Vol-3194/paper29

Paper

Paper
edit
description
id	Vol-3194/paper29
wikidataid	→Q117344926
title	Semantic Shift Detection in Vatican Publications: a Case Study from Leo XIII to Francis
pdfUrl	https://ceur-ws.org/Vol-3194/paper29.pdf
dblpUrl	https://dblp.org/rec/conf/sebd/CastanoFMP22
volume	Vol-3194→Vol-3194
session	→

Semantic Shift Detection in Vatican Publications: a Case Study from Leo XIII to Francis

Semantic Shift Detection in Vatican Publications: a
Case Study from Leo XIII to Francis
Silvana Castano1 , Alfio Ferrara1 , Stefano Montanelli1 and Francesco Periti1
1
 Università degli Studi di Milano
Department of Computer Science
Via Celoria, 18 - 20133 Milano, Italy


                                         Abstract
                                         In the recent years, word embedding models are being proposed to effectively detect language change and
                                         semantic shift in diachronic corpora. In this paper, we present a comparative analysis of different word
                                         embedding approaches by considering a case-study based on an Italian diachronic corpus of Vatican
                                         publications of Popes from Leo XIII to Francis (1898-2020). Four different approaches are considered,
                                         characterized by the adoption of different embedding models each one trained over the publications of a
                                         specific pope. The paper aims to explore whether and how word embedding techniques are successful in
                                         detecting semantic shifts over the language used by popes.

                                         Keywords
                                         Computational Humanities, Word Embeddings, Semantic Shift Detection




1. Introduction
In the recent years, the use of machine learning models in the field of Computational History is
gaining more and more attention [1]. In particular, the application of word embedding techniques
to the analysis of historical corpora is providing interesting and promising research results [2].
However, when historical corpora span different time periods, a number of linguistic issues can
emerge. A word can evolve across the years by acquiring/losing meanings or by changing the
context in which it is employed. For examples, the word gay shifted from meaning ‘cheerful’ to
‘homosexual’ during the 20th century, or the word girl having meant ‘young person of either
gender’ in the past [3]. We refer to this process as semantic shift. Although in the past decades
the automatic detection of semantic shift had been already investigating through data-driven
approaches [4, 5], solutions based on word embedding models are currently being proposed
and they are characterized by i) time-oriented splitting of a considered diachronic corpus into
sub-corpora in which a coherent language without semantic shifts can be assumed, and ii)
comparison of word embeddings derived from the sub-corpora to capture the semantic shift of
words across different time periods. These approaches leverage the idea that semantically-related
words are close the one to the others in the embedding space [6]. However, word embeddings
from different temporal vector spaces cannot be naturally compared due to their stochastic

SEBD 2022: The 30th Italian Symposium on Advanced Database Systems, June 19-22, 2022, Tirrenia (PI), Italy
" silvana.castano@unimi.it (S. Castano); alfio.ferrara@unimi.it (A. Ferrara); stefano.montanelli@unimi.it
(S. Montanelli); francesco.periti@unimi.it (F. Periti)
                                       © 2022 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
    CEUR
    Workshop
    Proceedings
                  http://ceur-ws.org
                  ISSN 1613-0073       CEUR Workshop Proceedings (CEUR-WS.org)
�nature. Consequently, different approaches have been proposed to enable the embedding
comparison across different models.

Motivations. In this paper, we present a comparative analysis of different word embedding
approaches by exploiting a diachronic corpus of Vatican publications from Leo XIII to Fran-
cis (1898-2020). The goal of the work is twofold. On one side, we aim at exploring whether
and how word embedding techniques are successful in detecting semantic shifts over official
documents and real documents that address a large audience over a long time period. Moreover,
the paper aims at comparing and discussing the effectiveness of different literature approaches
to capture the semantic shifts on a corpus of limited size and highly unbalanced nature like the
Vatican publications corpus. On the other side, the corpus of Vatican publications represents
a textual dataset of great interest, motivated not only by the exceptional historical depth of
the corpus, but also by two reasons concerned with the nature of Vatican documents. The
first reason is that the Catholic Church, through the writings of its popes, has always dealt
with the most relevant issues in the public debate of its time, alongside the themes of faith and
worship. Therefore, these writings constitute a historical source of primary importance for
reconstructing an important part of the human cultural history. The second reason is related to
the presence in the writings of the Holy See of terms and concepts that are characterized by
a poor semantic shift over time alongside others that have instead remarkably changed both
in terms of relevance and context. The former are mainly terms referring to the dogmas of
faith which, albeit with some variations, essentially remained stable in the discourse of the
popes. On the opposite, the latter are terms that describe well the way in which the attention
of the public discourse shifted over time to different topics, such as the environment, the role
of science, and many historical events of the human history. For these reasons, the corpus of
Vatican documents is a perfect laboratory for experimenting with the techniques of semantic
shift detection and this work constitutes a first step in the investigation of this very rich heritage
of human culture.
   The paper is organized as follows. In Section 2, we discuss the related work. In Section 3, we
present our case-study on Vatican publications. The methods used for the case study analysis is
described in Section 4. The results of the case study are presented in Section 5. In Section 6, we
finally provide our concluding remarks.


2. Related work
As a general remark, word embeddings approaches to semantic shift detection are based on
time-sliced corpora and separate embedding models. The comparison of different word rep-
resentations over time (one per model) is enforced through a distance measure such as for
example the cosine or jaccard similarity. A simple Non-Aligned (NA) method for semantic
shift detection is proposed in [7], where the use of a word over the time is detected through
the analysis of the word context in different time periods. In particular, the idea is to consider
the top-𝑛 neighbors of a word in each temporal embedding model and to measure the overlap
of these lists suggesting that smaller overlaps means drastic changes. However, an alternative
and more typical solution is based on the idea to align word representations (i.e., embeddings)
�which live in different temporal spaces before compared them. In [8], an Incremental Update
(IU) mechanism of the embedding models is proposed. After a model is trained on a first period,
it is then updated with data from the following time periods by saving its state as a new period
model each time. In [9], the idea is to align embedding models to a unique vector space using
heuristic local alignments per word based on the assumption that the set of nearest words in
the embedding space change for words that have a shift. Then, changes between periods are
detected by a distance-based distributional time series for each word in the corpus. The idea
of using a similar transformation in the temporal correspondence problem is proposed in [10],
where, given an input term (e.g., iPod) and a target time (e.g., 1980s), the task is to predict
the counterpart of the query that existed in the target time (e.g., walkman). The approach
in [11] relies on the orthogonal Procrustes (PR) as a global alignment mechanism for temporal
embedding spaces in the evaluation framework of different embedding techniques for detecting
semantic shifts. Further studies attempted to combine information captured by the embedding
models and the frequency of changes for capturing word shifts (e.g., [12, 13]).
    In [14], the idea of creating dynamic embedding models is proposed where data across all the
time periods are shared so that there is no need to align embedding spaces trained on separate
sub-corpora. A Bayesian version of the skip-gram model with a latent time series as prior is
proposed in [14]. Similarly, in [15], the authors propose to extend the skip-gram model by
modeling time as a continuous variable. In [16], a different approach is presented in which
word embeddings for each time period were not first learned, then aligned, but rather learned
and aligned at the same time. As a further approach, the idea of [17] is to train embeddings on
a corpus as a whole while tagging some word of interest with a special tag that indicate which
corpus it comes from. As a result, an individual time-dependent embedding is created for each
target word. To avoid the embedding alignment through orthogonal transformations, in [18],
the authors propose to compute Second-Order embeddings (SO), namely embeddings that
share the same temporal space since obtained by modeling the meaning of words by means of
their semantic similarity relations with all the other words in the vocabulary.
    As a final remark, we note that an increasing interest is emerging about the use of contex-
tualised pre-trained models for semantic shift detection [19, 20]. However, in this paper, such
approaches are not considered since recent comparisons show that static embedding models,
like Word2Vec, outperfomed the contextualised ones for semantic shift detection [21].


3. The Vatican corpus
The considered corpus of Vatican publications contains 27,831 documents extracted from the
digital archive of the Vatican website1 . The corpus consists of all the web-available documents
at downloading time from Leo XIII to Francis (1878-2020) and the popes represent a natural
criterion for splitting the corpus along the time, meaning that a separate sub-corpus is defined
for each pope with associated publications. Furthermore, we stress that the documents have
been downloaded in Italian. This choice is motivated as follows:
    • The documents on the Vatican website are available in various languages, including Italian,
      Latin, English, Spanish, and German. We decided to work with the Italian language since
   1
       https://www.vatican.va
�      a largest number of documents can be obtained in this language (consider that only 14,384
      documents are available in English).
    • In addition, although the official language of the Holy See is Latin, some of the available
      texts are not real official documents of the Catholic Church (e.g., encyclicals, apostolic
      constitutions, letters or exhortations), but they are about official documents of minor
      dogmatic importance (e.g., homilies, audiences, messages, biographies). Again, the number
      of available Latin documents about the publications from popes (i.e., 5,027 texts) is strongly
      less than the number of Italian documents.

A summary description of the considered Vatican corpus is provided in Table 1. Tokens rep-
resent the text units (i.e., words, terms) extracted from the Vatican documents through a text
lowercasing step. As a further feature of the considered Vatican corpus, we note that the size of
the sub-corpora from the popes varies from few documents (e.g., 19 documents from John Paul
I) up to some thousands (e.g., 15,307 documents from John Paul II), meaning that the overall
dataset is an example of unbalanced corpus.
                     Pope           Documents   Paragraphs     Tokens     Papacy beginning
                     Leo XIII           137        3,897       399,470       1878-02-20
                     Pius X              48        1,474       138,336       1903-08-04
                     Benedict XV        166        3,398       248,563       1914-09-03
                     Pius XI             95        4,371       370,728       1922-02-06
                     Pius XII           678       19,074      1,502,876      1939-03-02
                     John XXIII         377       14,014       702,157       1958-10-28
                     Paul VI           3,345      64,285      4,312,149      1963-06-21
                     John Paul I         19         370         23,862       1978-08-26
                     John Paul II     15,307      406,871    22,729,204      1978-10-16
                     Benedict XVI      3,008      68,055      4,713,339      2005-04-19
                     Francis           4,377      101,035     6,374,557      2013-03-13

Table 1
The considered corpus of Vatican publications.




4. Methods for the case study analysis
In this paper, we consider four different literature approaches to semantic shift detection for
application to the Vatican corpus. In particular, we selected a non-aligned (NA) approach [7] and
three different aligned solutions to make comparable the temporal vector spaces of different time
periods, namely Procrustes (PR) [11], Incremental Updates (IU) [8], and Second Order Embeddings
(SO) [18]. In our comparative analysis, the following data processing steps are executed, namely
time-oriented splitting, word embeddings construction and alignment, and semantic shifts detection.

Time-oriented splitting. The Vatican corpus is split by creating a separate sub-corpus for
each pontificate. Due to the short pontificate of John Paul I and the lack of documents from Pius
X and Pius XI, we decide to group their documents with those of the immediately preceding
popes. As a result, we merge the documents of John Paul I with those of Paul VI, the documents
of Pius XI with those of Benedict XV, and the documents of Pius X with those of Leo XIII,
respectively.
�Word embeddings construction. For each one of the considered approaches (i.e., NA, PR,
IU, SO), we train 100-dimensional word embeddings over each sub-corpus by exploiting the
Gensim’s implementation of Word2Vec.2


Word embeddings alignment. For the three aligned solutions, the alignment of embeddings
belonging to separate vector spaces is executed as follows.
   Procrustes (PR). We perform a cross-time alignment through the Procrustes implementation
available at www.github.com/williamleif/histwords. The Procrustes assumption is that each
word space has axes similar to the axes of the other word spaces, and two word spaces are
different due to a rotation of the axes:

                               𝑅(𝑡) = 𝑎𝑟𝑔 𝑚𝑖𝑛𝑄𝑇 𝑄=𝐼 ||𝑄𝑊 𝑡 − 𝑊 𝑡+1 ||𝐹

where 𝑊 𝑡 and 𝑊 𝑡+1 are matrices of word embeddings learn at year 𝑡 and 𝑡 + 1 respectively,
and Q is an orthogonal matrix that minimizes the Frobenius norm of the difference between
𝑊 𝑡 and 𝑊 𝑡+1 [11].

   Incremental Updates (IU). We consider the model on the sub-corpus related to Leo XIII (the
first pope in the dataset by time), and then we update the model with data of subsequent popes
saving its state each time as a new pope model. Each model 𝑚𝑖+1 is initialized with the word
vectors from the previous model 𝑚𝑖 [8].

   Second Order Embeddings (SO). As proposed in [18], we build second order embeddings by
modeling the words by means of their semantic similarity relations with all the other words
in the vocabulary. Denoting an embedding of a word 𝑤𝑖 at time period 𝑡 as 𝑤𝑖 (𝑡) ∈ R100 we
consider the vectors:

                    𝑤˜𝑖 (𝑡) = sim(𝑤𝑖 (𝑡), 𝑤1 (𝑡)), ..., sim(𝑤𝑖 (𝑡), 𝑤|𝑉 | (𝑡)
                             (︀                                               )︀


where 𝑉 is a common vocabulary of all the words in all the time periods and sim is a similarity
function such as the cosine similarity. For computational purposes, we define the common
vocabulary 𝑉 by relying on mutual information values computed between words and classes
of text (e.g., encyclicals, apostolic exhortations, homelies) associated with each pope. For
each class of text, we select the top-500 words by mutual information score. Similarly to the
experiment performed in [18], we only keep words associated with nouns, adjectives, and verbs.
Furthermore, we exclude stopwords and words shorter than 4 characters.

Semantic shifts detection. Word vectors from distinct time-sliced models cannot be directly
compared due to the stochastic nature of Word2Vec. This issue does not preclude the comparison
of distances between pair of words over time, which means that it is possible to compare the
semantic similarities of a pair of words in distinct models. For the sake of clarity, as an example,
we consider the case of temperature. Temperatures from different scales, such as Celsius and
    2
        https://radimrehurek.com/gensim/
�Kelvin, cannot be directly compared. They need to be aligned, i.e., one has to be converted to
the scale of the other. However, since scales are related to an additive constant, we can directly
compare deltas of temperatures computed in different scales.
  Similarly, we decide to exploit:
   1. non-aligned embeddings to analyze the relative position of word pairs (i.e., the distance
      between their vectors) in different vector spaces. With respect to the above example, this
      corresponds to compare temperature deltas in different scales;
   2. aligned embeddings to analyze the positions of a word over time (i.e., the distance between
      the vector of that word and itself in distinct aligned vector spaces). With respect to the
      above example, this corresponds to convert a temperature from a scale 𝑠1 to another 𝑠2
      before comparing it with another temperature in scale 𝑠2.
  Pairwise word similarity. We exploit non-aligned embeddings to compute the pairwise cosine
similarity between a pair of word vectors 𝑤1 and 𝑤2 across time in two different models 𝑚𝑖
and 𝑚𝑗 . In particular, as we chronologically trained the models pope by pope (they follow each
other over time without overlapping) we show how cosine similarity between word vectors
could highlight the strength of the relationship in the perspective of different popes.
                                                       𝑤1𝑚𝑖 · 𝑤2𝑚𝑗
                            𝑐𝑜𝑠𝑖𝑛𝑒(𝑤1𝑚𝑖 , 𝑤2𝑚𝑗 ) =
                                                     ||𝑤1𝑚𝑖 || ||𝑤2𝑚𝑗 ||

   Word context comparison. We exploit non-aligned embeddings to explore the context of words.
Given a word 𝑤, we investigate the top-𝑛 words corresponding to the 𝑛 closest vectors to the
vector of 𝑤 (i.e. the 𝑛 most similar words to 𝑤) in each embedding model. In other words, the 𝑛
closest vectors to the vector of 𝑤 are the top-𝑛 vectors with highest cosine similarity value from
that vector. Besides learning how neighbors change over time for different popes, we estimate
the context similarity of a given word 𝑤 between each pair of popes by computing the jaccard
similarity score between the 𝑛 most similar words to 𝑤 in their respective models 𝑚𝑖 and 𝑚𝑗 .

                                                       |𝑡𝑜𝑝-𝑛𝑚𝑖 ∩ 𝑡𝑜𝑝-𝑛𝑚𝑗 |
                      𝑗𝑎𝑐𝑐𝑎𝑟𝑑(𝑡𝑜𝑝-𝑛𝑚𝑖 , 𝑡𝑜𝑝-𝑛𝑚𝑗 ) =
                                                       |𝑡𝑜𝑝-𝑛𝑚𝑖 ∪ 𝑡𝑜𝑝-𝑛𝑚𝑗 |

  Self word similarity. The need of aligned embeddings rises to mutually compare words
over time. By relying on the cosine similarity, we detect meaning change independent from
neighboring words by considering the self similarity of a word 𝑤 throughout consecutive time
models 𝑚𝑖 , 𝑚𝑖+1 .
                                                     𝑤𝑚𝑖 · 𝑤𝑚𝑖+1
                          𝑐𝑜𝑠𝑖𝑛𝑒(𝑤𝑚𝑖 , 𝑤𝑚𝑖+1 ) =
                                                   ||𝑤𝑚𝑖 || ||𝑤𝑚𝑖+1 ||

5. Case study results
In this section, we discuss the results of the approaches presented in Section 4 for semantic shift
detection applied to the Vatican corpus of Section 3 according to pairwise word similarity, word
context comparison, and self word similarity. similarity.
�   One of the main problems in evaluating the results is that it is difficult to define a ground
truth that provides information about the expected shifts in the Vatican corpus. To address this
issue, we run our tests by exploiting three main categories of words.
    • Words representing long-term concepts in the Vatican publications (e.g., jesus,
      eucharist, ...). These terms represent central concepts in the Church, usually related to
      theological issues. For those terms, then, we expect to observe a limited shift of meaning
      in the publications of the different popes.
    • Words representing concepts from the past (e.g., heresy, perversion, ...). These
      terms are related to topics that have been central in the Vatican publications in the past,
      but that nowadays are less present in the popes publications in favour of new words that
      are more strictly related to events and social phenomena that are perceived as important
      at the present time. For those terms, we expect to observe a decreasing trend along the
      temporal dimension.
    • Words representing concepts from today (e.g., environment, science, ...), that are
      the opposite of the concepts from the past, namely words representing concepts that are
      important nowadays for which we expect to observe a growing trend along time.
   For the sake of readability, the considered words from the Vatican corpus are translated from
Italian to English.

Pairwise word similarity. In Figure 1, we show examples of word pairs taken from long-term
concepts (first row), concepts from the past (second row), and concepts from today (third row),
respectively. For each pair of words, we compare their cosine similarity in the models trained
on the different popes, exploiting both aligned and non-aligned embeddings. The relevant issue
in this experiment is that the comparison between words is based exclusively on their relative
position in the vector space. As a consequence, we do not have any information about the
stability of the meaning of each single word per se. The only information available is about
the meaning of a word with respect to the other in the same pair. As typically occurs for word
embedding methods, the proximity assumption holds. Thus, if two words are similar (i.e., their
are close in the vector space) we can derive that their meaning is also similar since the two
words are used in a similar context.
   Concerning long-term concepts (first row of Figure 1), the cosine similarity values for each
word pair are stable in time. In particular, we note that the pairs are essentially composed
by a word and its consolidated epithet (i.e., Virgin Mary, Jesus Christ) or alias (i.e.,
Eucharist, also called Most Blessed Sacrament). Such similarity values suggest that
the meaning shift for these long-term concepts is limited as expected.
   For the concepts from the past, the trend of the pairwise similarity between the considered
words is decreasing. In particular, we note that pairs having a strong similarity in the past can
be characterized by a lower similarity in the publications of recent popes, like ‹perversion,
novelty›. This means that these words were originally used in the same context, but their
linguistic and thematic context is changed along time, either because they are no more used
together or because one of the two, or both, are now rarely used.
   For the concepts from today, we observe the opposite phenomenon. The similarity of word
pairs increases over different pontificates, suggesting that the cultural changes that characterize
�Figure 1: Pairwise Word Similarity. The cosine similarity values between pairs of words in different
embedding models. The first, second, and third row shows pairs of words whose similarity value is
i) almost stable, ii) decreasing, and iii) increasing over time (over different pontificates), respectively.
Breaks in curves appear if a vector could not be found corresponding to a word 𝑤 in a certain model,
i.e., in a certain pope, or there was too little evidence for 𝑤. The values of the cosine similarity can be
found on the y-axis of each plot.


the 21th century have induced popes to increasingly use the two terms in similar contexts. A
pretty clear example of this behavior is given by the pair ‹science, technique› where we note
that the trend of the word technique is to become almost a synonym of the word science,
but only after the 70s with John Paul II. In this respect, it is also interesting to consider the new
words closely related to a certain pair introduced by a pope in comparison with the dictionary
of the previous one. The new words of a pope are determined as the set difference between
the subset 𝑠 of the vocabulary of a pope 𝑚𝑖 and the entire dictionary of the previous pope
𝑚𝑖−1 , where 𝑠 is the set of the 30 words closest to the mean vector of a certain word pair in the
embedding model related to 𝑚𝑖 . For example, with respect to the pair ‹environment, planet›,
Francis introduced the words amazonia, biodiversity , deforestation, ecosystem,
energetic, and oceans. With respect to the pair ‹sex, gender›, Francis introduced the words
mistreatment, and homosexuals; while for the pair ‹science, technique› John Paul II
introduced the words astronomy, biology, biomedical, branch, cosmology, computer
science, engineering, molecular, psychiatry, technological.
   As a further remark, we note that the same trend result in the relative position of the consid-
ered word pairs can be detected either using aligned or non-aligned embeddings.3
    3
        The breaks in the lines do not appear in IU models due to the main limitation of this approach: the recognition of
�Word context comparison. In Figure 2, we consider the target words jesus, environment,
heresy and we explore their context composed of 1, 5, and 10 most similar words in the different
embedding models, namely the words corresponding to the 1, 5, 10 vectors that closest to the
vector of the target. The color gradation describes the intensity of the jaccard similarity value
between any pair of popes. Obviously, the diagonal always shows the darkest color since the
jaccard similarity value between a pope and himself is equal to 1. About the word jesus,
the top-1 plot of Figure 2 show that all the popes except Francis share the most similar word.
This result confirms the observations of pairwise word similarity reported in Figure 1, where
the pair ‹jesus, christ›is almost unchanged from Leo XIII to Benedict XVI. Also when the




Figure 2: Word Context Comparison. Heatmaps of the jaccard similarity values computed between
any pair of popes about jesus (first row), environment (second row), and heresy (third row) with
respect to their top 1 (left column), 5 (middle column), and 10 (right column) similar words.


contexts of top-5 and top-10 words are considered, the stable behavior of jesus can be observed
over different pontificates (i.e., many dark areas can be recognized on the first row). On the

a word change is possible only if the word has enough occurrences in the considered time period. If the occurrences
of a word dramatically decrease (or completely disappear), its word vector will remain the same and hence it is not
possible to observe any change [8].
�opposite, the words environment and heresy are affected by semantic shifts. About the
word environment a shift can be observed in both Paul VI and Pius XII. About the word
heresy, the shift is less pronounced. The context of the word heresy is more similar in the
popes of the past, rather than in those of the recent periods. As an exception, the similarity
values between John Paul II and Benedict XVI denote a semantic shift in the context of the
considered target words. Exploring the closest common words to heresy from Benedict XV
and Pius XII, we find nestorio (i.e., the name of an Archbishop of Constantinople from which
Nestorianism - a doctrine condemned as a heretic by the Council of Ephesus in 431 - takes its
name), condamnation, apostasy. When John Paul II and Benedict XVI are considered, the
closest common words to heresy are arian and arianism that are about the heresy of Ario,
condemned as a heretic by the first council of Nicaea in 325.

Self word similarity. In this experiment, we consider the position of a word with respect
to itself, by measuring the similarity of a word vector at time 𝑡 with respect to the vector of
the same word at time 𝑡 − 1. In particular, in Figure 3, we observe the trend of the self cosine
similarity for the words environment, travel, and progress.The similarity measures are
computed by exploiting both the aligned methods (blue line) and the non-aligned one (green
line). Since Leo XIII is the first Pope of our corpus, it is not possible to calculate the self cosine
similarity of a word with respect to the model of the previous Pope. For this reason, the lines
reported in the figure start from Benedict XV instead of Leo XIII. Since in this experiment we
compare a word with itself in different models, we expect to observe high values of similarity
with a limited variation. However, this expectation is confirmed only for the aligned methods SO
and IU. This is due to the fact that independent models trained on different corpora of different
periods can be directly compared only when models are aligned as it occurs in SO and IU. In
the case of Procrustes PR, the low values of the self cosine similarity reveal that the alignment
mechanism adopted by this method is not suitable for small-sized, unbalanced datasets like
the considered Vatican corpus. According to the literature, low values of self similarity can be
associated with a semantic change of the considered word, while high values of self similarity
denote stable word meanings. As a result, we claim that successive increasing values of self
similarity suggest a strengthening of the word meaning, while successive decreasing values of
similarity suggest a weakening of the word meaning. About the considered target words, we
note that the trends of the self cosine similarity are different for IU and SO models, but they
share the increasing/decreasing direction of some shifts, such as for example between Paul VI
and John Paul II for the word environment. This can be interpreted as a consolidation of the
word meaning. Furthermore, both SO and IU models share shifts between John XXIII and Paul
VI for the word progress, but this behavior is less evident in the SO model. This can be due to
the dimensionality reduction applied when the second order embeddings are built.


6. Concluding remarks and future work
In this paper, we considered different approaches to semantic shift detection and we discussed
the results obtained on a corpus of Vatican publications related to popes from Leo XIII to Francis
(1878-2020). The results show that word embedding can be successfully employed in semantic
�Figure 3: Self cosine similarity. The self cosine similarity of the target words environment, travel,
and progress calculated over subsequent time/pope models according to Second Order embedding
(SO), Incremental Updates (IU), and orthogonal Procrustes (PR) (blue lines). (green) The self cosine
similarity of the same target words calculated over subsequent time/pope models according to the
non-aligned models (green lines)


shift detection, even when a small-sized, unbalanced dataset is considered like the Vatican
corpus. Both aligned and non-aligned approaches have been exploited in the proposed case
study. The results reveal that the alignment of embedding models over different vector spaces
is not required when we consider pairs of words belonging to different time periods. On the
opposite, to successfully detect the meaning shift of a word along time over different vector
spaces require the adoption of an alignment mechanism, so that the word vectors belonging
to different periods are comparable. However, when alignment approaches are adopted, our
results show that the change of a word over time can be noisy and the interpretation of the
word behavior can be difficult (e.g., see the case study results of the Procrustes method when
the self word similarity is considered).
   Ongoing and future work are focused on exploring semantic shift detection techniques by
relying on contextualized word embedding models like BERT. In this direction, BERT-like
models allow to capture the sense differentiations of a target word, meaning that they can detect
the different meanings of the considered target according to the different contexts in which
the word is used throughout the whole corpus. Furthermore, contextualized embeddings can
leverage the benefits of existing pre-trained models, thus avoiding the execution of a (costly)
training phase over each time-sliced sub-corpus.
�Acknowledgments
This paper is partially funded by the RECON project within the UNIMI-SEED research pro-
gramme.


References
 [1] C.-m. Au Yeung, A. Jatowt, Studying how the Past is Remembered: Towards Computational
     History Through Large Scale Text Mining, in: Proc. of the CIKM, ACM, 2011, pp. 1231–1240.
 [2] J. Bjerva, R. Praet, Word Embeddings Pointing the Way for Late Antiquity, in: Proc. of the
     LaTeCH, ACL, 2015, pp. 53–57.
 [3] N. Tahmasebi, L. Borin, A. Jatowt, Y. Xu, S. Hengchen (Eds.), Computational Approaches
     to Semantic Change, LSP, 2021.
 [4] E. Sagi, S. Kaufmann, B. Clark, Tracing Semantic Change with Latent Semantic Analysis,
     Current ethods in historical semantics 73 (2011) 161–183.
 [5] S. Mitra, R. Mitra, M. Riedl, C. Biemann, A. Mukherjee, P. Goyal, That’s sick dude!:
     Automatic Identification of Word Sense Change across Different Timescales, arXiv preprint
     arXiv:1405.4392 (2014).
 [6] T. Mikolov, K. Chen, G. Corrado, J. Dean, Efficient Estimation of Word Representations in
     Vector Space, in: ICLR Workshop Papers, 2013.
 [7] H. Gonen, G. Jawahar, D. Seddah, Y. Goldberg, Simple, Interpretable and Stable Method for
     Detecting Words with Usage Change across Corpora, in: Proc. of ACL, 2020, pp. 538–555.
 [8] Y. Kim, Y.-I. Chiu, K. Hanaki, D. Hegde, S. Petrov, Temporal Analysis of Language through
     Neural Language Models, arXiv preprint arXiv:1405.3515 (2014).
 [9] V. Kulkarni, R. Al-Rfou, B. Perozzi, S. Skiena, Statistically Significant Detection of Linguistic
     Change, in: Proc. of WWW, 2015, pp. 625–635.
[10] Y. Zhang, A. Jatowt, S. Bhowmick, K. Tanaka, Omnia Mutantur, Nihil Interit: Connecting
     Past with Present by Finding Corresponding Terms Across Time, in: Proc. of ACL, 2015,
     pp. 645–655.
[11] W. L. Hamilton, J. Leskovec, D. Jurafsky, Diachronic Word Embeddings Reveal Statistical
     Laws of Semantic Change, arXiv preprint arXiv:1605.09096 (2016).
[12] I. Stewart, D. Arendt, E. Bell, S. Volkova, Measuring, Predicting and Visualizing Short-Term
     Change in Word Representation and Usage in VKontakte Social Network, in: Proc. of
     ICWSM, 2017.
[13] A. Englhardt, J. Willkomm, M. Schäler, K. Böhm, Improving Semantic Change Analysis by
     Combining Word Embeddings and Word Frequencies, International Journal on Digital
     Libraries 21 (2020) 247–264.
[14] R. Bamler, S. Mandt, Dynamic Word Embeddings, in: Proc. of the ICML, 2017, pp. 380–389.
[15] A. Rosenfeld, K. Erk, Deep Neural Models of Semantic Shift, in: Proc. of the NAACL-HLT,
     ACL, 2018.
[16] Z. Yao, Y. Sun, W. Ding, N. Rao, H. Xiong, Dynamic Word Embeddings for Evolving
     Semantic Discovery, in: Proc. of the WSDM, ACM, 2018, pp. 673–681.
[17] H. Dubossarsky, S. Hengchen, N. Tahmasebi, D. Schlechtweg, Time-Out: Temporal
�     Referencing for Robust Modeling of Lexical Semantic Change, in: Proc. of the ACL, ACL,
     2019.
[18] S. Eger, A. Mehler, On the Linearity of Semantic Change: Investigating Meaning Variation
     via Dynamic Graph Models, arXiv preprint arXiv:1704.02497 (2017).
[19] R. Hu, S. Li, S. Liang, Diachronic Sense Modeling with Deep Contextualized Word Embed-
     dings: An Ecological View, in: Proc. of the ACL, ACL, 2019.
[20] S. Montariol, M. Martinc, L. Pivovarova, Scalable and Interpretable Semantic Change
     Detection, in: Proc. of the NAACL, ACL, 2021.
[21] D. Schlechtweg, B. McGillivray, S. Hengchen, H. Dubossarsky, N. Tahmasebi, SemEval-
     2020 Task 1: Unsupervised Lexical Semantic Change Detection, in: Proc. of the SemEval,
     ICCL, 2020.
�