=Paper=
{{Paper
|id=Vol-2493/system3
|storemode=property
|title=Translation inference through multi-lingual word embedding similarity
|pdfUrl=https://ceur-ws.org/Vol-2493/system3.pdf
|volume=Vol-2493
|authors=Kathrin Donandt,Christian Chiarcos
|dblpUrl=https://dblp.org/rec/conf/ldk/DonandtC19
}}
==Translation inference through multi-lingual word embedding similarity==
https://ceur-ws.org/Vol-2493/system3.pdf
Translation Inference through Multi-lingual
Word Embedding Similarity
Kathrin Donandt, Christian Chiarcos
Applied Computational Linguistics (ACoLi)
Goethe-Universität Frankfurt, Germany
{donandt|chiarcos}@cs.uni-frankfurt.de
Abstract. This paper describes our contribution to the Shared Task on
Translation Inference across Dictionaries (TIAD-2019). In our approach,
we construct a multi-lingual word embedding space by projecting new
languages in the feature space of a language for which a pretrained em-
bedding model exists. We use the similarity of the word embeddings to
predict candidate translations. Even if our projection methodology is
rather simplistic, our system outperforms the other participating sys-
tems with respect to the F1 measure for the language pairs which we
predicted.
1 Background
The Second Shared Task on Translation Inference across Dictionaries (TIAD-
2019) has been conducted in conjunction with the Conference on Language, Data
and Knowledge (LDK, Leipzig, Germany, May 2019). As in the first edition, the
objective is to automatically obtain new bilingual dictionaries based on existing
ones.
Our contribution is based on the application of a technology originally de-
veloped for a related, but broader problem, the identification of cognates in
dictionaries of languages that are either diachronically or culturally related with
each other. We consider two words from languages A and B to be cognates (in
a broad sense) if they share the same etymological origin, either because they
have been inherited from a language C which is ancestral to both A and B or
if they are loanwords from the same source.1 Cognates are characterized by a
systematic phonological relationship with each other, but also, a systematic se-
mantic relation. A typical example is German Bank and English bench which
1
Note that this technical definition is broader than the definition typically applied
in linguistics: In a strict sense, cognates are only those of the first case. However,
for languages with a long and intense history of contact, the differentiation between
cognates in a strict sense and mutual loans is not always clear-cut. A typical case is
the large number of common roots in Turkic and Mongolian which can be variously
attributed to either a loan between (proto-) languages or to their common (but hy-
pothetical) ancestral language. In both situations, however, the linguistic character-
istics (systematic phonological similarity and semantic relatedness) are comparable,
thus motivating a generalized, technical definition of cognates.
Copyright © 2019 for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
2 Kathrin Donandt, Christian Chiarcos
(in one sense of the German word) are semantically identical, and they reflect
the same sound correspondence of k and ch that we also find in word pairs
such as Kinn/chin. An example for the second case is English bank (as in river
bank) and German -bank in compounds such as Sandbank ‘sandbank’. Again,
senses are similar (an accumulation of soil in or at a water stream), but here,
the English word is a loan (either from Low German or Scandinavian). For Bank
and bank as ‘financial institute’, the situation is more complex: In both German
and English, this is a loan from Italian banca, but the origin of this word is a
Germanic (Langobardian) word with the meaning ‘bench’, i.e., the place where
clients had to wait in order to receive financial support. Cognate candidates can
be identified by means of phonological and semantic similarity metrics, and the
latter are the basis for the implementation that we describe with this paper. Our
research has been conducted in the context of the Independent Research Group
‘Linked Open Dictionaries’ (LiODi), funded by the German Federal Ministry of
Education and Science (BMBF), and applied to language contact studies in the
Caucasus area.
Out of the context of this project, one aspect that we appreciate in the TIAD
shared task is that it does not depend nor require the use of parallel corpus
data. This is in fact the situation for most languages studied in our project.
For many low-resource languages, the only language data that is available is
provided in the form of dictionaries (or plain word lists) on the one hand and
in the form of monolingual, non-translated text on the other, often in the form
of field recordings. Parallel (translated) data, however, is scarce – if available
at all (and if so, often limited to religious texts, e.g., translations of the Bible
with limited representativeness for the language as a whole and not necessarily
written by a native speaker). The technological challenges of the TIAD task thus
closely mirror the technical (data) challenges in the LiODi project, except that
it is limited for translation equivalence whereas we aim to include less direct
semantic relations, as well.
In our approach, we make use of the similarity of cross-lingual word embed-
dings – genuine and projected ones – in order to build new bilingual dictionaries.
The projection of words of one language into the embedding space of another and
thus the generation of a multi-lingual word embedding space is a widely discussed
topic. For an overview of different methodologies of cross-lingual word embed-
ding generation, we refer to [3]. Different from more sophisticated projection
methods, we use the provided bilingual resources and obtain word embeddings
for a new language by summing up the vector representation(s) of the transla-
tion(s) into a language for which we already have either pretrained or projected
word embeddings. It should be noted that our simple embedding-based approach
allows us to generalize beyond translation information provided by dictionaries,
as it naturally includes similarity assessments in accordance with the underlying
embedding space. However, as such more remotely related terms receive scores
much lower than those of words between which a translation path over different
dictionaries exists, the threshold-based implementation we apply for the TIAD
shared task systematically restricts our results to the latter.
Translation Inference through Multi-lingual Word Embedding Similarity 3
Despite its simplicity, our implementation achieves good results in compari-
son to the other systems submitted to TIAD. When similarity beyond translation
equivalence is excluded (as by applying a selection threshold), our system is ba-
sically a neural reconstruction of [4], which has also been used as a baseline in
this task (see Sect. 4.2).
2 Data & Preprocessing
We use the TSV edition of the dictionaries provided by the task organizers.
Whereas we only use the languages and language pairs provided in these dic-
tionaries, it would be possible to add more language pairs to be processed by
our approach, as long as they are provided in the TIAD-TSV format. In the
context of the H2020 project ‘Pret-a-LLOD. Ready-to-use Multilingual Linked
Language Data for Knowledge Services across Sectors’, we are currently in the
process of creating such translation data, with OntoLex-Lemon and TIAD-TSV
data for 130 language varieties and 373 language pairs already being avail-
able from http://github.com/acoli-repo/acoli-dicts/. For the TIAD-2019 task,
selected parts of this data have been taken into consideration, however, only for
estimating prediction thresholds (Sect. 4.1), not as a source of external lexico-
graphical information.
For initializing the embedding space, we employ readily available monolingual
embeddings from a particular ‘reference language’. Different reference languages
would be feasible in this context, and many different types of embeddings are
available for each. Here, we build on the 50 dimensional GloVe v.1.2 embed-
dings for English, trained on Wikipedia and Gigaword [2].2 Our implementation
can be applied to any other sort of embeddings as long as they are provided in
the same format (one word plus its embedding per line, the first column hold-
ing the word [without whitespaces] followed by whitespace-separated columns
of doubles). Only for reasons of time, we chose the lowest-dimensional embed-
dings provided, and no other reference languages nor embeddings have been
experimented with. We would expect that higher-dimensional embeddings, and
embeddings from other languages with many dictionary pairs (e.g., Spanish)
would outperform our implementation, and this should be addressed in subse-
quent research.
In order to make our approach computationally more efficient, we prune
the embedding space before running our experiments from words not occurring
in the (English) dictionary data we are working with. We add multi-word ex-
pressions consisting of words present in the model by taking the sum of their
representations. As we want to project other languages into the semantic space
of our reference language, we need to store the information to which language a
word of the model belongs to. We solve this by adding a language tag to the word:
Example entry in the original embedding model:
2
Download link: http://nlp.stanford.edu/data/glove.6B.zip
4 Kathrin Donandt, Christian Chiarcos
house 0.60137 0.28521 -0.032038 -0.43026 ... -1.0278 0.039922 0.20018
Same entry in the updated embedding model:
”house”@en 0.60137 0.28521 -0.032038 -0.43026 ... -1.0278 0.039922 0.20018
3 Approach
We developed a baseline approach, and submitted the results of this approach to
the Shared Task organizers. Later on, we elaborated our baseline and compared
its performance to our baseline’s performance by applying an interal evaluation
procedure (4.3). The results of our elaboration were not submitted to the Shared
Task organizers due to time constraints.
3.1 Baseline
Using the pretrained embedding model for our reference language, we generate
word embeddings for all the other languages of the provided dictionaries. We
project the words of these languages into the embedding space of the reference
language, e.g. for the Basque word abentura and its English translations adven-
ture and venture (according to the Basque-English Apertium dictionary), we
calculate the new embedding by summing up the embeddings of the possible
translations into English:
*
”abentura”@eu = * *
”adventure”@en + ”venture”@en
In a first iteration, we thus project all those languages into the embedding
space of our reference language English for which a dictionary with English
translations exist (here: EU → EN, EO → EN).3 In order to improve coverage,
we also included the inverse direction (EN → CA, EN → ES, EN → GL). We refer
to these languages (CA, EU, ES, EO, GL) as ‘first-level’ languages. In the second
(and subsequent) iterations, we then project all the languages into the enriched
embedding space for which at least one dictionary exists that connects it with
another language in the embedding space. If more than one dictionary does exist
(e.g., for French EO → FR, FR → ES, FR → CAT), we sum over all translations
in the embedding space. AN, AST, FR and IT are the ‘second-level languages’
for which embeddings are created in this way. Note that once embeddings for a
particular language are created, they are not updated nor extended any more,
even if in a further iteration additional dictionaries also containing this language
are encountered. For example, we obtain EO embeddings from EO → EN and
nothing else, even if after adding CA to the embedding space, EO could be
reached via EN → CA → EO, as well. This also means that lexical material for
Esparanto provided by the CA → EO dictionary which is not covered in the EO
3
In the following, we refer to the languages solely by using the language abbreviations
used in the Shared Task description (https://tiad2019.unizar.es/task.html)
Translation Inference through Multi-lingual Word Embedding Similarity 5
→ EN dictionary will not be represented in the embedding space. Again, this is
an implementation decision taken in the interest of time and to be reconsidered
in subsequent research. With a larger set of dictionaries and language pairs, we
would continue these iterations until no more languages can be reached in the
language graph represented by the dictionaries.
In order to generate a new bilingual dictionary A → B, we use this enriched,
multilingual word embeddings and predict candidate translations of a word a in
language A by choosing the words in language B whose embeddings are most
similar to the embedding of a, and which have the same part-of-speech as a. As
similarity measure we chose cosine similarity, which is ignorant of vector length,
and faster in lookup than Euclidian distance. This also allows us to add rather
than average in the process of projecting new language into the embedding space.
It should be noted that speed is a decisive criterion in the intended application
of the technique, as we aim to provide on-the-fly cognate detection.
3.2 Experiments with sense clustering
We elaborated our baseline approach by taking into consideration the sense(s) of
a word to be translated. We follow the definition of senses in the TIAD-OntoLex
edition of the Apertium dictionaries by assuming that in a bilingual dictionary,
each translation b1 , b2 , ... provided for a source language word a entails a spe-
cific sense ab1 , ab2 , .... The number of OntoLex senses in the TIAD data is thus
identical to the number of translations provided. Unfortunately, the Apertium
data does not provide sense-level linking beyond different bilingual dictionaries,
so that sense-level linking cannot be used as a feature for translation inference.
We do not have information on whether sense ab1 induced from the A → B dic-
tionary has any relation with the sense ac15 induced from a A → C dictionary.
However, if the basic assumption is correct, then we can approximate the
factual number of senses of a source language word a as the maximum number n
of translations in any bilingual dictionary. Assuming that translations referring
to the same word sense tend to be closer in the embedding space, we can use
this information for disambiguation and try to identify the most representative
translation for every sense. We do so by performing a lookup in the dictionaries
to get all translations, and then cluster (the embeddings of) these translations
into n clusters. For every cluster, we then return the target language word(s)
closest to the cluster center.
We generate “sense embeddings” for a as follows:
1. Look up all possible translations of a in the dictionaries, e.g.
(a, b11 ), ..., (a, b1n ), (a, b21 ), ..., (a, b2m ), etc.,
where
bij : the j’th translation of a in a dictionary A → Bi (or Bi → A)
N : number of dictionaries where A appears as source or target language;
i = 1, ..., N .
6 Kathrin Donandt, Christian Chiarcos
2. Apply k-means clustering4 over all possible translations bij , where
k = maxi Ji ,
where Ji : number of entries containing a in A → Bi .
3. For each of the k cluster centers, take the closest embedding (irrespective of
its language) as a sense embedding of word a. This is the basis for finding
the closest target language embedding, resp., word, per sense cluster.
We then predict possible translations of a given word in a source language into
a specified target language by looking up the most similar word embeddings of
the target language to each of the sense embeddings of the word in the source
language.
While the baseline predicted n words in the target language as candidate
translations, we now predict at least n/k words for each of the k sense embed-
dings, with distance relative to the ‘sense cluster’ rather than the source language
embedding.
4 Results
We first ran an internal evaluation in order to get an idea of the quality of our ap-
proach. After that, we determine a cosine similarity threshold for our predicted
translations in order to select the best translation candidates. Later, we use the
internal evaluation strategy again to compare our baseline with the elaboration.
Unlike most other systems in TIAD-2019, we provide only the data points ex-
ceeding our thresholds, whereas a lower threshold (and parameter estimation by
the task organizers) may have produced better results on the blind test set.
4.1 Threshold Estimation
Our baseline approach returns the n most similar word embeddings of the target
language to the word embedding to be translated, but obviously, not all of these
might be good translation candidates. In order to filter out good candidates,
we determine a threshold for the cosine similarity value. Therefore, we take the
EN → PT, PT → FR, FR → EN dictionaries available from FreeDict,5 using
the TIAD-TSV version we provide under https://github.com/acoli-repo/acoli-
dicts. If both translation directions are available for a language pair, we join the
4
We use the scikit-learn implementation of k-means clustering (https://scikit-
learn.org/stable/modules/generated/sklearn.cluster.KMeans.html) with default pa-
rameters.
5
http://www.freedict.de/
Translation Inference through Multi-lingual Word Embedding Similarity 7
two dictionaries to obtain a higher coverage. As expected, the results in table
1 show decreasing cosine similarity decrease is correlated with lower precision
and higher recall. We chose the cosine similarity threshold 0.97, a rather high
threshold in order to achieve sufficient precision, without suffering too much from
the drop in recall in terms of F1. Yet, considering recall here is debatable: It is
not as expressive in this scenario as it requires a full coverage of the languages
by the dictionary, which is not the case for dictionaries in general (see also the
discussion in the TIAD-2017 overview report [1]).
Table 1. Comparison of Cosine Similarity Thresholds, using FreeDict dictionaries as
evaluation
Translation Threshold Precision Recall F1
0.98 37.00% 12.00% 18.12%
0.97 36.00% 12.00% 18.00%
0.96 35.00% 13.00% 18.96%
FR → PT 0.95 34.00% 13.00% 18.81%
0.9 27.00% 14.00% 18.44%
0.98 47.00% 10.00% 16.49%
0.97 47.00% 11.00% 17.83%
0.96 46.00% 11.00% 17.75%
PT → EN 0.95 46.00% 12.00% 19.03%
0.9 39.00% 13.00% 19.50%
0.98 19.00% 12.00% 14.71%
0.97 19.00% 13.00% 15.44%
0.96 19.00% 13.00% 15.44%
EN → FR 0.95 19.00% 14.00% 16.12%
0.9 17.00% 17.00% 17.00%
4.2 Shared Task Evaluation
We generated translations in one direction per language pair: EN → FR, FR →
PT and PT → EN. We submitted the baseline implementation with a threshold
of 0.97 to the TIAD Shared Task, where we achieved overall precision of 0.64,
recall of 0.22, F1 of 0.32 and coverage of 0.43. Unsurprisingly, all these scores
are substantially higher than the numbers we obtained for evaluation against
FreeDict data. In comparison with the performance of the other participants’
systems, our system shows one of the highest overall performance regarding the
F1-measure, see Tab. 2. It should be noted that the OTIC and W2VEC imple-
mentations were provided by the task organizers as baseline implementations,
but have not been outperformed by any participant system. For our implemen-
tation, a similar performance as OTIC [4] would be expected as we consider our
approach a neural reconstruction of the OTIC approach.
The “One Time Inverse Consultation” (OTIC) method [4] is a graph-based
approach, where lexical overlap in the pivot language is used for scoring trans-
8 Kathrin Donandt, Christian Chiarcos
lation candidates from source and target language. With some degree of simpli-
fication, the score is for a source language word a and a target language word
b is based on the relative number of pivot language elements provided both as
translations of a and b relative to the number of pivot language elements pro-
vided as translations of a or b. In combination with a prediction threshold, this
lexical overlap score is then used for predicting translation pairs.
In our approach, this resembles the way how multilingual embeddings are
constructed: A non-English word is represented as the sum (of the embeddings)
of the (directly or indirectly associated) English words. If two words are identical
in their English translations, they thus have cosine similarity 1, every English
translation they do not share leads (when added to the embedding) to a slight
decrease in this score. In our approach, this decrease will be smaller for seman-
tically related words and larger for non-related words, but if we assume that,
on average, the deviation from a perfect match per non-shared English transla-
tion is equal, the decrease in cosine similarity will directly reflect the number of
non-shared English translations relative to the number of shared English trans-
lations. We suspect that the observed drop in F1 in comparison to OTIC is due
to the construction method of the multilingual embedding space, where dictio-
naries connecting two first-, resp., two second-level languages are being ignored,
as well as due basing this on English as the only pivot language, whereas OTIC
can adopt different pivots depending on the structure of the dictionary graph.
In the light of this explanation, it is somewhat surprising to see that the
evaluation results of the Shared Task organizers seem to confirm the trend that
2nd level languages predict better translation candidates, as previously noticed
in Sect. 4.3: For PT → FR, we get the highest F1 value, even if for both of these
languages, the embeddings are produced in the second iteration of our algorithm.
Table 2. TIAD-2019 Shared Task results, top 5 systems in terms of F1, according to
https://tiad2019.unizar.es/results.html
System Precision Recall F1 Coverage
OTIC 0.64 0.26 0.37 0.45
W2VEC 0.66 0.24 0.35 0.51
FRANKFURT 0.64 0.22 0.32 0.43
LyS-DT 0.36 0.31 0.32 0.64
LyS-ES 0.33 0.30 0.31 0.64
Overall, the scores achieved by all systems (as well as the baselines) in the
TIAD-2019 shared task were considerably low, possibly indicating conceptual
differences in the composition of Apertium data and the blind test set. We
thus performed an additional evaluation on the Apertium data itself. The sense-
clustering extension was evaluated only in this internal evaluation.
Translation Inference through Multi-lingual Word Embedding Similarity 9
4.3 Internal Evaluation
To assess the quality of a newly generated bilingual dictionary by our approach,
we leave out one of the provided dictionaries when calculating the embeddings
for the languages based on the reference language model and try to reconstruct
it using a similarity threshold of 0.97 for predicting translation pairs. Only in
this evaluation, we compare our base implementation with the sense-clustering
extensions, see Tab. 3.
For more than two thirds of the language pairs (72%, 13/18), the sense clus-
tering yields a slight improvement in F1, but only at a marginal scale. In general,
precision and recall vary substantially. We obtain the highest precision for EU
→ EN in both the baseline and extension. The fact that high-precision language
pairs show a (small) drop in precision in the sense clustering extension is possibly
related to the fact that most of them involve the reference language. Dictionaries
including OC have comparably high recall, PT → CA and FR → CA also obtain
high recall. The highest precision values (over 69%) are obtained by dictionaries
including EN for the baseline, precision drops for these dictionaries in the sense
extension, while recall increases in most of them.
The fact that embeddings for a language are not trained on their own data as
it is the case for the reference language might have a small effect on precision, but
the F1 values of language pairs that do include the reference language English
are not in general higher than of pairs that do not. Our “2nd level languages”
AN, AST, FR, IT, OC, PT and RO are not added in the first iteration like CA,
EO, ES, EU and GL (“1st level languages”) and have to be generated using
already projected word embeddings instead of those of the reference language.
The number of pivots used in the projection procedure does not seem to be neg-
atively correlated with the quality of the generated dictionaries though, at least
for those languages we could evaluate in our internal evaluation (we could not
include AN, AST, IT, and RO as they are connected to only one other language,
resp., and thus, as there is no alternative way of getting their embeddings with
our projection procedure if their dictionaries are excluded from the translation
graph): The highest F1 value is obtained by predicting CA and ES, both 1st
level languages, from OC, a 2nd level language, whereas the lowest F1 value is
obtained by predicting FR (2nd level language) from EO (1st level language).
In general, predicting dictionaries with EO as input yields low F1 values, and
all except one (FR → ES) predicted dictionary from 2nd level languages yield
higher recall values than most of the other predicted dictionaries.
5 Discussion & Conclusion
As the results of our internal evaluation (and later on also of the submission)
show, the fact that a word embedding was projected and the number of pivot
languages necessary for the projection do not seem to worsen the quality of the
generation of translation pairs. We have therefore concluded that our approach
is a viable way for generating new translation pairs. We might, however, expect
10 Kathrin Donandt, Christian Chiarcos
Table 3. Internal Evaluation for reference language (EN), first-level languages (CA,
EO, ES, EU, GL), and second-level languages (FR, OC, PT)
Baseline With sense clustering
Dictionary Precision Recall F1 Precision Recall F1
EU, EN 84.25% 31.67% 46.03% 82.83% 30.10% 44.16%
EN, GL 81.05% 24.93% 38.13% 80.00% 28.47% 41.99%
ES, PT 68.29% 37.50% 48.41% 75.85% 35.97% 48.79%
EN, CA 72.09% 30.16% 42.52% 68.18% 31.04% 42.66%
EN, ES 71.19% 40.08% 51.29% 66.48% 43.01% 52.23%
EO, EN 69.15% 19.58% 30.52% 66.16% 18.27% 28.64%
ES, CA 62.32% 39.92% 48.67% 67.83% 37.00% 47.88%
ES, GL 58.17% 44.36% 50.33% 67.59% 38.45% 49.02%
PT, GL 59.17% 39.14% 47.11% 66.27% 37.06% 47.53%
EO, FR 61.07% 12.30% 20.48% 66.15% 17.19% 27.29%
OC, CA 59.03% 59.05% 59.04% 65.94% 52.26% 58.31%
OC, ES 55.03% 54.72% 54.87% 65.35% 55.81% 60.21%
PT, CA 56.26% 57.85% 57.04% 63.15% 53.20% 57.75%
EO, CA 53.40% 25.40% 34.42% 56.22% 31.63% 40.48%
EO, ES 51.47% 30.34% 38.18% 53.13% 35.00% 42.20%
FR, ES 49.88% 23.40% 31.86% 52.96% 25.89% 34.78%
EU, ES 52.34% 33.38% 40.76% 47.97% 36.65% 41.55%
FR, CA 51.34% 56.36% 53.73% 50.26% 61.10% 55.15%
better results when using more sophisticated projection methodologies, for ex-
ample by learning projection matrices jointly for the languages to be projected
into the English embedding space - this remains to be tested. A critical point of
our algorithm is to generate a multi-lingual embedding space in a greedy fashion,
other languages are projected into the space of the reference language: In each
iteration, we only consider the dictionaries whose target language is in the lan-
guage set of already added languages plus the reference language, e.g. we obtain
EO from EO → EN and nothing else. We did not exploit yet any data provided
by the dictionaries EO → CA, EO → ES and EO → FR; the integration of these
dictionaries and therefore, the usage of all available resources would be a more
comprehensive approach and probably improve the quality of our generation
of candidate translation pairs. Choosing an alternative pretrained model for the
reference language with a higher-dimensional embedding space would be another
way to improve – GloVe embeddings are available in higher dimensions as well,
and likewise, alternative embeddings could be explored. Furthermore, one could
use more than just one reference language, and create a joint embedding space
by means of either concatenation (e.g., of English-based and Spanish-based em-
beddings for each word) or by training a joint embedding space as mentioned
above.
The sense clustering we used in the elaboration of our baseline approach
fails to yield significant improvements. One possible explanation for this is that
we have too little input for clustering. Another cause might be that senses are
Translation Inference through Multi-lingual Word Embedding Similarity 11
represented by written representation pairs as described in Sect. 3.2. This re-
sults in more emphasis on certain “senses” (pairs of written representations)
when predicting translations. In general, the remarkably poor performance of
all submitted systems (and the provided baseline implementations) on the blind
test set may reflect the possibly quite distinct nature of training and blind test
data: The blind test data originates from proprietary learner dictionaries pro-
vided by KDictionaries, whereas the Apertium dictionaries that constitute the
training data have been designed for machine translation. Both are similar on a
superficial level, but it is not unlikely that learner’s dictionaries put a stronger
emphasis on covering the different semantic senses in their translations in a
near-exhaustive way (regardless of their real-world frequency), whereas MT dic-
tionaries do have a preference for covering prototypical senses (being ignorant
against infrequent/domain-specific uses may actually improve the system for
most of its applications). In order to quantify the impact of sense granularity
independently from proprietary data, we would encourage the task organizers to
provide an provide open test set along with the blind test set.
As mentioned above, we developed our system originally in the context of
our research project for the purpose of facilitating the search for semantically
similar words. Vector representations of words in a semantic space are particu-
larly promising when searching for remote semantic links, e.g., as required for
cognate detection, but maybe less so for finding literal translations. Yet, as the
evaluation indicates, our system also produces acceptable results for the task
of translation inference across dictionaries in the sense that we rank among the
best-performing implementations for this task.
Our contribution and the key benefit of our approach and is to be seen in
the fact that it is slim, fast, and trivially extensible beyond literal translations:
Translation inference (like cognate detection) requires a simple lookup in the
embedding space. We see that as an important component of on-the-fly cognate
detection, as it allows to identify semantically related form forms over a mul-
tilingual graph of dictionaries without actually traversing this graph at query
time.
Acknowledgments
The research described in this paper was primarily conducted in the project
‘Linked Open Dictionaries’ (LiODi, 2015-2020), funded by the German Ministry
for Education and Research (BMBF) as an Independent Research Group on
eHumanities. The conversion of FreeDict dictionaries into TIAD-TSV data that
we used for estimating the prediction threshold was performed in the context of
the Research and Innovation Action “Pret-a-LLOD. Ready-to-use Multilingual
Linked Language Data for Knowledge Services across Sectors” funded in the
European Union’s Horizon 2020 research and innovation programme under grant
agreement No 825182.
12 Kathrin Donandt, Christian Chiarcos
References
1. Alper, M.: Auto-generating bilingual dictionaries: Results of the TIAD-2017 shared
task baseline algorithm. In: Proceedings of the LDK 2017 Workshops: 1st Work-
shop on the OntoLex Model (OntoLex-2017), Shared Task on Translation Inference
Across Dictionaries & Challenges for Wordnets co-located with 1st Conference on
Language, Data and Knowledge (LDK 2017), Galway, Ireland, June 18, 2017. pp.
85–93 (2017), http://ceur-ws.org/Vol-1899/baseline report.pdf
2. Pennington, J., Socher, R., Manning, C.D.: Glove: Global vectors for word repre-
sentation. In: Empirical Methods in Natural Language Processing (EMNLP). pp.
1532–1543 (2014), http://www.aclweb.org/anthology/D14-1162
3. Ruder, S.: A survey of cross-lingual embedding models. CoRR abs/1706.04902
(2017), http://arxiv.org/abs/1706.04902
4. Tanaka, K., Umemura, K.: Construction of a bilingual dictionary intermediated by a
third language. In: Proceedings of the 15th Conference on Computational Linguistics
- Volume 1. pp. 297–303. COLING ’94, Association for Computational Linguistics,
Stroudsburg, PA, USA (1994)