This paper discusses the results of the linguoquantitative multiparametric profiling of linguistic constructions, focusing on English 'detached nonfinite/ nonverbal with explicit subject'constructions and adopting cognitive-quantitative construction grammar as a theoretical and methodological foundation. Despite extensive research into the linguistic diversity of the syntactic patterns under analysis, a comprehensive parametrization of their linguistic profiles to identify the determining properties that influence their linguistic behavior in present-day English has yet to be carried out. Thus, the statistical platform R was used to achieve two goals: 1) to perform a linguoquantitative parametrization of the formal properties of English 'detached nonfinite/ nonverbal with explicit subject'-constructions based on the corpus data; 2) to verify the results of the linguoquantitative parametrization in a machine learning model and establish the properties with the most significant capacities to differentiate between the DNF/NVESconstructions beyond the corpus. The results of the study prove that the operationalized parameters (factors/factor values) demonstrate different determinative capacity; thus, the linguistic profiles of the DNF/NVES-constructions are distinguished by high, medium, and low linguistic homogeneity, which allows their classification according to the degree of proximity/remoteness in the constructional network. The operationalized linguistic parameters are quite reliable in differentiating types of the DNF/NVES-constructions beyond the corpus.
The development of modern linguistics, particularly construction grammar, is accompanied
by a discussion about increasing the objectivity of research data and finding ways to
improve research precision [1, p. 149; 2]. As a result, the methodology for analyzing
linguistic phenomena is being refined and statistically reliable tools are being actively
employed to verify scientific theories and hypotheses. Traditional methods of language
analysis are being complemented by innovative quantitative corpus-linguistic
methods [
This study aims to discuss the results of the linguoquantitative multiparametric profiling of linguistic constructions, focusing on English ‘detached nonfinite/ nonverbal with explicit subject’-constructions (DNF/NVES-constructions) and adopting cognitive-quantitative construction grammar as a theoretical and methodological foundation. With this in mind, the following objectives are attained: 1) to perform a linguoquantitative parametrization of the formal properties of English ‘detached nonfinite/ nonverbal with explicit subject’constructions based on the corpus data; 2) to verify the results of the linguoquantitative parametrization in a machine learning model and establish the properties with the greatest capacity to differentiate between the DNF/NVES-constructions beyond the corpus. By integrating quantitative corpus linguistics and machine learning approaches, this study advances understanding of the determining properties that define the linguistic behavior of the analyzed constructions in present-day English.
The parameterization method, widely used in engineering and exact sciences, is also applied in various fields of linguistics, such as linguistic modeling (Yatsenko (2011)), lexicology and lexicography (Boychuk (2011), Ivakhnenko (2016), Kupriianov (2019)), stylistics and genre studies (Romanchenko, Stryi (2022)), analysis of the individual writer’s style (Buk (2021), Davydenko (2014), Tkachenko (2018)), cognitive linguistics (Harmash (2015)), corpus studies (Luchyk, Ostapova (2017)), and forensic linguistics (Azhniuk (2017)). Quantitative corpus-based studies employ parameterization to identify, define, and quantify the essential properties (“diagnostic features” [9, p. 35]) of a linguistic unit.
English ‘detached nonfinite/nonverbal with explicit subject’-constructions ([[AUGwith] [NPthe bats] [XPtaking turns to be the starved victim]]; [[NPheart] [XPthumping]]; [[AUGdespite] NP[oil] [XPbeing the lifeblood of industrial (modern) society]]; [[AUGwhat with] [NPmy three sons] [XPbeing away in the Army]]) as complex clause-level constructions possess several idiosyncratic properties that set them apart from other complex syntactic units. Different aspects of these syntactic patterns in both diachrony and synchrony have been studied from the standpoint of various linguistic approaches and frameworks such as traditional grammar (Stump (1985), Quirk, Greenbaum, Leech, Svartvik (1985), Kortmann (1991)), generative grammar (Riemsdijk (1981), Beukema, Hoekstra (1984), Felser, Britain (2007), Nakagawa (2011)), corpus linguistics (van de Pol (2012, 2014), van de Pol & Petré (2015)), systemic functional grammar (He, Wu (2015), He, Yang (2015)), and construction grammar (Riehemann, Bender (1999), Bouzada-Jabois, Guerra (2016)). In addition, the analyzed syntactic units have been considered in the dimensions of linguotypology (Haff (2012), Hasselgård (2012)), translation studies (Davydiuk (2010)) and discourse structure analysis (Asher, Lascarides (2003)). Although several studies have been undertaken, the linguistic versatility of nonfinite/nonverbal syntactic patterns with an explicit subject in English raises a number of questions that have not yet been finally resolved. Primarily, most research has concentrated on the qualitative rather than quantitative aspects of these units, resulting in a gap in understanding their functional and contextual characteristics. Moreover, a comprehensive parametrization of their linguistic profiles to identify the determining properties that influence their linguistic behavior in present-day English and may reflect speakers’ preferences in categorizing their linguistic experience has yet to be performed.
Cognitive-quantitative construction grammar (CQCxG) is a novel research framework in
cognitive-quantitative grammar studies. This framework triangulates the theoretical and
methodological underpinnings of cognitive-semiotic grammar approaches with analytical
and research tools of quantitative corpus linguistics to investigate general and idiosyncratic
properties of linguistic constructions. Cognitive-quantitative construction grammar
revitalizes the traditional concept of a construction, promoting it to the status of the basic
unit for language representation and analysis. Constructions are conceptualized as holistic
semiotic models, emergent cognitively entrenched symbolic units conventionally used in a
language community, and exhibit pairings of generalized form and meaning/function (plane
of expression and plane of content) [
The most effective analytical and research tool for examining the essential properties of
a construction is a comprehensive methodology for multiparametric constructional
profiling. In quantitative corpus studies, ‘profiling’ refers to the process of establishing
specific linguistic properties at a particular language level based on quantitative indicators
of this property (parameter) realization in the corpus, whereas the ‘profile’ of a linguistic
unit is a set of established quantitative indicators [
Multiparametric profiling is based on the procedure of a linguoquantitative parameterization. The procedure entails identifying and statistically verifying a set of essential linguistic properties (parameters/ factors/ factor values) of the plane of expression (form) and the plane of content (meaning/function) of a linguistic construction that constitute its linguistic profile. Thus, a construction’s linguistic profile is an inventory of its formal and semantic properties (parameters) (morphosyntactic, positional, relational, referential, distributional, syntactic-functional, collocational-collexeme and cognitivesemantic), along with corresponding quantitative indicators obtained through their linguoquantitative verification in corpus data. Linguistic parameters of a construction are realized in linguistic features at a particular language level – factors, which are then manifested in specific language categories – factor values.
Nonfinite/nonverbal syntactic patterns with an explicit subject are considered
syntagmatically and semantically complex clause-level constructions in CQCxG and are
referred to as “D(etached) N(on)F(inite)/N(on)V(erbal) (with) E(xplicit)
S(ubject)”constructions (DNF/NVES-constructions). The argument-predicate structure of the
DNF/NVES-constructions minimally consists of a predicate expressed by a nonfinite
(NF)/nonverbal (NV) phrase (XP) and a subject (the external argument of the
nonfinite/nonverbal predicate) expressed by a (pro)nominal phrase (NP). These
clauselevel constructions are partially schematic, represented by obligatory lexically unspecified
slots [SubjNP] and [PredNF/NV], with an open slot for an augmentor [Aug/ØAug] that in
present-day English is expressed by a limited number of units {AUG: with, without, despite,
what with}. The constructions represent a syntactically independent configuration,
detached from a matrix clause by intonation or a punctuation mark. The morphosyntactic
arrangement of the components is displayed as [[Aug/ØAug][SubjNP][PredNF/NV]] (e.g.,
[Augwith][Subjher eyesNP][PredNVopenAdj] (BNC, GOS); [Augdespite] [Subjdesparate
attemptsNP][PredNFto reviveto-Inf her] (BNC, JYB); [Augwhat_with][Subjdelays][PredNFgetting
startedPII] (BNC, HPP); [ØAug][SubjheartNP][PredNFthumpingPI widely] (BNC, EWH). The
DNF/NVES-constructions constitute a taxonomic constructional network in which
individual constructions are projected onto the network as nodes with different degrees of
schematicity, lexical specification, and productivity [
The quantitative multiparametric constructional profiling methodology is employed to establish the essential properties of the DNF/NVES-constructions that determine their linguistic behavior in contemporary English. Multiparametric profiling of the DNF/NVESconstructions entails linguoquantitative parameterization of their linguistic (formal and semantic/functional) properties that comprise their constructional multiparametric linguistic profiles, followed by verification of the obtained data through a machine learning experiment.
The procedure for linguoquantitative parameterization of a constructional profile is applied
to a research sample of the DNF/NVES-constructions obtained from the British National
Corpus [
In the current study, the primary focus is on the plane of expression (form) of the investigated constructions, while the properties of the content plane need a different methodology. The inventory of formal parameters (factors/factor values) is determined by the linguistic and constructional nature of the constructions under study. The DNF/NVESconstructions, structurally complex clause-level constructions, are distinguished by seven parameters of the plane of expression, which define their morphosyntactic, relational, referential, syntactic-functional, positional, and distributional properties. The inventory of the specified parameters is not exhaustive, but it is sufficient for an objective examination of the linguistic behavior of the DNF/NVES-constructions in contemporary English.
A considerable number of parameters (factors/factor values) specified to describe linguistic profiles of the DNF/NVES-constructions in combination with a large amount of quantitative data cannot be objectively analyzed without using complex statistical procedures and appropriate computer programs for statistical processing of linguistic data. As a result, each linguistic parameter (factor / factor value) is submitted to computerized quantitative verification and subsequent qualitative interpretation.
One of the most widely used analytical tools for quantitative processing of empirical data
in Western corpus-oriented linguistics and usage-based construction grammar is the
statistical data analysis system R (R Development Core Team) [
Parametrization of linguistic profiles of the DNF/NVES-constructions is carried out according to the following algorithm.
Step 1. Operationalization of the parameter by identifying the factors of its linguistic manifestation and defining the values that a particular factor acquires at the appropriate level of the linguistic structure.
Step 2. Quantitative analysis of the realization (frequency) of a particular parameter (factor/factor value) in the corpus.
Step 3. Statistical analysis of the data obtained using multivariate analysis of variance (MANOVA), one-factor analysis of variance (ANOVA), and Tukey's multiple comparison method, quantified with the computer statistical data analysis system R.
Step 4. Interpretation of quantitative indicators and identification of essential parameters (factors/factor values) that determine the degree of proximity/remoteness between constructions in a constructional network.
Step 5. Verification of the linguoquantitative data in a machine experiment to establish factors/factor values with the highest capacity to distinguish between the constructions beyond the corpus.
The application of the algorithm to the factors "Part of speech representation of the
subject" (SubjPOS) of the morphosyntactic parameter and the factor "Register Distribution"
(RegDSTN) of the distributional parameter of the DNF/NVES-constructions has already been
extensively discussed in our prior works [
5.1. Linguistic profiles of the DNF/NVES-constructions: a computerized linguoquantitative parametrization The linguistic profiles of the DNF/NVES-constructions are parametrized through the quantitative verification of 7 parameters (morphosyntactic, relational, referential, syntactic-functional, positional, distributional, and punctuational), manifested in 12 factors and 34 factor values as shown in Table 1. The quantitative data of the operationalized parameters and their respective factors/ factor values retrieved from the BNC are standardized by logarithmization. Subsequently, the null (H0) and alternative (H1) statistical hypotheses are formulated for each of the identified factors:
Н0: The quantitative differences between the analyzed DNF/NVES-constructions (dt-øaug
and any detected quantitative differences are random.
Н1: The quantitative differences between the analyzed DNF/NVES-constructions (dt-øaug
and the differences found are essential.
The formulated hypotheses are tested using multivariate analysis of variance (MANOVA), and the quantified findings are displayed in Table 1.
Punctuation marking (PUNC) Position to Matrix Clause (SentPSN) Coreference with Matrix Clause (CoREFR) Discourse Mode Distribution (DiscMdDSTN) Text Type Distribution (TxtTpDSTN) Syntactic Function to Matrix Clause (FSYN) Syntactic Relation with Matrix Clause (SynREL) accepted accepted
The statistical analysis of the linguistic profiles of the DNF/NVES-constructions did not reveal statistically significant differences in the realization of 3 factor values (Nominative case (Nom), Accusative case (Acc), Full coreference (CorrefFull)) out of 32. The quantitative correlations between these linguistic properties do not differentiate the linguistic profiles of the DNF/NVES-constructions and suggest general regularities of the subject’s linguistic embodiment and the reference relations between the DNF/NVES-constructions and the corresponding matrix clauses.
The one-way ANOVA indicates the existence of differences but does not explain where these differences are most prominent. To solve this issue, the post hoc Tukey test is employed to prevent erroneous rejection of the null hypothesis.
The Tukey’s multiple comparison method detects the DNF/NVES-constructions that exhibit statistically significant differences in the realization of factor values. The Tukey’s test is quantified by comparing the indicators for a specific factor value in pairs of constructions. It enables the establishment, with a 95% confidence level, which linguistic features are determining for specific constructions. The multiple comparison method is used to analyze ten pairs of the DNF/NVES-constructions: 1) dt-what_with-SubjPredNF/NV–cxn and dt-despite-SubjPredNF/NV–cxn; 2) dt-with-SubjPredNF/NV–cxn and dt-despite-SubjPredNF/NV– cxn; 3) dt-øaug-SubjPredNF/NV–cxn and dt-despite-SubjPredNF/NV–cxn; 4) dt-withoutSubjPredNF/NV–cxn and dt-despite-SubjPredNF/NV–cxn; 5) dt-with-SubjPredNF/NV–cxn and dtwhat_with-SubjPredNF/NV–cxn; 6) dt-øaug-SubjPredNF/NV–cxn and dt-what_withSubjPredNF/NV–cxn; 7) dt-without-SubjPredNF/NV–cxn and dt-what_with-SubjPredNF/NV–cxn; 8) dt-øaug-SubjPredNF/NV–cxn and dt-with-SubjPredNF/NV–cxn; 9) dt-without-SubjPredNF/NV–cxn and dt-with-SubjPredNF/NV–cxn; 10) dt-without-SubjPredNF/NV–cxn and dt-øaugSubjPredNF/NV–cxn.
According to the degree of proximity/remoteness by the number of statistically significant differences in the realization of the specified factor/factor values, we distinguish constructional linguistic profiles with high, medium, and low linguistic homogeneity.
The constructions dt-despite-SubjPredNF/NV–cxn, dt-without-SubjPredNF/NV–cxn, dtwhat_with-SubjPredNF/NV–cxn form a group with high linguistic homogeneity, showing no statistically significant differences in the analyzed factors/ factor values realization. The medium linguistic homogeneity group consists of constructions such as dt-øaugSubjPredNF/NV–cxn and dt-with-SubjPredNF/NV–cxn, which revealed statistically significant differences in 6 factor values. The group with low linguistic homogeneity includes two subgroups of the constructions: 1) subgroup dt-with-SubjPredNF/NV–cxn and dt-despiteSubjPredNF/NV–cxn, dt-without-SubjPredNF/NV–cxn, dt-what_with-SubjPredNF/NV–cxn, where significant differences were recorded between the with- and what_with-augmented constructions by 31 factor values, between the with- and despite- augmented constructions by 30, and between the with- and without-augmented constructions by 27; 2) dt-øaugSubjPredNF/NV–cxn та dt-despite-SubjPredNF/NV–cxn, dt-without-SubjPredNF/NV–cxn, dtwhat_with-SubjPredNF/NV–cxn, between which differences in 13, 17, and 20 factor values were registered, respectively.
The statistical analysis of linguistic profiles of the DNF/NVES-constructions
demonstrates that some factors/factor values significantly influence their linguistic
behavior. Determining parameters (factors/factor values) of the plane of expression of the
analyzed constructions define the degree of proximity and remoteness of the constructions.
To validate the obtained results, a machine learning experiment is conducted to establish
the operationalized factors with the most significant capacity to differentiate between the
DNF/NVES-constructions beyond the corpus.
5.2. Linguistic profiles of the DNF/NVES-constructions: a machine-learning model
The use of machine learning technologies alongside traditional statistical approaches in
language study is becoming more prevalent [
Figure 1 displays the results of the distribution of the DNF/NVES-constructions for each factor, where significant differences were found with a one-factor ANOVA. All data are best divided along the first two axes LD1 and LD2, simplifying the graphical representation of information and enabling result comparison.
Model accuracy = 0,71
• – dt-despite-SubjPredNF/NV–cxn, • – dt-what_with-SubjPredNF/NV–cxn • – dt-with-SubjPredNF/NV–cxn, • – dt-øaug-SubjPredNF/NV–cxn, • – dt-without-SubjPredNF/NV–cxn
Figure 1 shows a distinct separation of dt-with-SubjPredNF/NV–cxn and dt-øaugSubjPredNF/NV–cxn from other constructions submitted to the linear analysis. However, in the factor "Syntactic Function to Matrix Clause" (FSYN), the dt-despite-SubjPredNF/NV–cxn is also distinguished.
The confusion matrices are generated for all specified factors to validate the results reached from the analysis of graphic materials. Due to space limitations, only the confusion matrices with the highest and lowest model accuracy are presented in Table 3 and Table 4. The analysis of the confusion matrices reveals that for with- and øaug-augmented constructions the Recall and Precision values are pretty high, particularly for models with an overall accuracy of 0,7 and higher. It suggests a high probability of their correct extraction by the created machine learning models. The specified factors may be ranked based on the model accuracy: the factor ‘Syntactic Relation with Matrix Clause’ with the model’s accuracy of 0,91 is characterized by the highest capacity to differentiate the analyzed constructions. The factors ‘Subject Determiner’ (0,71), ‘Position to Matrix Clause’ (0,71), ‘Coreference with Matrix Clause’ (0,69), ‘Text Distribution’ (0,66), ‘Syntactic Function to Matrix Clause’ (0,63) are less reliable in differentiating between the DNF/NVESconstructions. The factors ‘Punctuation marking’ (0,4) and ‘Discourse Mode Distribution’ (0,4) show the lowest differentiating capacity.
The results of the machine learning experiment are very similar to those obtained by the linguoquantitative parameterization of the constructional profiles. However, the overall efficiency of the constructed machine learning model to solve the problem of distinguishing the types of the DNF/NVES-constructions beyond the corpus is not sufficient. The model effectively distinguishes the dt-øaug-SubjPredNF/NV–cxn and dt-with-SubjPredNF/NV–cxn constructions, despite insufficient overall accuracy. However, dt-despite-SubjPredNF/NV–cxn, dt-what_with-SubjPredNF/NV–cxn, and dt-without-SubjPredNF/NV–cxn are more challenging to classify. Provided an effective model is constructed, the specified factors of the operationalized linguistic parameters can be used to distinguish types of the DNF/NVESconstructions in the present-day English usage.
The formal parameters (factors/factor values) are determined by the linguistic and constructional nature of the DNF/NVES-constructions as complex clausal constructions. The analysis includes seven parameters of the expression plane of the DNF/NVES-constructions, which define their morphosyntactic, relational, referential, syntactic-functional, positional, punctuational, and distributional features. The operationalized parameters (factors/factor values) reveal different determinative capacities; thus, the linguistic profiles of the DNF/NVES-constructions are characterized by a certain degree of proximity/remoteness in the constructional network, which allows their categorization according to the degree of linguistic homogeneity: 1) high (dt- despite-SubjPredNF/NV–cxn, dt- without-SubjPredNF/NV– cxn, dt- what_with-SubjPredNF/NV–cxn); 2) medium (dt- øaug-SubjPredNF/NV–cxn and dt- with-SubjPredNF/NV–cxn); 3) low (subgroup dt- with-SubjPredNF/NV–cxn and dt- despiteSubjPredNF/NV–cxn, dt- without-SubjPredNF/NV–cxn, dt- what_with-SubjPredNF/NV–cxn; subgroup dt- øaug-SubjPredNF/NV–cxn and dt- despite-SubjPredNF/NV–cxn, dt- withoutSubjPredNF/NV–cxn, dt- what_with-SubjPredNF/NV–cxn). The specified factors of the operationalized linguistic parameters can be utilized to differentiate types of the DNF/NVES-constructions beyond the corpus in current English usage. Differences in the quantitative realization of individual factors/factor values within one parameter of a specific DNF/NVES-construction are determined by intra-constructional variability. In contrast, quantitative differences in the realization of factors/factor values within one parameter between different DNF/NVES-constructions are determined by interconstructional variability.
The results of this study indicate a need for further research on the discussed issues. In future studies, it will be interesting to generate other types of machine learning models and assess their effectiveness in distinguishing between the types of the DNF/NVESconstructions based on the data sets for the operationalized parameters/factors/factor values extracted from the corpus. [27] V. Addanki, S. Durgapu, K. Dorasanaiah, S. Abhishek, Safeguarding SMS: A Dynamic Duo Approach to Tackle Spam Using LDA and QDA. 2023 Innovations in Power and Advanced Computing Technologies (i-PACT). IEEE, 2023 1–6. doi: 10.1109/iPACT58649.2023.10434334. [28] Cran.r-project.org. Package MASS, 2024. URL: https://cran.rproject.org/web/packages/MASS/MASS.pdf [29] M. Kuhn, J. Wing, S. Weston, A. Williams, C. Keefer, A. Engelhardt, T. Cooper, et al.
Package “caret”: Classification and Regression Training. 2023. URL: https://cran.rproject.org/web/packages/caret/caret.pdf [30] H. Wickham, ggplot2. Elegant Graphics for Data Analysis, Springer, 2016. doi: 10.1007/978-3-319-24277-4.