight-fold reduction of the grating reflectivity,
0.933	In this Letter, we propose a new grating coupler concept which achieves for the first time, to the best of our knowledge, both broadband operation bandwidth and high coupling efficiency, simultaneously. The coupler operation principle is based on the use of the zero diffraction order, instead of the first order that is typically employed in conventional surface grating couplers. In this specific (zero-order) regime, the radiation angle variation with wavelength is minimized, hence substantially broadening the coupler bandwidth.
0.932	The coupler exploits the principle of subwavelength refractive index engineering which mitigates loss and wavelength resonances by suppressing diffraction effects. The proposed technique allows precise control of the modal field at the coupler tip to optimize coupling efficiency between the optical fiber and the silicon chip, for both TE and TM polarizations simultaneously, which helps minimizing the polarization dependence.
0.931	In this paper, we propose a new approach which, building upon these techniques, exploits the unique advantage of dispersion properties of SWG waveguides. For the first time, a Bragg grating comprising SWG metamaterial waveguide core loaded with lateral segments is proposed an analyzed.
0.831	Here we use, for the first time, the dispersive properties of sub-wavelength gratings to achieve a fivefold enhancement in the operation bandwidth of a silicon-on-insulator directional coupler. This approach does not compromise the size or the phase response of the device.
0.664	We have demonstrated a set of suspended MIR silicon SWG engineered devices.
0.661	Here, we propose a new type of sensing waveguide based on subwavelength gratings (SWGs).
0.661	Here we propose, for the first time, the use of sub-wavelength gratings (SWGs) to design broadband directional couplers.
0.647	An important advantage of our SWG structures is that they can be fabricated with a single etch step.

below show examples of claims that are most highly associated with each of the three topics.

A visual examination of the claims will demonstrate their thematic relevance to each specific topic. It is worth pointing out that the topic modeling results enabled an interpretation that was quite intuitive given the specific nature of the invention. This is another indication that topic modeling could be considered a valuable semantic technology that could enhance the application of ANT in the context of scientific inventions. The method demonstrates an impressive ability to process, categorize, and enhance the comprehensibility of textual data, uncover latent meanings, and facilitate the contextual interpretation of the results.

Finally, the analysis of the temporal evolution of the content of the research claims leads to another interesting observation that is worth highlighting in the context of the adopted ANT methodology. In the first half of the study period (2006–2018), the authors did not refer to their invention as a SWG photonic metamaterial but as a new SWG waveguiding mechanism or principle. They started referring to a ‘novel photonic metamaterial’ in the second half of that period. How do we explain this transition? According to the PRO, they used the term SWG (sub-wavelength grating) in order to clearly distinguish their invention from other micro-photonic structures (e.g., photonic crystals which are based on a different so-called bandgap effect (https://en.wikipedia.org/wiki/Photonic_crystal, accessed on 16 March 2025)), as they believed that the term SWG describes their new structures most accurately. It was only after the photonics research community recognized the novelty and the uniqueness of the SWG waveguiding concept and its related structures, often referred to as on-chip metamaterials, that the inventors transitioned to using the SWG metamaterial terminology. In ANT terms, the new concept needed to stabilize through the publication of articles and the filing of patents that demonstrated its multiple potential applications before the invention could be named as a newly emerging actant in the photonics research and technology world. The role of the patents here is particularly relevant since they frame the ownership of the intellectual property and firmly document the content of the invention. The emergence of SWG photonic metamaterial could be associated with a double reciprocal effect: first, the publications of articles and patents, as well as the experimental validation of the novel photonics designs based on the SWG waveguiding effect enabled the emergence of the new SWG metamaterial; second, the transition to the photonic metamaterial language has marked the actual existence of the SWG photonic metamaterials. One can identify a striking similarity with one of Bruno Latour’s most radical claims that microbes did not exist before Louis Pasteur’s work, meaning that while microbes were always present, their existence as a scientific concept and the understanding of their role as an actant in disease were socially constructed through Pasteur’s research and the subsequent acceptance of his theories by the medical community. (https://direct.mit.edu/posc/article/31/1/9/114165/Bruno-Latour-s-Science-Is-Politics-By-Other-Means, accessed on 16 March 2025). Even though the cases are quite different, the comparison helps in highlighting the value of the ANT perspective in the context of our study.

3.4. Entanglement Between Human and Non-Human Actors

The identification of the two non-human actors (photonics design/simulations and fabrication equipment/techniques) offers an opportunity to discuss the effect of the entanglement between human and non-human actors [8,9]. Before doing that, we should point out that this topic requires a more detailed discussion that would go beyond the scope of this article. However, we could mention that the employment of both non-human actors above required solid professional expertise acquired over many decades. At the same time, the personal predisposition of researchers plays a key role. The interviews with our two respondents have shown that the PRO has been highly proficient in mastering design and simulation tools as well as in linking his deep knowledge of theoretical electromagnetics and optics to the theoretical foundation and design context of photonics simulation tools. In addition, the Silicon Photonics Research group at NRC is known for its world-class expertise in numerical modeling and simulations of micro-photonics devices. This is a factor that could explain the motivation of the PRO to explore the capabilities of new photonics design solutions. He interacted with a research scientist at Optiwave Systems (a photonics design automation company based in Ottawa, ON, Canada) and adopted a newly developed commercial FDTD simulation tool to co-design the new devices employing the SWG waveguiding mechanism. At the same time, he did not hesitate to shift to an alternative design and simulation tool provider when the experimental logic of the research project required him to do so. In addition, the micro-photonics fabrication facilities at NRC are unique in Canada and this part of North America. The Advanced Electronics and Photonics Research Centre has been hosting the Canadian Photonics Fabrication Centre, which operates world-class fabrication facilities that enable PRO’s team to meet the challenge of fabricating the SWG photonic metamaterial prototypes. In this sense, our empirical observations support previous ANT research [8,9], which suggests that the quality of the entanglement between human and non-human actors affects the quality of the translation outcomes. This finding contributes to the answer to the third research question: How did the entanglement between human and non-human actors affect the course of action leading to the invention?

4. Conclusions

In this article, we adopted the ANT perspective to explore the case of the invention of the SWG photonic metamaterials by researchers from the Photonics Research Group in the Advanced Electronics and Photonics Research Centre at the National Research Council (NRC) of Canada. The research questions focus on examining (1) the key human and non-human actors that have affected the pursuit of the invention and (2) how topic modeling of the contributions and novelty claims articulated in the published research articles can help in identifying some of the key non-human actants in the invention; (3) how the entanglement between human and non-human actors affected the course of action leading to the invention. To answer these questions, we have: first, performed unstructured interviews and follow-up discussions with the key inventor and two domain experts who were deeply familiar with the context of both the inventors and the invention; second, examined the authorship of all 22 research publications co-authored by the researchers involved in the pursuit of the invention; third, performed a topic modeling analysis on the corpus of 104 research contributions and novelty claims and interpreted the emerging topics in the context of the ANT perspective.

The study provides specific answers to all three research questions, which are highlighted in the results section. It enhances the scope of the traditional descriptive way of applying ANT by highlighting the role of the perseverance and ingenuity of the key inventor and the synergy between different actors in affecting the successful pursuit of the invention, which refines ANT’s traditional understanding of an unconditional symmetry between impacts of human and non-human actors [7,8,9]. The study distinguishes between the traditional explanatory (descriptive) way of applying ANT to examine how a specific invention has happened and a more proactive (anticipatory) way of applying ANT that could help in identifying potential future actors and directions that may affect the ultimate adoption of the invention [5]. In addition, it demonstrates the possibility of adopting topic modeling as a semantic technology that could help identify key actors and factors in the pursuit of scientific inventions. The adoption of an NLP-based analytics approach only demonstrates the value of text analytics approaches in addition to human-based coding and thematic analysis of text documents.

One limitation of the study is that the unstructured interviews included a relatively small number of respondents and only one of the co-authors, the Principal Research Officer (PRO). The quality and depth of PRO’s insights were highly valuable from an ANT perspective but could be associated with potential data bias. On the positive side, the two domain experts involved in this research were deeply familiar with the context of the research group and the nature of the invention. Thus, their insights were highly beneficial in providing a broader independent perspective in answering the initial research questions. Future studies should explore the opportunity for a broader involvement of a larger number of relevant respondents.

Another limitation is the relatively small corpus of novelty/contribution claims that was subjected to topic modeling analysis. Even though the corpus was limited to 104 research claims, the emerging topic model structure was quite straightforward. There is, however, something more interesting, such as a limitation related to the selection of the specific corpus of text documents. While the topic modeling does a great job of identifying non-human actants in the novelty and contributions claims extracted from the relevant research articles, it identifies only actants that have been explicitly discussed as part of these claims. The point is that there are other non-human actants that are not usually discussed as part of the novelty and contribution claims. Our study offers a basis for future studies that could address these methodological issues.

The overall research design adopted in this article provides an example of an actionable semantic technology framework enabling the categorization and processing of textual data, uncovering latent meaning, and enhancing its comprehensibility in ways that could complement human textual thematic analysis.