Periodization of constructional productivity in diachronic ...fperek.net/pdfs/slides-icame38.pdf ·...
24
Periodization of constructional productivity in diachronic corpora Florent Perek University of Birmingham
Transcript of Periodization of constructional productivity in diachronic ...fperek.net/pdfs/slides-icame38.pdf ·...
Periodization of constructional productivity in diachronic corpora Florent Perek University of Birmingham
Overview
o New method for diachronic studies
o Aim: identify stages of language change in the productivity of grammatical constructions
o Two case studies
Corpus-based studies of language change o Typical corpus-based studies of language change
– Extract tokens from a diachronic corpus
– Classify these tokens according to some criterion
– Compare the state of the language at different points in time
o Assess stages of language change – When was it relatively stable, and for how long?
– When did it change (and how)?
Manual periodization
o Normalised frequency of the hell-construction in the COHA “Verb the hell out of”, e.g., You scared the hell out of me!
0.0
0.5
1.0
1.5
2.0
2.5
3.0
Decades
Nor
mal
ised
freq
uenc
y (p
er M
W)
1930 1940 1950 1960 1970 1980 1990 2000
Problems with manual periodization
o Stages are not always clear to discern
o Potentially subjective: what are the criteria for splitting periods? – Different possible groupings for the same data
– Comparison between studies
o More complex when multiple variables are considered e.g., token frequency + type frequency
Periodization
o This problem was first exposed by Gries & Hilpert (2008)
o They introduce “variability-based neighbour clustering” (VNC) as a method for automatic periodization
o Variant of agglomerative clustering algorithm – Periods are grouped according to their similarity, following
some pre-defined criteria
– Only time-adjacent periods can be merged
Gries, S., & Hilpert, M. (2008). The Identification of Stages in Diachronic Data: Variability-based Neighbor Clustering. Corpora, 3, 59–81.
The VNC algorithm
o Starting point: data partitioned into “natural” time periods (years, decades, etc.)
1. Look at all pairs of adjacent periods (e.g., 1930s-1940s, 1940s-1950s, etc.). Measure their similarity according to some quantifiable property/ies.
2. Merge the two periods that are the most similar.
3. Calculate the properties of the merger as the mean values of its constituent periods.
o Repeat until all periods have been merged.
VNC: an example
o VNC with one variable: frequency of the hell-construction
Token frequency (per million words)
Decades
Sum
med
dis
tanc
e (S
D)
1930 1940 1950 1960 1970 1980 1990 2000
0.0
1.0
2.0
0.0
1.0
2.0
3.0
VNC
o Two kinds of uses of VNC in the literature – To partition data in a principled way for further analysis
– To uncover patterns of change and/or compare changes
o So far mostly based on quantitative variables – Frequencies: tokens, types, hapax legomena, etc.
– Frequency distributions of lexical items, collexeme analysis
o Lines up with usage-based linguistics: grammatical representations are shaped by frequency
o Frequency = good starting point for looking at the history of constructions, but do not tell the whole story
Productivity
o Especially true for the study of productivity – The property of a construction to attract new lexical fillers
– E.g., verbs in the way-construction (Israel 1996)
They hacked their way through the jungle. (16th century)
She talked her way into the club. (19th century)
o Type frequency often taken as an indicator of productivity – Number of different items, but not how different they are
– Need to consider the semantic diversity of the distribution
Israel, M. (1996). The way constructions grow. In A. Goldberg (ed.), Conceptual structure, discourse and language. Stanford, CA: CSLI Publications, 217-230.
Operationalizing word meaning
o Distributional semantics (Lenci 2008) – “You shall know a word by the company it keeps.”
(Firth 1957: 11)
– Words that occur in similar contexts tend to have related meanings (Miller & Charles 1991)
o Captures the meaning of words through their distribution in a large corpus
o Proposal: use distributional semantics to build representations of the semantic range of a construction
Firth, J.R. (1957). A synopsis of linguistic theory 1930-1955. In Studies in Linguistic Analysis, pp. 1-32. Oxford: Philological Society.
Lenci, A. (2008). Distributional semantics in linguistic and cognitive research. Rivista di Linguistica, 20(1), 1–31. Miller, G. & W. Charles (1991). Contextual correlates of semantic similarity. Language and Cognitive Processes, 6(1), 1-28.
“Bag of words” approach
o Distributional data extracted from COHA (Davies 2010); 400 MW from 1810 to 2009
o Collocates of all verbs in a 2-word window
o Restricted to the 10,000 most frequent nouns, verbs, adjectives and adverbs
the upper crust; cut a lip in it ; and ornament growing season. “I spend a lot of my garden time and disdainful port; looked intrepidly and indignantly mocking me? What! I marry a woman sixty-four years old that they no longer fight against it ; it is embalmed
Davies, M. (2010). The Corpus of Historical American English: 400 million words, 1810-2009. Available online at http://corpus.byu.edu/coha/
Distributional semantic model
o Co-occurrence frequencies turned into PPMI scores
o 10,000 columns of the co-occurrence matrix reduced to 300 distributional-semantic features with SVD
o In the distributional semantic model, each verb corresponds to an array of 300 values, i.e., a vector
o Semantically similar words tend to have similar values in the same features
(column1) (column2) (column3) (column300)
find 15.59443 -2.022215 0.561186 ... -0.5778517 carry 21.82777 4.714768 -11.974389 ... -0.5226300 answer 11.66246 2.008967 8.810539 ... -0.2389049 push 22.09577 13.130336 -6.027978 ... 0.8539545 ... ... ... ... ... ...
Period vectors
o For each period, extract the semantic vector of each verb in the distribution of the construction
o Add all vectors and divide by the number of verbs: this is the period vector
o “Semantic average” of the distribution; reflects semantic properties of the verbs attested in the period
(column1) (column2) (column3) (column300)
make 14.09814 -4.231832 -1.844898 ... 0.06963598 find 15.59443 -2.022215 0.561186 ... -0.5778517 push 22.09577 13.130336 -6.027978 ... 0.8539545 Sum 51.78834 6.876289 -7.311691 ... 0.3457388 /3 17.26278 2.292096 -2.43723 ... 0.1152463 period vector
Distributional period clustering
o The VNC algorithm is run on the period vectors
o Similarity is measured by cosines between vectors
o The output dendrogram shows the semantic history of the construction: – Early mergers correspond to periods of semantic stability.
– Late mergers of large clusters indicate semantic shifts.
Two case studies
o Both using COHA, focusing on verbs in two constructions
o The hell-construction V the hell out of NP You scared the hell out of me!
I enjoyed the hell out of that show.
They beat the hell out of him.
o The way-construction V one’s way PP They hacked their way through the jungle.
She talked her way into the club.
Restricted to the “path-creation” interpretation: the verb describes an action that enables motion
(vs. manner: They trudged their way through the snow)
The hell-construction
Gradual expansion and “plateauing” rather than brutal shifts
VNC dendrogram
Decades
Sum
med
cos
ine
dist
ance
1930
1940
1950
1960
1970
1980
1990
2000
0.0
0.4
0.8
1.2
Scree plot
ClustersC
osin
e di
stan
ce
1 2 3 4 5 6 7
0.10
0.15
0.20
0.25
0.23
0.2
0.150.13
0.160.14 0.14
Token frequency (per million words)
Decades
Sum
med
dis
tanc
e (S
D)
1930 1950 1970 1990
0.0
1.0
2.0
●
● ●
●●
●
●
●
0.0
1.0
2.0
3.0
Scree plot
Clusters
Dis
tanc
e (S
D)
1 2 3 4 5 6 7 8
0.0
0.2
0.4
0.6
0.8
0.9
0.40.4 0.4
0.2
0.1 0.1
Type frequency
Decades
Sum
med
dis
tanc
e (S
D)
1930 1950 1970 1990
05
1020
30
●
●
●●
●
●
●
●
010
2030
40
Scree plot
Clusters
Dis
tanc
e (S
D)
1 2 3 4 5 6 7 8
02
46
810
10.6
64.9
4.2
2.6
1 0.7
Hapax legomena
Decades
Sum
med
dis
tanc
e (S
D)
1930 1950 1970 1990
05
1015
20●
●
● ●
●
●
●
●
05
1015
2025
30
Scree plot
Clusters
Dis
tanc
e (S
D)
1 2 3 4 5 6 7 8
02
46
7.5
4
2.92.1
1.7 1.7
0
The hell-construction
o The shape of the dendrogram reflects gradual expansion rather than brutal shifts (cf. Perek 2014, 2016)
o Construction centered on the same semantic classes, with new members joining the periphery
o Vs. two-way split obtained with quantitative measures
o Questions the practice of using quantitative data for the initial partitioning
Perek, F. (2014). Vector spaces for historical linguistics: Using distributional semantics to study syntactic productivity in diachrony. In Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics, Baltimore, Maryland USA, June 23-25 2014 (pp. 309-314).
Perek, F. (2016). Using distributional semantics to study syntactic productivity in diachrony: A case study. Linguistics, 54(1), 149–188.
The way-construction VNC dendrogram
Decades
Sum
med
cos
ine
dist
ance
1830
1840
1850
1860
1870
1880
1890
1900
1910
1920
1930
1940
1950
1960
1970
1980
1990
2000
0.0
0.5
1.0
1.5
Scree plot
Clusters
Cos
ine
dist
ance
1 3 5 7 9 11 13 15 17
0.06
0.07
0.08
0.09
0.10
0.11
0.094
0.104
0.0980.1 0.1
0.102
0.089
0.0830.083
0.072
0.0780.076
0.071
0.0790.0770.074
0.065
1830s – 1870s Concrete, physical actions, literal creation of a path: hew, shape, explore, carve, track, enforce, shoulder, etc.
1890s – 2000s More abstract: communication, social interaction, etc.: joke, bellow, chatter, snarl, spit, laugh, talk, bully, etc.
1880s: transition period More abstract verbs than the previous period: buy, smell, stammer, beg, think, pay, etc. More concrete verbs than the next period: bore, pierce, feel, wear, melt, trace, burn, etc.
The way-construction
o Change from mostly concrete to more abstract verbs (in line with Israel 1996, Perek aop)
o How does distributional semantics compare to collostructional analysis for periodization? – Which verbs occur more distinctively frequently in each
decade than in the others? (Hilpert 2006)
– Each verb receives an association score in each decade
– The distribution of collexemes can be used as input for VNC (Hilpert 2012): change in lexico-grammatical associations
Hilpert, M. 2006. Distinctive collexeme analysis and diachrony. Corpus Linguistics and Linguistic Theory 2(2). 243–57. Hilpert, M. 2012. Diachronic collostructional analysis. How to use it, and how to deal with confounding factors. In K. Allan &
J. Robynson (eds.), Current Methods in Historical Semantics, 133–160. Berlin: Mouton de Gruyter. Perek, F. (ahead-of-print). Recent change in the productivity and schematicity of the way-construction: a distributional
semantic analysis. Corpus Linguistics and Linguistic Theory.
VNC with collostructional analysis VNC dendrogram
Decades
Cumu
lated
cosin
e dist
ance
s
1830
1840
1850
1860
1870
1880
1890
1900
1910
1920
1930
1940
1950
1960
1970
1980
1990
2000
02
46
8
Scree plot
Clusters
Cosin
e dist
ance
1 2 3 4 5 6 7 8 9 11 13 15 17
0.40.6
0.8
0.966
0.6710.6210.612
0.5860.533
0.470.497
0.470.4610.4670.4650.4180.413
0.356
0.2410.241
Physical change of state: cut, hew, tear, cleave, break, pierce, burst, etc. Semantically neutral verbs: take, find, win, make
talk, buy, negotiate, lie (1930s-2000s)
Haphazard list of more abstract verbs:
earn, sing, advertise, brew, declaim, experiment (1910s-1920s)
work, pick (1930s-2000s)
VNC with collostructional analysis
o Some evidence of a shift from concrete to abstract verbs
o But it is attested later than in the distributional VNC
o Semantic classes are less clearly identifiable
o With collostructional analysis, the detection of changes is highly dependent on token frequency – Frequency associations are not always semantically relevant
– “Real” change is only exemplified by high-frequency types
– The timing of these changes is delayed, until sufficient frequency is reached
Conclusion
o Distributional period clustering captures semantic changes in the productivity of constructions
o Represents a step forward from regular VNC
o Results confirm previous studies
o Two advantages – Semantic changes are inferred mathematically rather than
assessed impressionistically
– Changes can be dated more precisely
… paper (with Martin Hilpert) under review, downloadable at www.fperek.net
Thanks for your attention! [email protected] www.fperek.net
