=Paper=
{{Paper
|id=Vol-1749/paper23
|storemode=property
|title=Spammare senza pietà – Corpus based analysis of English, unacclimatised verb Loans in Italian and Creation of a Reference Lexicon
|pdfUrl=https://ceur-ws.org/Vol-1749/paper23.pdf
|volume=Vol-1749
|authors=Anna Fantini
|dblpUrl=https://dblp.org/rec/conf/clic-it/Fantini16
}}
==Spammare senza pietà – Corpus based analysis of English, unacclimatised verb Loans in Italian and Creation of a Reference Lexicon==
<pdf width="1500px">https://ceur-ws.org/Vol-1749/paper23.pdf</pdf>
<pre>
       Spammare senza pietà - Corpus based analysis of English, un-
    acclimatised verb loans in Italian and creation of a reference lexicon

                                        Anna Fantini
                                 Università degli studi di Pavia
                          anna.fantini01@universitadipavia.it

                                                        cal, phonetic, and lexical – in the language is still
                       Abstract                         in progress. For instance, they tend to retain the
                                                        phonetic property of the original word, especially
     English. We describe the lexical resource          of the lexical root (to spawn > spawnare
     created to investigate the semantic changes of     /spo’nare/).
     90 English, un-acclimatised verb loans in          The new word serves as an alternative – usually
     Italian. Final results and interesting observa-    a hyponym – of an already existing term1. As for
     tions concerning the annotation task are dis-      the concept of loan acclimatisation, the literature
     cussed.
                                                        states that it involves the role of the new term in
                                                        the target language. Therefore, Gusmani speaks
     Italiano. Descriviamo la risorsa lessicale
     creata per indagare in italiano il cambiamen-
                                                        of acclimatisation only with regard to the lexicon
     to semantico di 90 prestiti verbali inglesi non    and its connection with speakers’ usage: the
     acclimatati. Illustriamo i risultati finali e le   more they familiarise with the loan, the more the
     interessanti osservazioni emerse dall'esperi-      latter gets acclimatised. It follows that – for very
     mento di annotazione.                              recent, scarcely integrated loans – the majority of
                                                        speakers, as well as linguistic authorities, do not
1     Introduction                                      perceive the influence of English as an enrich-
                                                        ment of the lexical heritage but mostly as a nui-
The case of language borrowing was investigated         sance. If it is true that a number of reports de-
in depth by Gusmani (1983), who argues that a           scribe the interference of English over Italian as
linguistic loan is an interference phenomenon,          an impoverishment, some attempts have also
connected with contact and mutual influence of          been made to study the less acclimatized loans
different languages. According to his study, the        themselves. It is thus of interest to examine why
motivations behind the origin of a loan lie in the      this kind of loan infiltrates the Italian language,
individual act of a speaker or of a group of            how the speakers cope with the new word and
speakers. The need to resort to a foreign alterna-      what is the semantics of the loan in the target
tive derives from the prestige held by the latter       language. The aim of the present paper is to give
against an equivalent word in the mother tongue         a detailed account of how the meaning of a verb
of the speaker (or from the absence all together        loan changes (and if it changes) in the target lan-
of an alternative, as in our work: “Se mi vede,         guage and to offer a reliable source of lexical
Miki mi banna (<to ban)” vs. *“Se mi vede Miki          information in the form of an electronic lexicon
mi bandisce”).                                          built for the occasion. Section 2 details the meth-
Facts show that language borrowing is particular-       od used to collect suitable data; section 3 illus-
ly common among specialized languages, more             trates the structure and functions of the lexicon;
so if they are linked to technical contexts.            section 4 provides the results of our analysis as
This is extremely visible within the computer           well as the annotation task performed with our
context. The main focus of this paper is the in-        data; section 5 discusses our findings and section
formal variety of Italian as used by communities        6 finally provides a conclusion.
of online video-gamers, computer experts and
amateurs, forum users, etc.; a specialized lan-         2       Methods
guage linked to technical context populated with
partially integrated and un-acclimatised English        In order to investigate the semantics of English
verb loans.                                             un-acclimatized verb loans, we examined their
These kinds of (mostly) lexical influences are so       occurrence in a monitor web-corpus created for
recent that their structure is hardly stable, and the
process of integration – graphical, morphologi-         1
                                                            E.g. googlare < to google as hyponym of cercare.
this purpose, following the guidelines and in-                    3     The Lexicon
structions of previous Corpus Linguistics works
(Baroni and Kilgarriff, 2001; Lenci et al., 2012;                 After extracting the semantic patterns for each
McEnery, Xiao, Tono, 2006; Pomikálek, 2011;                       lemma from the corpus, we stored the informa-
Pustejovsky and Stubbs, 2012). The corpus con-                    tion in an electronic lexicon, built using the
tains 6 transcriptions from a total of 194,07                     software Personal Lexicon 2.7.1, a language
minutes of audio material, collected with consen-                 learning resource developed by Alexander Smith
sual but unaware recordings and then transcribed                  between 2007 and 2015. The software comes
using the software Elan 4.9.6 (Wittenburg et al.,                 both in free and registered versions, the current
2006), plus 129 texts obtained through the                        lexicon has been compiled – and it will be con-
Sketch Engine web-crawler, suitably set. We ex-                   sultable – using the free version.6
tracted a sum of 90 different verb lemmas (542                    The lexicon is designed to give a precise ac-
different word forms), for a total of 1327 occur-                 count of every semantic feature and every mean-
rences. The annotation involves a POS level –                     ing variation of the verb loans. As the reader will
limited to the sentence containing the loan – a                   see observing Figure 1, each entry is character-
loan-type level – describing three degrees of lan-                ized by the following elements (some of them
guage integration2 – a semantic type level3 and a                 pre-named          in         the        software):
thematic role level 4 . The last two levels have
been annotated using the tags proposed in Jezek
and Nissim (2014) and Jezek and Vieu (2013)
respectively.
Every text has been annotated using the Mae
software (Stubbs 2011). An annotation task was
conducted using a sample of the corpus (see sec-
tion 4.2), its agreement result being only partially
positive but interesting nonetheless from a lin-
guistic point of view.
The next part of our research involved the analy-                    “Figure 1. The spammare 2b lexical entry”
sis of the semantic patterns for each lemma5, thus                     The entry citation form, with the number
compiling one or more data-driven senses for                             of the sense or of the sub-sense7;
every verb. The senses obtained were classified
                                                                       The Pronunciation of the citation form;
according to Verb Net’s semantic class hierar-
                                                                       The Class (pre-named) as in the loan type
chy. The assumptions underlying this in-
                                                                         which it belongs to (whether it is fully in-
vestigation are grounded on Corpus Pattern
                                                                         tegrated or only partially integrated);
Analysis (CPA) and Computational Lexicogra-
phy (Hanks 2008; Hanks 2012; Jezek 2011).                              The Root element, as in the lexical English
Verb patterns have – in general – the following                          root it comes from;
structure, where:                                                      The Theme (pre-named), as in the Verb
(1) Spammare 2b                                                          Net class it was reduced to;
Agent[PERSON]                         V_spammare                       The Definition box, containing the lexical
(Theme[ARTEFACT | ABSTRACT]).                                            definition and the verb pattern;
We have chosen all uppercase for the semantic                          The Related entries in the lexicon, all ac-
type, and first letter uppercase for the thematic                        cessible through hyperlink;
role, extended to every argument of the verb.                          The Personal Examples, used to extract the
Round brackets contain the possible optional                             pattern.
arguments of the verb.                                            In figure 2 we show the Conjugations tab (pre-
                                                                  named) that includes all the syntactic comple-
2
                                                                  ments of the verb and their semantic properties
  Totally integrated, e.g. spammare; partially integrated         (thematic roles and semantic types).
(grafic), e.g. trackare; partially integrated (phonetics), e.g.
spawnare /sp'nare/.
3
  E.g. Person, Artefact, Location, Abstract, etc.
4
  E.g. Agent, Patient, Goal, Source, Duration, ect.
5                                                                 6
  From Hanks, Pustejovsky 2005, a pattern is intended here         The resource is not yet available for public consultation..
as an argument structure with specifications of both the          7
                                                                   Sub-senses corresponding to GRADIT’s accezioni, the
thematic roles and the semantic types of each argument            progressively numbered paragraphs a sense is pos-sibly
positions.                                                        divided into.
                                                            Group 2 Diversified sense           25

                                                            Group 3 New sense                   26

                                                                     New v. and new sense 7

                                                          Senses     157
   “Figure 2. Lexicon conjugation tab”                      “Table 1. Senses sorted according to their se-
We listed the grammatical subject, the direct ob-       mantic behaviour”
ject, the indirect object and up to ten different           Group 1 coincides with 63% of the total (99
indirect complements. Notice that the semantic          senses out of 157), 11 senses have also new verb
type slot may specify more than one element, in         forms (bishottare, autospottare), 78 senses occur
which case we used the | separator.                     in 1 to 10 examples – they often have new verb
In figure 3 we illustrate the Themes section of         forms (riloggare) or a very specific meaning
the lexicon, with a partial list of Verb Net classes    (rippare)8. In this group there is the highest per-
and sub-classes used in the resource.                   centage (41%) of monosemic verbs.
                                                            Group 2 coincides with 15% of the total (25
                                                        senses out of 157), 19 senses occur only in 1 to
                                                        10 examples – they do not have a very specific
                                                        meaning, but may be considered as hyponyms of
                                                        Italian verbs (farmare2 of sfruttare) 9 . In this
                                                        group there is the lowest percentage (16%) of
                                                        monosemic verbs – their distribution proved not
                                                        to be directly proportional to the senses’ quanti-
                                                        ty.
                                                            Group 3 coincides with 25% of the total (33
“Figure 3. Lexicon Theme section”                       senses out of 157), 24 senses occur only in 1 to
Clicking on each one produces the list of entries       10 examples yet we also have the senses occur-
belonging to that particular class; this list appears   ring in the highest number of examples (drop-
in the third section of the lexicon, the Lexical        pare 1 10 and 2 11 with 189 examples). In this
Items column storing all the entries ordered al-        group 27% of the verbs are monosemic.
phabetically.
                                                        4.2     The inter annotator agreement
4     The Results                                          The semantic annotation task was conducted
In this section we report the results of both the       following the methodology of Pustejovksy and
semantic analysis of the loans and the annotation       Stubbs (2012); only a sample of 440 random oc-
task.                                                   currences was annotated by 9 groups of an-
                                                        annotators, each constituted by 3 people. They
4.1    Quantitative considerations                      were given guidelines explaining the method and
   The lexicon contains 157 senses for a total of       the tagsets, and they were asked to separately
90 verbs. As shown in Table 1, the 157 senses           annotate the semantic type and the thematic role
have been classified into 3 groups according to         of each verbal argument. We used Fleiss’k algo-
three main criteria about the degree of semantic        rithm to calculate the agreement12 (Artstein and
conservativeness of the loan:                           Poesio, 2008), the values being interpreted ac-
   1 The meaning remains the same as the origi-         cording to Landis and Koch (1977). We already
nal verb.                                               said that the results have been only partially posi-
   2 The meaning remains linked to the original         tive, in particular – as for the thematic role – on-
one, but it diversifies to some degree.
   3 The meaning changes to the point that it be-
                                                        8
                                                          E.g. “Non potendo accedere al CD-Rom non posso rippa-
                                                        re niente” .
comes a new meaning altogether                          9
                                                          E.g. “Ok, farmerò i campi di battaglia eterni”.
  Group # Type                        Numbers           10
                                                           E.g. “Non molto tempo fa ho droppato i bracciali”.
  Group 1 Same sense                  88                11
                                                           E.g. “Non è difficile droppare un computer privato”.
                                                        12
                                                           We choose Geertzen, J. (2012) online resource for
             New verb form            11                agreement evaluation .
ly one group reached the 0.6 threshold consid-               cloccare 14 (potenziare and overcloccare being
ered acceptable with semantic annotation, the                almost synonyms). The thematic role level is the
others showing moderate agreement and fair                   most problematic, obtaining substantial agree-
agreement (one group only). For the semantic                 ment only in one case; the semantic type level on
type annotation, three groups reached the 0.6                the other end is perceived as a less abstract, more
value, four groups showed moderate agreement                 transparent concept and the annotation is slightly
and two groups showed fair agreement. Nonethe-               better, with three groups over 0.6. What is really
less, we could make interesting linguistic consid-           interesting is that the group which performed
erations.                                                    best with thematic roles is also the one which did
                                                             worst with semantic types. Moreover, the groups
5     Discussion: the semantic behaviour of                  which performed best with semantic types
      the loans                                              showed only fair to moderate agreement in the-
                                                             matic roles. We observe a – general and group
Let us consider the case of spammare13: all its              wise – performance improvement with the se-
three main senses are distributed among the three            mantic type level.
groups mentioned in Table 1, the semantic be-                This is because assigning a thematic role requires
haviour shows not only a certain degree of con-              a deeper reflection and some of the roles may be
servativeness, but also a great degree of diversi-           ambiguous (for example, Beneficiary and Goal).
fication (just 18 occurrence out of 76 keep the              The creation and combination of more specific
original meaning) and thus of acclimatisation.               sub-types and sub-roles – targeted to this kind of
Italian speakers apply a saving strategy: the                verbs – could help resolve the ambiguity hinder-
monosemous loans are also the most conserva-                 ing agreement (for example, Person split into
tive, while diversification often results in                 Authority and Subordinate)15.
polysemous verbs – it seems that, once a seman-                 Furthermore the un-acclimatisation of the
tic change starts, the speaker continues to use the          loans leads to somewhat different uses and dif-
loan until it reaches a definitive meaning, even-            ferent meanings among the speakers. This hap-
tually becoming acclimatized. The cognitive ef-              pens either between different communities, either
fort behind this process is very high, but it also           between different speakers of the same commu-
implies a certain linguistic confidence. Of course           nity. Other significant observations emerged on
it is less arduous to produce a loan whose sense             the frequency of roles and types and on their co-
is strictly linked to the original verb's one, thus          occurrence: the most used roles are Agent (often
generating many monosemous loans. Neverthe-                  erroneously) and Patient (often in the place of a
less we wonder whether – aside from being eco-               more neutral Theme). The types most used are
nomically convenient – is it also strategically and          Person, Artefact and Abstract. The Agent-Person
linguistically sensible to produce just monose-              combination is the most frequent, even if the role
mous loans instead of using semantically diversi-            is often wrongly assigned. Great uncertainty
fied ones. Is it sensible to keep numerous and               emerged in assigning the correct type to argu-
specific loans, when there can be fewer and pol-             ments whose referents are intangible informatics
ysemous ones? Further investigations of English              entities, e.g. nicknames, server, updates, etc. or
un-acclimatised verb loans may answer part of                characters of a game, e.g. boss, Pokémon, etc.
these questions.                                                Last but not least, it was possible to already
5.1    Interesting observations about the an-                identify primitive verb classes, depending on the
       notation task                                         roles and the types assigned to verb arguments,
                                                             i.e. verbs of change of state with a Patient role
We feel that the only partially satisfying results           and possibly a Beneficiary role, or verbs of crea-
may depend on the tricky lexical meaning of                  tion with a Result role and occasionally an Agent
each loan. It is clearly easier to annotate the ar-          role.
gument structure of a well-known verb like po-
tenziare, rather than the one of the loan over-


                                                             14
                                                                E.g. “Prima di lanciare il tutto ho overcloccato la scheda
13
   E.g. 1 “Magari capiterà di spammare un oggetto”           video”.
2 “Iniziano a spammare pubblicità a tutti gli iscritti.”     15
                                                                From section 1 “Se mi vede Miki mi banna” (bannare 1a)
3 “Questa è un'abilità controversa [...] ma non va spamma-   Miki annotated as Agent-Authority and the personal pro-
ta” .                                                        noun as Beneficiary-Subordinate.
6    Conclusions
                                                           Thierry Fontenelle. 2011. Lexicography. In J. Simp-
The peculiarities of each group of annotators               son, ed The Routledge Handbook of Applied Lin-
lead to a thought-provoking analysis of the tags-           guistics. Routledge, Taylor & Francis Group. Oxon.
                                                            pp. 53 – 66.
ets and the semantic notions themselves. The
analysis of the semantic behaviour of the loans            Jeoren Geertzen. 2012. Inter-Rater Agreement with
unveils deeper questions on the speaker’s strate-            multiple raters and variables. Retrieved November
gy: if it is easier to reuse a loan whose meaning            27, 2015, from https://nlp-ml.io/jg/software/ira/
already exists, why are there also loans with new
or diversified meaning? Furthermore it came to             Roberto Gusmani. 1983. Saggi sull’interferenza lin-
our attention that yes, many loans are linked to           guistica (vol.1 e vol.2). Le lettere.
the informatics/technical context (rippare, slog-
gare, etc.) but others can be considered as hypo-          Patrick Hanks. 2012. How people use words to make
nyms of already existing Italian verbs, whose               meanings: Semantic types meet valencies. In A.
                                                            Boulton & J. Thomas, eds. Input, Process and
meaning is rather general (cheattare for barare,            Product: Developments in Teaching and Language
whinare for lamentarsi, etc.). It is possible to            Corpora. Masaryk University Press.
already discern synonymic and antonymic rela-
tions between the loans themselves: craftare,
farmare 2, spawnare 1 or sbuggare-buggare.                 Patrick Hanks. 2008. Lexical Patterns: from Hornby
                                                             to Hunston and beyond. In E. Bernal & J. DeCe-
Finally, the most productive verb classes in our
                                                             saris, eds.Proceedings of Euralex 2008. Barcellona,
corpus are the verbs of change of state, the crea-           Universitat Pompeu Fabra.
tion verbs and the verbs of killing.
                                                           Patrick Hanks and Elisabetta Jezek. 2010. What lexi-
References                                                  cal sets tell us about conceptual categories. In Lexis:
                                                            E-journal in English lexicology, 4: Corpus Linguis-
Ron Artstein and Massimo Poesio. 2008. Inter-coder          tics and the Lexicon.
 agreement for computational linguistics. Computa-
 tional Linguistics, 34(4), 555-596.                       Patrick Hanks and James Pustejovsky. 2005. A Pat-
                                                            tern Dictionary for Natural Language Processing.
Joe Barcroft, Gretchen Sunderman and Norbert                Revue française de linguistique appliquée, 10(2).
  Schmitt. 2011. Lexis. In J. Simpson, ed. The
  Routledge Handbook of Applied Linguistics.               Martin Haspelmath. 2009. Lexical borrowing: con-
  Routledge Taylor & Francis Group. Oxon., pp. 597          cepts and issues. Loanwords in the world’s lan-
  – 610.                                                    guages: a comparative handbook.

Marco Baroni. Italian tagset documentation. Available      Martin Haspelmath. 2008. Loanword typology: steps
 at:http://sslmit.unibo.it/~baroni/collocazioni/itwac.ta    toward a systematic cross-linguistic study of lexical
 g set.txt.                                                 borrowability. Aspects of language contact: new
                                                            theoretical, methodological and empirical findings
Marco Baroni and Adam Kilgarriff. 2001. Large lin-          with special focus on Romancisation processes.
 guistically processed web corpora for multiple lan-        Mouton de Greuyter.
 guage.
                                                           Elisabetta Jezek. 2011. Lessico. Classi di parole,
Raffaella Bombi. 2003. Anglicismi come banco di             strutture, combinazioni. Il Mulino, Bologna.
 prova dell’interferenza linguistica. Italiano e ingle-
 se a confronto, Firenze, Franco Cesati Editore. 101-      Elisabetta Jezek. 2010. Struttura argomentale dei ver-
 125.                                                       bi. In L. Renzi – G. Salvi (a cura di) Grammatica
                                                            dell’Italiano Antico. Bologna: Il Mulino, 77-122.
Raffaella Bombi. 2005. La linguistica del contatto:
 tipologie di anglicismi nell’italiano contemporaneo       Elisabetta Jezek and Malvina Nissim. 2014. Linee
 e riflessi metalinguistici. (Vol. 11). Il calamo.          guida per l’annotazione degli argomenti del verbo

Tullio De Mauro. 1999–2007. GRADIT - Grande Di-            Elisabetta Jezek and Laure Vieu. 2013 . Lista di ruoli
zionario Italiano Dell’uso. 6 vols. Torino: UTET.           semantici per Senso Comune.

Joseph L. Fleiss. 1971 . Measuring nominal scale           Adam Kilgarrif. 2001. Web as corpus. Proceedings of
agreement among many raters. Psychological Bul-             corpus linguistics 2001. Corpus Linguistics. Read-
letin, 76:378-382.                                          ing in a Widening Discipline.
Alan Kirkness. 2004. Lexicography. In A. Davies &
 C. Elder, eds. The Handbook of Applied Linguis-
 tics. Blackwell Handbooks in Linguistics. Black-
 well Publishing, pp. 73 – 101.

Klaus Krippendorf. 1980. Content Analysis: An Intro-
 duction to its Methodology. Sage Publications

Alessandro Lenci, Simonetta Montemagni and Vito
  Pirrelli. 2012. Testo e computer. Elementi di lin-
  guistica computazionale. Carocci editore.
Beth Levin. 1993. English verb classes and alterna-
 tions: a preliminary investigation. University of
 Chicago Press.

Anthony McEnery, Richard Xiao and Yukio Tono.
 2006. Corpus based language studies. An advanced
 resource book. Routledge, Taylor & Francis Group.
 Oxon. Routledge Applied Linguistics.

Jan Pomikálek. 2011. Removing Boilerplate and Du-
  plicate Content from Web Corpora. Ph.D. thesis.
  Brno, Repubblica Ceca: Masaryk University. Avail-
  able at:
  http://is.muni.cz/th/45523/fi_d/phdthesis.pdf.

James Pustejovsky and Amber Stubbs. 2012. Natural
  language annotation for machine learning.
  O’Reilly.

Renata Savy. 2006. Specifiche per la trascrizione or-
 tografica annotata dei testi raccolti. In Progetto
 CLIPS. Corpora e Lessici dell’Italiano Parlato e
 Scritto.

Amber Stubbs. 2011. MAE and MAI: Lightweight
 Annotation and Adjudication Tools.

Michael Stubbs. 2004. Language Corpora. In A. Da-
 vies & C. Elder, eds. The Handbook of Applied Lin-
 guistics. Blackwell Handbooks in Linguistics.
 Blackwell Publishing, pp. 125 – 152.

Peter Wittenburg, Hennie Brugman, Albert Russel,
  Alex Klassmann and Han Sloetjes. 2006. ELAN: a
  Professional Framework for Multimodality Re-
  search. In: Proceedings of LREC 2006, Fifth Inter-
  national Conference on Language Resources and
  Evaluation.

</pre>