Comparing constructicons: A cluster analysis of the causative constructions with doen in Netherlandic and Belgian Dutch Natalia Levshina
Outline 1. Dutch causative Cx with doen 2. Data and method 3. Quantitative analyses: - Netherlandic doen - Belgian doen 4. Is it done with doen?
Causative doen in Dutch Haar stem deed het glas barsten. her voice made the glass break
Semantic variation physical causation affective causation volitional causation inducive causation (?) (Verhagen & Kemmer 1997)
Semantic variation physical causation affective causation volitional causation Haar stem deed het glas barsten. inducive causation (?) (Verhagen & Kemmer 1997)
Semantic variation physical causation affective causation volitional causation Haar stem deed me aan Lady Gaga denken. inducive causation (?) (Verhagen & Kemmer 1997)
Semantic variation physical causation affective causation volitional causation De politie deed de auto stoppen. inducive causation (?) (Verhagen & Kemmer 1997)
Semantic variation physical causation affective causation volitional causation Ze strooien poeder op je vel en doen je slapen op bevel. inducive causation (?) (Verhagen & Kemmer 1997)
Lectal variation doen is more common in Belgian Dutch than in Netherlandic Dutch (Speelman & Geeraerts 2009) doen is more common in formal/written than in informal/spoken Dutch (Levshina 2011) Frequency of causative doen, per million words Belgium Netherlands 0 50 100 150 Newspapers Usenet Spoken dialogues
Are there also differences in the semantic structure of the regional varieties?
Outline 1. Dutch causative Cx with doen 2. Data and method 3. Quantitative analyses: - Netherlandic doen - Belgian doen 4. Is it done with doen?
Corpus data Netherlands Belgium Twente News Corpus Leuven News Corpus Usenet.nl Usenet.be CGN-NL spontaneous conversations CGN-BE spontaneous conversations Total: 66 observations Total: 66 observations
Variables Causer, Causee, Affectee: sem. class, person, number, definiteness, POS, synt. expression Effected Predicate: transitivity, prepositional complements, semantic class of the caused event, lemma Coreferentiality and possession relations btw. the participants Causee only: intentionality, semantic role Negation, adverbial modifiers Mood, tense, type of the clause and sentence 35 categorical variables in total
Analytical Procedure Data frame with observations (rows) and variables (columns)
Analytical Procedure Data frame with observations (rows) and variables (columns) Matrix of distances between the observations based on Gower's distance metric
Analytical Procedure Data frame with observations (rows) and variables (columns) Matrix of distances between the observations based on Gower's distance metric Multivariate analysis (hclust) to explore the semantic structure
Analytical Procedure Data frame with observations (rows) and variables (columns) Matrix of distances between the observations based on Gower's distance metric Multivariate analyses (hclust) to explore the semantic structure Comparison of NL and BE clustering solutions
Cluster validation clusterboot{fpc}, Hennig 2007 (modified) random sampling from 66 observations with replacement random sampling from 66 original + 33 (66) additional observations calculate the Jaccard coefficient for each cluster: how similar a cluster in the original solution to its most similar cluster in the bootstrapping solution run it n (100) times calculate the average Jaccard coefficient
Outline 1. Dutch causative Cx with doen 2. Data and method 3. Quantitative analyses: - Netherlandic doen - Belgian doen 4. Is it done with doen?
Semantic structure of NL doen Cluster Dendrogram Height 0 1 2 3 4 5 0.5<J<0.6 0.6<J<0.7 0.7<J<0.8 0.8<J<0.9 J>0.9 dist.nl hclust (*, "ward")
Highly stable clusters J > 0.8
Highly stable clusters J > 0.8 affective causation
Highly stable clusters J > 0.8 doen denken: ja, dat doet mij denken aan dat boek
Highly stable clusters J > 0.8 implicit Causee Het nummer doet met zijn bombast aan zijn oude band denken,
Highly stable clusters J > 0.8 scherpe tweestemmigheid, ook, die me doet verlangen naar Gram Parsons
Highly stable clusters J > 0.8 non-affective causation
Highly stable clusters J > 0.8 Een verhaal in De Telegraaf deed het aandeel met 18.3 procent duikelen.
Highly stable clusters J > 0.8 bij ons was't de druppel die de emmer deed overlopen.
Outline 1. Dutch causative Cx with doen 2. Data and method 3. Quantitative analyses: - Netherlandic doen - Belgian doen 4. Is it done with doen?
Semantic structure of BE doen Cluster Dendrogram Height 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 0.5<J<0.6 0.6<J<0.7 0.7<J<0.8 0.8<J<0.9 J>0.9 dist.be hclust (*, "ward")
Highly stable clusters (J > 0.80)
Highly stable clusters (J > 0.80) doen denken aan
Highly stable clusters (J > 0.80) dat doet mij d'raan denken dat...
Highly stable clusters (J > 0.80) Dat doet mij eventjes denken aan een anekdote
Highly stable clusters (J > 0.80) nu pas weet ik aan wie uncle mij doet denken: hij is David Brent!
Highly stable clusters (J > 0.80) lexically specific clusters: doen vermoeden, hoop doet leven, etc.
Interim conclusions previous classifications of causation events (Verhagen & Kemmer 1997) find some support only in the case of the NL doen the Belgian sample yields a weaker cluster structure than the Netherlandic one more evidence?
Neighbour-joining algorithm D377 D172 D142 D304 D422 D178 D617 D355 D193 D69 D197 D155 D357 D157 D50 D141 D319 D611 D614 D574 D322 D152 D467 D488 D729 D707 D473 D470 D131 D135 D712 D111 D126 D726 D714 D118 D270 D469 D277 D267 D110 D259 D465 D466 D499 D500 D502 D504 D505 D506 D507 D511 D516 D517 D518 D520 D521 D522 D523 D526 D531 D533 D537 D540 D542 D548 D28 D338 D74 D580 D204 D6 D375 D327 D604 D653 D326 D158 D58 D370 D35 D177 D45 D328 D635 D303 D183 D63 D686 D430 D213 D684 D447 D443 D244 D456 D231 D685 D674 D228 D215 D99 D680 D695 D212 D455 D104 D462 D700 D664 D535 D544 D495 D512 D510 D514 D532 D503 D530 D552 D539 D538 D515 D546 D547 D509 D527 D550 D543 D524 D549 D536 NL BE
Outline 1. Dutch causative Cx with doen 2. Data and method 3. Quantitative analyses: - Netherlandic doen - Belgian doen 4. Is it done with doen?
Entrenchment of doen-schema NL BE
Productivity = entrenchment
Diachronic evidence semantic shrinking of doen (Duinhoven 1994; Verhagen 2000) e.g. the inducive use (interpersonal causation) more common in the 18 th century, now laten is preferred doen is more frequent in formal texts (more archaic features) doen is more frequent in BE (a more archaic variety) is the weaker schema of doen in NL yet another symptom that it is done with doen?
Conclusions socially and geographically uniform taxonomic networks of constructions are rather an exception than a rule the global constructicon of a language is a dynamic heterogeneous network of networks (cf. the Internet) an analysis of lectal variation in the use of constructions should also include a comparison of the constructional networks this procedure can be a tool for tracking down ongoing constructional changes experimental support is needed (weights of variables)
Thank you! natalevs@gmail.com