time in translationtime-in-translation.hum.uu.nl/static/time-in-translation...time and tense human...
TRANSCRIPT
![Page 1: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/1.jpg)
Time in Translation
Henriëtte de Swart
Utrecht Institute of Linguistics (UiL OTS)
UiL OTS colloquium, February 16, 2017
![Page 2: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/2.jpg)
Joint work with:
Martijn van der Klis
Programmer at the Digital
Humanities Lab at Utrecht
University, PhD in NWO-TinT
Bert Le Bruyn
Semantics/corpus linguistics/
L2 acquisition at Utrecht
University, co-PI NWO-TinT
![Page 3: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/3.jpg)
Time and Tense
Human life is marked by time passing by.
Language has ways to express temporal reference
window on human cognition.
Lexical means:
Words that refer to a time span (yesterday)
an event that took place in time(World War II),
that connect events in time (before, since).
Tense: grammatical expression of temporal reference.
Deixis: present (now), past, future.
![Page 4: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/4.jpg)
Language & cognition: the puzzle of cross-linguistic variation
Assume all human beings share the same cognitive
abilities,
including experience of past-present-future,
conceptualization of states as situations that ‘hold’,
and events as what ‘happens’), etc.
But then: why do languages vary so widely in their
grammaticalization of these notions?
![Page 5: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/5.jpg)
Macro-perspective: tenseless languages
Comrie (1985): all cultures have a conceptualization
of time, not all languages have tense.
Wo shuaiduan-le tui [Mandarin Chinese]
I break – LE leg
‘I broke my leg.’ (still in plaster)
Wo shuaiduan-guo tui
I break.GUO leg
‘I broke my leg.’ (once, fully healed now)
![Page 6: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/6.jpg)
Micro-perspective: have PERFECT
Morpho-syntax: auxiliary (have/be) + participle.
Core of perfect meaning: past event + present state.
1) Mary has visited Paris. [experiential perfect]
(her past visit to Paris is relevant now)
2) Mary has moved to Paris. [resultative perfect]
(she currently lives in Paris)
3) Mary has been living in Paris since 2010. (she currently lives in Paris) [continuative perfect]
![Page 7: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/7.jpg)
Micro-variation: PERFECT, PRESENT
Linguistic theory claims: continuative PERFECT meaning
conveyed by PRESENT in Dutch, French.
a.Rien ne bouge. [PRESENT] [French]
b.Nothing has moved. [PERFECT] [English]
c.Niets beweegt. [PRESENT] [Dutch]
(No et moi, Delphine de Vigan, 2007)
Manual research on literary translations confirms that not
all languages have a continuative PERFECT.
![Page 8: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/8.jpg)
Micro-variation: PERFECT, PAST, PRESENT
Multilingual conversation data: subtitles of tv series.
a. In case you hadn’t noticed, we just got a confession. [Eng]
b. Falls es ihnen entging, er hat gestanden. [German]
c. Si vous ne l’avez pas remarqué, on a des aveux. [French]
Result meaning not always conveyed by PERFECT: PAST tense implies result, PRESENT tense implies preceding event.
Manual research confirms that PERFECT use in Dutch/German/French is neither a subset, nor a superset of PERFECT use in English. Differences between these languages.
![Page 9: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/9.jpg)
Problem statement
Current approaches tend to look at the PERFECT in
isolation.
They ignore variation within a language:
Competition with other tense-aspect forms in the grammar (in particular PAST and PRESENT)
They ignore variation between languages:
Same meaning can be conveyed by other tense-aspect forms.
![Page 10: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/10.jpg)
The PERFECT – the TinT approach
Semantic maps (Haspelmath 1997)
Allows to see variation within and between languages
No preconceptualized meanings, but generate
maps directly from the data
translation corpora: same meaning, different forms
Final goal: compositional semantics of the PERFECT
across languages
Linking distribution to grammar
![Page 11: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/11.jpg)
Practical applications
L2 acquisition & education (school grammars),
human/machine translation (discourse quality),
intercultural communication (timeline in conversation)
..
![Page 12: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/12.jpg)
Time in Translation
Translation equivalents provide us with form variation across languages in contexts where the meaning is stable.
Digital resources: Europarl, Subtitle corpus
5 languages (English, Dutch, German, French, Spanish) and 3 verb forms (PAST, PRESENT, PRESENT PERFECT).
Genre variation through different corpora
EuroParl (Tiedemann 2012): formal spoken language
OpenSubtitles (Tiedemann 2016): informal conversation
Literary corpora (following de Swart 2007): story telling
![Page 13: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/13.jpg)
Methodology I
Extraction of PERFECTs from multilingual corpora
Word-level alignment of verb phrases
Tense attribution for translations
Discard ‘other’ translations (nominalizations, ‘free’ translations)
Manual/automatic tagging (language-specific)
Create a dissimilarity matrix
Multidimensional scaling: project variation onto a semantic
map.
Methodology inspired by Wälchli & Cysouw 2012
![Page 14: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/14.jpg)
Methodology II
Distance function: we define a pair of source and translations to be maximally similar (d=0) if all tenses match up.
d(1, 2) = 2/5, d(1, 3) = 2/5, d(2, 3) = 4/5
# German English French Spanish Dutch
1 Perfekt Present perfect Passé composé Pretérito perfecto
conpuesto Voltooid
tegenwoordige tijd
2 Präterium Simple past Passé composé Pretérito perfecto
conpuesto Voltooid
tegenwoordige tijd
3 Perfekt Present perfect Passé récent Pasado receinte Voltooid
tegenwoordige tijd
![Page 15: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/15.jpg)
Classification of tense forms into
PERFECT, PRESENT, PAST
DE EN ES FR NL
PERFECT 360 347 371 481 438
PRESENT 19 18 47 20 20
PAST 124 146 89 8 36
PLUPERFECT 4 1 3 2 18
other 5 - 2 1
Distribution of tense forms across languages
![Page 16: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/16.jpg)
PERFECT map: English (plot1/2)
![Page 17: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/17.jpg)
Screenshot data point: complete example
![Page 18: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/18.jpg)
Perfect map: French (plot 1/2)
![Page 19: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/19.jpg)
Perfect map: German (plot 1/2)
![Page 20: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/20.jpg)
Visualization
We visualize the results of MDS on a scatter plot.
We use the tenses per language as labels.
Click/unclick tenses to filter them.
We can choose which dimensions of the MDS to show.
We can choose which language to show.
![Page 21: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/21.jpg)
Filters
![Page 22: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/22.jpg)
Mixed tenses in the 5-tuple
![Page 23: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/23.jpg)
Different tenses in 5-tuple
![Page 24: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/24.jpg)
How to interpret the dimensions?
MDS plots dissimilarity along (statistical) dimensions – no
interpretation given.
We ran MDS for 5 dimensions. Plots shown so far:
dimensions 1 and 2.
Plot 1/2 carves out the opposition between PAST, PRESENT
and PERFECT in German, but not in English or Spanish
(‘confetti’).
Which dimensions lead to clustering in which language?
![Page 25: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/25.jpg)
German plot 1/2
![Page 26: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/26.jpg)
English plot 2/3
![Page 27: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/27.jpg)
Dutch plot 3/4
![Page 28: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/28.jpg)
Cross-linguistic variation correlates
with dimensions
The plots show that languages are sensitive to
different dimensions for carving out the domains of
use of their PAST, PRESENT, PERFECT.
In MDS, higher dimensions account for more variation
than lower dimensions. But: tense use is complex.
Tentative conclusion: the dimensions that German
tense use is sensitive to are more frequent in our
dataset than those that English or Dutch are sensitive
to.
![Page 29: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/29.jpg)
Diving into the data (1)
The maps show more central and more peripheral uses of
the English Perfect.
The more the points are distant from each other, the
more their tenses are different.
We can investigate ‘outliers’ to find cases where one
language is different from the other languages.
E.g. confirm that English requires a PAST with a locating
time adverbial, whereas German, Dutch and French
tolerate a PERFECT in this configuration. Spanish patterns
with English (see Schaden 2009).
![Page 30: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/30.jpg)
![Page 31: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/31.jpg)
![Page 32: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/32.jpg)
Both English and Spanish should allow time adverbials
that refer to ‘now’; confirmed in Europarl.
![Page 33: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/33.jpg)
Subtle differences in what counts as ‘now’
![Page 34: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/34.jpg)
Diving into the data (2)
We can look at special verb forms in English, and find
out how other languages deal with them in a
different tense-aspect grammar.
E.g. English Present Perfect Continuous has been
claimed to translate as a PRESENT in French (cf.
Nishyama & Koenig 2010), and manual translation
data confirm this for Dutch too. But German
maintains a PERFECT in sentences that contain a seit
(‘since’) adverbial. Note the special construction in
Spanish.
![Page 35: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/35.jpg)
Continuative reading: Dutch, French Present
![Page 36: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/36.jpg)
Diving into the data (3)
We can use the maps to find datapoints with new insights,
not yet in the literature.
More examples of continuative readings can be found
once we take into account the aspectual class of the verb
(no tagging yet – manual research so far): stative verbs
also give rise to continuative readings.
Interestingly, such continuative readings do not only have
PRESENT translations, but also PAST ones.
Implications for theories on PERFECT and (stative) PAST.
![Page 37: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/37.jpg)
![Page 38: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/38.jpg)
Screenshot of ‘have known’ with past translation.
![Page 39: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/39.jpg)
Diving into the data (4)
English is unique for having a Present Perfect
Continuous, French and Spanish are special for
having a recent PAST.
We can use the maps to find out how other
languages deal with them in a different tense-
aspect grammar.
![Page 40: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/40.jpg)
![Page 41: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/41.jpg)
![Page 42: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/42.jpg)
Recency as a dimension of the PAST or the
PERFECT?
Where French and Spanish use a recent PAST verb form,
English, German and Dutch use a PERFECT.
Languages who use a PERFECT to refer to a recent past
use an additional time adverbial: just, gerade,
kortgeleden.
Tentative conclusion: ‘recent past’ can be a dimension
of the PAST or of the PERFECT, but in both cases recency
requires additional marking.
![Page 43: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/43.jpg)
Optionality
Languages may vary in tense use without an obvious reason:
what is driving optionality?
a) I am commanded to take you hence from this place, to the Tower. [EN] [present]
b) Ich habe den Befehl Euch von hier fortzubringen, in den Tower. [DE] [present]
c) Ik heb het bevel gekregen u naar de Tower te brengen [NL] [perfect]
Result may be encoded (in PERFECT) or entailed (lexical aspect).
Languages have multiple, overlapping strategies to mark
certain meaning configurations. Interaction with other features
of the grammar: lexicon, active/passive, clause type,…
![Page 44: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/44.jpg)
Future research
Lexical annotation. Annotate verbs for aspectual class, identify locating time adverbials, identify aspectually sensitive adverbials (since, for, already, just, always, negation), and investigate interaction with PAST/PRESENT/PERFECT.
Syntactic annotation. Investigate the interaction of (i) voice (active/passive, see CLIN 2015 presentation) and (ii) clause type (main/subordinate, e.g. conditional, declarative/interrogative) with PAST/PRESENT/PERFECT.
Discourse annotation. Investigate contexts in which PERFECT is used (speech acts, interaction with register)
![Page 45: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/45.jpg)
Work in progress..
More on our research programma on the website:
http://timealign.pythonanywhere.com/
We are just getting started, so feedback is
appreciated!
![Page 46: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/46.jpg)
Methodology Time in Translation
Step 1: PERFECT extractor
Step 2: word-level alignment
Step 3: tense attribution
Step 4: dissimilarity matrix
Step 5: multidimensional scaling
![Page 47: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/47.jpg)
Step 1) Extraction of PERFECTs
Stand-alone application: PerfectExtractor
Simple principle:
Look for an auxiliary verb (HAVE/BE)
Find a neighboring past participle
Takes care of words in between, lexical restrictions, reversed
order, passive present perfects.
Uses config files to discern between languages and corpora
Source code on GitHub: https://github.com/UUDigitalHumanitieslab/time-in-translation
![Page 48: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/48.jpg)
Step 2) Word-level alignment
Web application: TimeAlign
Allows for manual alignment of translations of PERFECTs.
Source code on GitHub:
https://github.com/UUDigitalHumanitieslab/timealign
![Page 49: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/49.jpg)
Step 3) Tense attribution
To allow for calculating distances between fragments,
we need to assign a tense to every selected
translation.
Algorithm for English, French and Dutch
Manually for German and Spanish
POS-tags do not discern between PRESENT and PAST
tense for auxiliary verbs
![Page 50: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/50.jpg)
Step 4) Dissimilarity matrix
Distance function: we define a pair of source and
translations to be maximally similar (d=0) if all tenses
match up.
d(1, 2) = 2/5, d(1, 3) = 2/5, d(2, 3) = 4/5
# German English French Spanish Dutch
1 Perfekt Present perfect Passé composé Pretérito perfecto
conpuesto Voltooid
tegenwoordige tijd
2 Präterium Simple past Passé composé Pretérito perfecto
conpuesto Voltooid
tegenwoordige tijd
3 Perfekt Present perfect Passé récent Pasado receinte Voltooid
tegenwoordige tijd
![Page 51: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/51.jpg)
Step 4) Dissimilarity matrix
Using our defined distance function, we can create a
dissimilarity matrix.
#1 #2 #3
#1 - 2/5 2/5
#2 2/5 - 4/5
#3 2/5 4/5 -
![Page 52: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/52.jpg)
Step 5) Multidimensional Scaling
Multidimensional scaling (MDS) tries to create a low-
dimensional representation of the data, while
respecting distances in the original high-dimensional
space.
![Page 53: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/53.jpg)
Assigning tenses
To allow for calculating distances between fragments,
we need to assign a tense to every selected
translation.
Automatically for English, French and Dutch; via
POS-tags.
Manually for German and Spanish (POS-tags in
those languages do not discern between PRESENT
and PAST tense).
![Page 54: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/54.jpg)
Matrix of tenses constitutes input for our (dis)similarity matrix.
![Page 55: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/55.jpg)
Distance measure We define a pair of source and translations to be maximally similar if all the tenses match up.
d(1, 2) = 3/5, d(1, 3) = 3/5, d(2, 3) = 1/5
# German English French Spanish Dutch
1 Perfekt Present
perfect
Passé
composé
Pretérito
perfecto
conpuesto
Voltooid
tegenwoord
ige tijd
2 Präterium Simple past Passé
composé
Pretérito
perfecto
conpuesto
Voltooid
tegenwoord
ige tijd
3 Perfekt Present
perfect
Passé
récent
Pasado
receinte
Voltooid
tegenwoord
ige tijd
![Page 56: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/56.jpg)
Dissimilarity matrix
Using our defined distance measure, we can create a
dissimilarity matrix, by substracting the distance from 1.
#1 #2 #3
#1 - 1 – 3/5 = 2/5 1 – 3/5 = 2/5
#2 1 – 3/5 = 2/5 - 1 – 1/5 = 4/5
#3 1 – 3/5 = 2/5 1 – 1/5 =4/5 -
![Page 57: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/57.jpg)
Multidimensional Scaling (MDS)
Multidimensional scaling (e.g. Wälchli & Cysouw 2012)
tries to create a low-dimensional representation of the
data, while respecting distances in the original high-
dimensional space.
MDS takes the dissimilarity matrix as input.
We choose to scale down to 5 dimensions.
We use MDS implementation in the Scikit-learn
package: http://scikit-learn.org
![Page 58: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/58.jpg)
Finding relevant dimensions
Irrelevant dimensions: confetti
Relevant dimensions: 3 spaces on the map for PAST,
PRESENT and PERFECT.
Strategy for finding relevant dimensions (per language)
set dimensions to 1/1, 2/2, etc. – creates a diagonal.
Unclick perfect; if lower/upper half of diagonal
disappears, this dimension models the perfect/non-
perfect dimension.
Unclick perfect, and then unclick either past or present: if
lower/upper half of diagonal disappears, this dimension
models the past/present dimension.
![Page 59: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/59.jpg)
German dimension 1 (all)
![Page 60: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/60.jpg)
German dimension 1 without perfect
![Page 61: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/61.jpg)
German dimension 2 (all)
![Page 62: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/62.jpg)
German dimension 2 without perfect
![Page 63: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/63.jpg)
German dimension 2: only past
![Page 64: Time in Translationtime-in-translation.hum.uu.nl/static/Time-in-Translation...Time and Tense Human life is marked by time passing by. Language has ways to express temporal reference](https://reader030.vdocuments.site/reader030/viewer/2022040303/5e93262b1e0d7d46f0025b41/html5/thumbnails/64.jpg)
German dimension 2: only present