ArticlePDF Available

Figures

Content may be subject to copyright.
LLM Special Issue 2017 ISSN: 0023-1959
Journal of the Linguistic Society of Papua New Guinea
ISSN 0023-1959
Special Issue 2017
Péter Maitz & Craig A. Volker (eds.):
Language Contact in the German Colonies:
Papua New Guinea and beyond
LLM Special Issue 2017
65
DOCUMENTING
UNSERDEUTSCH (RABAUL CREOLE GERMAN):
A WORKSHOP REPORT
Angelika Götze & Siegwalt Lindenfelser & Salome Lipfert &
Katharina Neumeier & Werner König & Péter Maitz
University of Augsburg
peter.maitz@philhist.uni-augsburg.de
ABSTRACT
This paper provides insights into the ongoing international research project
Unserdeutsch (Rabaul Creole German): Documentation of a highly
endangered creole language in Papua New Guinea, based at the University
of Augsburg, Germany. It elaborates on the different stages of the project,
ranging from fieldwork to corpus development, thereby outlining the
methods and software background used for the intended purposes. In doing
so, we also give some approaches to solving specific problems, which have
arisen in the course of practical work until now.
1
KEYWORDS
Unserdeutsch, corpus development, creole, fieldwork, German-based,
language endangerment, language documentation, spoken language data,
transcription
1
Research underlying this article has been funded by the Deutsche Forschungs-
gemeinschaft/German Research Foundation (MA 6769/1-1). We are very grateful to Eva
Schenzinger, Susanne Klohn and Craig Volker for their assistance in the translation of the
article and we wish to thank Thomas Schmidt for his helpful advice. Of course, any
remaining shortcomings are entirely ours.
Documenting Unserdeutsch: A workshop report
66
1 INTRODUCTION
At least half of the world’s languages can be considered endangered, i.e.
facing the brink of language dormancy or extinction (cf. Thomason 2015: 2).
For linguists, the impending loss of language variety means a race against
time in saving valuable data. The realization of this risky situation gave rise
to the linguistic subfield of language documentation in the 1990s (cf. Austin
2014: 58), which is concerned with the methods, tools, and theoretical
underpinnings for compiling a representative and lasting multipurpose record
of a natural language or one of its varieties (Gippert & Himmelmann &
Mosel 2006: v).
Efforts in language documentation, above all, strengthen the empirical
foundations of those branches of linguistics and related disciplines which
heavily draw on data of little-known speech communities (e.g., linguistic
typology, cognitive anthropology, etc.) in that they significantly improve
accountability (verifiability) and economizing research resources” (Himmel-
mann 2006: 1). Whalen (2004) even sees an upcoming revolution looming
on the linguistic horizon, initiated by the study of endangered languages
based upon large amounts of language data recently being shared on a global
scale.
Language documentation may also support work in related disciplines,
such as oral history and anthropology. Moreover, it constitutes “a necessary
first step toward language maintenance and revitalization or as a safeguard
against complete language loss(cf. Austin & Grenoble 2007: 17). Doing
research not only on, but likewise for and with a speech community is the
basic idea (cf. Dwyer 2006: 32).
Unserdeutsch (Rabaul Creole German), the creole language this paper
deals with, can be classified as “severely” to critically endangered (cf.
Maitz & Volker 2017) according to the UNESCO Language Vitality Index
(cf. UNESCO 2003). Unserdeutsch is therefore an urgent case for language
documentation. This applies even more as Unserdeutsch has several
characteristics that make it stand out in comparison to other languages.
LLM Special Issue 2017
67
2 THE UNSERDEUTSCH LANGUAGE
Unserdeutsch is the only known German-based creole language in the world.
It emerged among mixed-race children at a missionary station of the Catholic
Sacred Heart Missionaries (MSC) in Vunapope at the beginning of the 20th
century. Vunapope, today part of Kokopo, is located in the northeast of the
island New Britain (Gazelle Peninsula) in the Bismarck Archipelago. For
generations, Unserdeutsch has been a historical connection between Papua
New Guinea and Germany, even with the vast majority of the speakers living
in Australia today. Tok Pisin, as the most important substrate language,
shaped especially the grammar and phonology of Unserdeutsch (for aspects
of the linguistic structure of Unserdeutsch refer to Lindenfelser & Maitz, this
issue).
Despite its linguistically exceptional position, the language has hardly
been recognized by linguists since its first documentation at the end of the
1970s. The only more or less broad description of at least the most important
basics of the language is an unpublished master’s thesis from the early 1980s
(cf. Volker 1982). More than 35 years later, Unserdeutsch is now at the edge
of its extinction: Hardly any fluent speaker is younger than 65 years, and the
language has not been passed on to the following generations for decades.
It is urgent to document the language and to do research now. The
ongoing decline of the number of speakers, and the increasing attrition show
that the data collection cannot have been delayed. Fortunately, The German
Research Foundation (Deutsche Forschungsgemeinschaft, DFG) the
central funding agency for scientific research projects in Germany – granted
a considerable third-party project in order to document Unserdeutsch (cf.
Maitz et al. 2016). Maitz & Volker (2017) show the status quo of the
Unserdeutsch research, and first new findings concerning language structure
and language endangerment. This workshop report reflects the project status
as of spring 2017.
3 PROJECT OVERVIEW
The goal of the research project under the leadership of Péter Maitz and
Werner König (both at the University of Augsburg) is the comprehensive
Documenting Unserdeutsch: A workshop report
68
documentation of Unserdeutsch. The funding period for the project covers
three years (October 2015 to September 2018). The close cooperation with
Craig A. Volker (The Cairns Institute of the James Cook University), the
Institute for the German Language (Institut für Deutsche Sprache, IDS), the
main non-university institution in Germany for studying and documenting
the German language, and the advisory support of Peter Mühlhäusler, a
leading expert in the field of Pacific and creole linguistics, contribute to the
project with additional expertise.
The documentation of Unserdeutsch can be divided into three main
stages. The first project stage, already far advanced, covers data collection
by carrying out fieldwork in the Pacific. This was followed by the current
second stage, which is the development of an annotated corpus of
Unserdeutsch. In the third and last stage, a systematic description of the
language based on the corpus will be carried out. For this description, an
application for the funding of a subsequent project will be made, as the
current project covers mainly the development of the corpus. Of course, the
three project stages cannot be hermetically separated from each other but
overlap one another. A flexible approach has seemed reasonable: While
developing the corpus (stage 2), one can keep collecting new data if new
speakers come to be known or if deficiencies, e.g., in terms of the balancing
of the corpus (such as a balanced consideration of different varieties in the
post-creole continuum) become visible. Furthermore, qualitatively poorer
recordings can be replaced by better ones. The language description (stage 3)
can also be started gradually during stage 2 (cf. Lindenfelser & Maitz, this
issue). The basic linguistic structures have already become apparent while
dealing with the data in the course of the corpus development. Even within
the current project stage, e.g., in the stage of the corpus development, it is
possible to start working on one of the following stages early. The advantage
here is that problems that arise can be recognized at an early stage.
Even now, in the first third of the funding period, some first positive
results and effects of the Unserdeutsch project can be stated:
(a) Through the project, Unserdeutsch has attracted public interest and
generated great interest beyond our expectations, not only within the
scientific community but also in the public. Numerous print and audio-
visual media in the German-speaking countries, Australia, Papua New
LLM Special Issue 2017
69
Guinea, and beyond have been reporting about the language, the speech
community, and the project since it was launched. This is a factor that is
of great importance in giving prestige to the project, and at the same time
a good basis for efforts to officially accept Unserdeutsch as an
endangered language and to take revitalising measures.
(b) The speakersattitudes towards their language has changed positively.
Until now, they have thought that Unserdeutsch was, in contrast to hohe
Deutsch (Standard German), a deficient L-variety. Therefore, at the
beginning of the project some persons did not want to speak
Unserdeutsch with a competent Standard German speaker being present
(linguistic shame). Through the interest in research and documentation of
their language, the speakers have become aware of their language, its
uniqueness, and the cultural heritage it transports. The desire has arisen
to preserve Unserdeutsch for their descendants and to maintain the
language in their community.
(c) Last but not least, the social relationships within the group have been re-
intensified through the project. The Unserdeutsch community was close-
knit until the post-war period, as all speakers were living in and around
the Vunapope Mission. With the emigration of most of the speakers as a
consequence of the independence of Papua New Guinea in 1975, social
contacts and with them community ties were more and more weakened
by geographical dispersion. Now, the group identity is being revived,
strengthened by its own closed Facebook group, where memories and
news can be shared and discussed.
4 DATA COLLECTION, FIELDWORK
Unserdeutsch is spoken by only about 100 elderly people today. There are
only a few fluent or semi-fluent speakers below the age of 65. Most of the
speakers families who lived in and around Vunapope have emigrated to
Australia as a result of the independence of Papua New Guinea, and live
today in the metropolitan areas along the east coast of Australia in and around
Brisbane, the Gold Coast, Cairns and Sydney. Only a small number remained
in Papua New Guinea, now scattered on different islands of the country. The
main distribution area of the language has thus shifted from Papua New
Documenting Unserdeutsch: A workshop report
70
Guinea to eastern Australia since 1975. For this reason, fieldwork is taking
place mainly in Australia. Only a small handful of speakers have been
interviewed in Papua New Guinea, among those living in Kokopo, Kavieng
and the Duke of York Islands.
A large part of the data could be collected from 2014 to 2017 during five
fieldwork trips. We started with the transcription of the first recordings made
during the first fieldwork trips in spring 2016. At this stage, there is only one
further fieldwork trip scheduled for autumn 2017. The data collection is done
by means of a method triangulation, with a partly controlled narrative
interview at its core. In order to reduce the relative unnaturalness of the
survey situation and to improve the authenticity of the data, two peer group
members (friends or relatives) are interviewed at the same time, so that a
conversation can develop during the interview. The interviewer’s questions
mainly relate to the linguistic and social circumstances at the mission in
Vunapope, to the past and the present life of the speakers, and to their families
and the speech community as a whole. By this thematic focus, valuable
metalinguistic information relevant for the linguistic interpretation of the
primary data can be gained. The interview is supplemented by a questionnaire
with about 320 stimuli (sentences and phrases) in English and Tok Pisin, the
principal languages of Papua New Guinea spoken by all Unserdeutsch
speakers. The language of the stimuli is determined or selected by the
interviewees depending on their preference and competence. These stimuli
are reproduced orally by the interviewees in Unserdeutsch and recorded. The
stimuli are designed in a way that the basic vocabulary and the most
important morphological and syntactic variables are elicited. At the end, the
interviewees fill out a questionnaire to provide metalinguistic data
concerning their language biography, their self-identity, and their language
attitudes. This information is essential for both the reconstruction of the
language history of the community, which is to be reconstructed
systematically as part of a doctoral dissertation project, as well as for the
appropriate linguistic interpretation of the primary linguistic data.
Approximately 50 hours of spontaneous speech and the according
transcripts are to be integrated in the corpus. Basilectal, mesolectal and
acrolectal varieties of Unserdeutsch will be represented in these 50 hours so
that, if possible, the whole variation along the post-creole continuum is
displayed. When collecting data, one is inevitably confronted with numerous
LLM Special Issue 2017
71
methodological problems. The first and probably biggest one is the
observer’s paradox (Labov 1972): We try to record natural, spontaneous
speech in an investigative situation, which is unnatural when produced only
by the presence of an outsider investigator. Unserdeutsch is, and always was,
the medium of informal oral in-group communication within the Vunapope
mixed-race community. Therefore, the language does not show any
considerable stylistic variation. However, the observer’s paradox is a severe
methodological challenge, especially regarding mesolectal and acrolectal
speakers, who control a wider range of the creole continuum, and, at the same
time, have a greater language awareness.
A second methodological difficulty arises from standard language
ideology being present within the speech community and especially among
acrolectal speakers:
(1) immä wi geht spiel-en mit alle kind fi die,
every_time 1PL go play-V with PL child-PL of 3PL
die sa: in mein haus du spreh-en deutsch odä
3PL say in 1SG.POSS haus 2SG speak-V German or
englisch, kein kaputt-e deutsch
English no broken-ATTR German
‘Every time we went playing with their children, they [their parents]
said: In my house, you speak German or English, no broken German.
This leads the speakers to classify Standard German as Proper German and
Unserdeutsch as its corrupted form. This becomes apparent inter alia in emic
language designations for Unserdeutsch, such as Falsche Deutsch (“wrong
German”) or Kaputtene Deutsch (“broken German”). This standard language
ideology leads to the problem that especially acrolectal and mesolectal
speakers, who already show a bigger language awareness, might consciously
avoid basilectal features in formal out-group communication, including the
interview situations.
These difficulties have been dealt with in four different ways during
fieldwork. Firstly, we try to establish a trustful relationship with the speakers
through informal personal meetings (e.g., gatherings, lunch, dinner etc.)
before the interviews. Secondly, most of the interviews take place at the
Documenting Unserdeutsch: A workshop report
72
speakershomes, i.e. in a familiar environment. Thirdly, the interviews are,
as already mentioned, conducted with two familiar peer-group members. All
these measures are supposed to reduce the unnaturalness of the recording
situation. Last but not least, the interviewers speak (as much as they can)
Unserdeutsch during the interviews. This is a necessary strategy as Standard
German is only partly intelligible for the speakers of Unserdeutsch.
Moreover, the use of the in-group language as a contextualization cue should
contribute to dissolve the group boundary between interviewer and
interviewee (Gumperz 1982). In addition, the avoidance of Standard German
as an interview language is supposed to prevent echo effects, and, hence, the
evocation of unauthentic acrolectal features.
At the beginning of the fieldwork conducted for the Augsburg
Unserdeutsch Project in autumn 2014, Unserdeutsch was hardly documented.
A large part of the recordings made by Volker at the end of the 1970s and the
beginning of the 1980s had been lost; one copy is missing due to the
carelessness of the university library in Australia where it had been deposited,
another copy was destroyed by the great volcanic eruption in Rabaul in 1994.
Only a small rest containing about six hours of bad or very bad quality has
survived.
2
Almost half of this material, however, is actually Standard
German, because the older generation interviewed at that time was also able
to speak Standard German fluently thanks to school lessons partly held in
German in Vunapope during the interwar period.
Since the beginning of the project, 62 hours of recordings with 52
speakers in total have been made from 2014 to 2017; 47 hours of these are
narrative interviews and about 15 hours are questionnaires. The latter will not
be integrated into the corpus but will be essential when describing the
language structure after the language documentation.
The data show considerable variation in numerous respects. The first
concern the languages used. Unserdeutsch, Tok Pisin and English play an
important role in the language biography of all speakers. This trilingual
competence shows varying symmetry or asymmetry depending on the
speakers’ individual language biographies.
2
The remaining data is now stored at the Archive of Spoken German (Archiv für
Gesprochenes Deutsch, AGD) in Mannheim; cf. the corpus German in Oceania (Deutsch
in Ozeanien, OZ): http://agd.ids-mannheim.de/korpus_index.shtml (13.11.2017).
LLM Special Issue 2017
73
This is why speakers shift or switch from Unserdeutsch to English or Tok
Pisin to a different extent and often vary between two or three of these
languages during the interviews:
(2) orait i wid ni resign i wid bleib
TP UD UD UD EN UD UD UD
all_right 1SG AUX.FUT NEG resign 1SG AUX.FUT stay
arbeit weiter […] aba i own-im de trade store
UD UD UD UD EN-TP UD EN EN
work further but 1SG own-TR ART.DEF trade store
‘All right, I will not resign; I will stay and work further on […], but I
(will) own the trade store.’
Secondly, the speakers use lexically and grammatically differently elaborated
basilectal, mesolectal or acrolectal varieties of Unserdeutsch. Thirdly, the
recordings differ in terms of the extent and kind of attrition phenomena:
lexical gaps, problems with word finding, code-switching, and phonological,
grammatical and/or lexical interference of English and/or Tok Pisin
languages that have been used as functional first languages for decades.
5 DEVELOPMENT OF THE UNSERDEUTSCH CORPUS
The development of the Unserdeutsch corpus can be divided into five phases:
1. Transcription, 2. Normalization, 3. Lemmatization, 4. Annotation, and 5.
Database implementation. The most time-consuming part is the present phase
of transcription.
Phase 1 – Transcription
The transcription of the audio recordings is conducted with the Partitur Editor
of the software EXMARaLDA (Extensible Markup Language for Discourse
Analysis). EXMARaLDA (cf. www.exmaralda.org) is a free tool for
computer-assisted speech transcription and for the administration of spoken
language corpora (cf. Schmidt & Wörner 2014). It is compatible with all
common operating systems and offers interfaces to other common
transcription tools and formats (e.g., ELAN, Praat, TEI, Transcriber) as well
Documenting Unserdeutsch: A workshop report
74
as to standard applications such as Microsoft Word, internet browsers and
text editors via import and export functions. The transcription appears time-
aligned to the audio file in musical score (Partitur) notation:
Figure 1. The work surface of the EXMARaLDA Partitur Editor
One transcription tier is assigned to every speaker, further description tiers
and annotation tiers can be added optionally. The possibility to play intervals
in a modified speed has been helpful. Work is done with two different files
that are linked to one another: the recording itself and the transcription file.
EXMARaLDA has dedicated support for a number of widely used
transcription systems such as GAT 2 (Gesprächsanalytisches Transkriptions-
system 2, cf. Selting et al. 2009) or HIAT (Halbinterpretative Arbeits-
transkriptionen, cf. Ehlich & Rehbein 1976) that are commonly used in
German-speaking countries and beyond. For the purpose of the Unserdeutsch
project, the system cGAT (cf. Schmidt et al. 2015) was taken as a basis, which
was developed in the style of GAT 2 minimum transcript. It presumes the use
LLM Special Issue 2017
75
of standard orthography when a deviation is not significant from the standard
phonology. Expressions that show significant deviation are noted in a so-
called “literary transcription”, i.e. phonetically using the Latin alphabet (a
modified standard orthography). It is based on grapheme-phoneme
correspondences; the literary transcription is therefore easily applicable and
readable. It makes sure that structurally relevant features of Unserdeutsch –
but not every small phonetic alternation that are irrelevant for grammatical
description purposes – are visible in the transcript and searchable. This
concerns phonetic phenomena such as the g-spirantization (Standard German
Berg ‘mountaintranscribed as <berch> according to the actual pronunciation
influenced by northern German varieties), the delabialization of labial vowels
(Standard German Frühstück ‘breakfasttranscribed as <frihstick>) or the
loss of final consonants (Standard German Nachmittag ‘afternoon
transcribed as <namitta>). Sequences or words in English and Tok Pisin are
reproduced according to the standard orthography of the respective language,
since these languages are not in the focus of the Unserdeutsch project. For
the notation of Unserdeutsch, some additional conventions were established
in the interest of a more detailed possibility for analysis. In planning the time
needed for transcription, a ratio of 1:60 has turned out to be realistic. For one
minute of recordings, 60 minutes of transcription are therefore required
(including correction steps). This is an average value, which varies depending
on the comprehensibility of the recordings, the number of the people
speaking, and the languages used. When choosing recordings for the initial
phase of the transcriptions, we consciously started with a range of different
speakers who represent various varieties along the creole continuum. By this
means, we can start at an early stage to do first language analyses on a semi-
representative basis and to describe the variation along the creole continuum
of Unserdeutsch. Furthermore, it has become apparent that it makes sense to
specialize the transcribers in terms of the recordings: The necessary period
of acclimatization with a new recording decreases when individual speakers
or the speech variety (basilectal vs. acrolectal speakers) are familiar to the
person transcribing. The same is true for the distribution of the recordings
with extensive code-switching to English or especially to Tok Pisin.
Therefore, not everyone transcribing has to become acquainted with Tok
Pisin equally well. By this means, every transcriber becomes an expert for a
special part of the transcription and loses less time when dealing with
Documenting Unserdeutsch: A workshop report
76
problems. As soon as the planned amount of data is completely transcribed
and reviewed, shorter relevant passages will be additionally transcribed
phonetically using the IPA. This is crucial for a detailed phonological
description of Unserdeutsch. EXMARaLDA supports phonetic transcription,
but we intend to use the software Praat, which was created especially for that
purpose. Praat offers detailed measuring, e.g., in terms of vowel quantity
which is especially relevant for the analysis of Unserdeutsch, as the data
show different vowel lengths, but the phonological distinctiveness of the
vowel quantity seems very doubtful at first sight, at least in the basilect (cf.
Maitz & Volker forthc.).
Phase 2 – Normalization
The use of literary transcription means a high error ratio for automatic part-
of-speech tagging (cf. Westpfahl & Schmidt 2013: 140), because there are no
lexicon entries for spoken forms. It thus becomes necessary to normalize the
transcribed forms before doing further steps of annotation. That is, every
transcribed form is assigned to its equivalent in standard orthography which,
at a later stage, offers expanded possibilities for searching the corpus. In order
to speed up this manually time-consuming process, the tool OrthoNormal
was developed at the Institute for German Language (IDS) in Mannheim (cf.
Schmidt 2012: 239–240).
Figure 2. The work surface of OrthoNormal in normalization mode
LLM Special Issue 2017
77
The normalization with this tool is semi-automatic, as it suggests forms by
itself, and identical forms can be assigned to their standard form in only one
step in a whole transcript. Moreover, the program proposes equivalents on
the basis of an integrated lexicon, which keeps growing by the assignment of
further forms. By this means, fewer and fewer forms have to be typed in
manually. The word suggestions are sorted by frequency of use as well.
However, until now only a Standard German lexicon has been implemented:
The lexicon for the substrate language Tok Pisin and the adstrat language
English would have to be built as part of the normalization of the
Unserdeutsch data in the first place. In this way a foundation for the use of
the tool in English-speaking areas and in the Pacific could be laid. But since
we only used standard spellings for Tok Pisin and English in the
transcriptions, we can get around this task: The forms no longer need to be
normalized, since there are no deviating spelling forms in the transcripts.
Phase 3 – Lemmatization
The normalized forms are the basis for an error-free automatic
lemmatization. This happens by the already established tool TreeTagger. An
interface for the TreeTagger functions has been integrated into OrthoNormal,
so the lemmatization can easily be done there, too. The assignment of tokens
to their lemmas provides further possibilities for an effective search of the
corpus at a later stage.
Phase 4 – Annotation
In order to guarantee the searchability of the corpus by morphosyntactic
categories, independent from individual word forms, we do a part-of-speech
tagging of the data. This also happens automatized with TreeTagger with its
interface in OrthoNormal. The Stuttgart-Tübingen-Tagset will be applied for
this purpose, which is already an established standard for the German
language (STTS, cf. Schiller et al. 1999). To be more precise, we use the
expanded STTS 2.0, which has been designed for the annotation of spoken
language in particular, optimizing the tagset for that purpose by introducing
slight modifications and by adding further categories of spoken language (cf.
Westpfahl 2014; Westpfahl et al. 2017).
Furthermore, a manual annotation of transfer phenomena on the lexical
level will be performed, which can also be done with the tool OrthoNormal.
Documenting Unserdeutsch: A workshop report
78
By this means, the occurrence of Tok Pisin and English lexemes will be
marked. This will later facilitate research, for instance on the integration of
lexical material from English or Tok Pisin into Unserdeutsch, e.g., when
creating a dictionary.
Transcription
i
hat
ein
sch (.) schtore
Meaning
‘I‘
‘have‘
‘a‘
store
Normalization
[ich]
[habe]
[einen]
[Store]
Lemmatization
[ich]
[haben]
[ein]
[Store]
POS-Tagging
PPER
VVFIN
ART
NN
Figure 3. Normalization, lemmatization and POS-Tagging of transcribed word forms
Phase 5 – Database implementation
After completion and a period in which access will be restricted to selected
direct collaborators of the project, the Unserdeutsch corpus will be made
available for international research and teaching (i.e. non-commercial use)
via the Database for Spoken German (URL: http://dgd.ids-mannheim.de).
The DGD as a corpus management system digitally provides a part of the
Archive of Spoken German (Archiv für Gesprochenes Deutsch, AGD) that is
part of the IDS in Mannheim. The access is web-based and free of charge
after a one-time registration. The database provides the transcripts aligned
with the recordings so that they can be read and listened to in the browser.
Furthermore, there is additional material and extensive metadata. The core of
the database is an elaborated search function, which allows complex search
queries within the metadata, the transcripts, and their annotation layers (cf.
Schmidt 2014: 1453–1454). Here, different variables and wildcards can be
used. The structure-sensitive token search provides the possibility to search
on different annotation levels, including transcribed, normalized and
lemmatized forms as well as POS tags. The context-sensitive search displays
co-occurrences in a KWIC-view. An individual working area provides the
possibility to compile one’s own virtual corpora. They can even be
downloaded on request for offline work.
LLM Special Issue 2017
79
6 PRACTICAL ISSUES 1: ETHICO-JURISTIC ASPECTS
Particular project specific issues have arisen in this practical fieldwork-based
data collection. One of them refers to the ethical and legal dimension of the
project in terms of data protection (for general ethical aspects in language
documentation refer to Dwyer 2006). The collection, storage, processing and
publication of personal data are subject to the consent of the concerned
people according to German law as well as to the regulation of data protection
of the European Union guaranteeing the right of informational self-
determination. This obviously includes the building of a corpus (cf. DFG
2013). A mutual agreement must therefore be established between
researchers and informants. This contains data protection statements on the
researchers’ part and the written consent of the interviewed speakers, as was
done in the Unserdeutsch project.
Normally, this must be guaranteed by anonymization and
pseudonymization so that the identities of the speakers are, if at all, only
possible with tremendous effort. Unserdeutsch is however a special case: The
speakers care for the documentation of their biographic stories and their
language, especially for their own descendants. Therefore, there is no need
for them to conceal their identities. The reference persons explicitly agreed
to have their names and pictures published in connection with the
Unserdeutsch data on the homepage of the IDS and on the university project
website. Therefore, the obligation for the anonymization and pseudo-
nymization of transcripts and metadata, as well as the necessity to fade out
or to blend the recordings at passages that are sensitive for data security is
not applicable. Nonetheless, during transcription all occurring names are
assigned to a separate tier used solely for names. This gives the possibility to
fade out this tier or to blend it at a later stage, thus granting complete
anonymization. This could be necessary if the declaration of consent in this
respect was withdrawn, or if the data were to be passed on for (non-
commercial) purposes that do not concern the project the latter in
consultation with the speakers of course.
Apart from such data that allow the identification of the informants
(personal names, precise names of location), the handling of other personal
data has to be considered carefully, as well, even more so, as, because of the
arrangements just described, the full identification of the speakers is
Documenting Unserdeutsch: A workshop report
80
achievable without any effort. According to the German laws, “information
on racial or ethnic origin, political opinions, religious or philosophical
convictions, […] health, sex life” (Federal Data Protection Act, Section 35,
2) are included. First, as a matter of principle, information about the racial
or ethnic origin” of the Unserdeutsch speakers is crucial for research
questions on the social context of Unserdeutsch, its functions and its
emergence. Apart from that, the speakers talk openly about their lives during
the interviews. Thus, further sensitive information occurs in the recordings,
not only about the speakers themselves but also about persons related to
them.
As the loss of essential context information has to be avoided, this ethical
and legal issue has to be dealt with before the corpus goes online. If the
obliteration of sensitive passages is inevitable, this has to be conducted
manually. In case of doubt, the participants must be asked or their
descendants, in case of speakers having already passed away. Besides the
speakers themselves and their personal environment, sensitive content might
involve other parties as well, such as the MSC mission and its former staff.
Some interviews may even contain politically delicate information, for
example, when speakers talk about war crimes during the Japanese
occupation of New Britain in the course of World War II.
7 PRACTICAL ISSUES 2: TRANSCRIPTION PROBLEMS
Besides the common challenges emerging when transcribing spoken
language, some project-specific complications have arisen. Here four of these
shall be described.
(1) Dealing with homography
Some of the speakers switch between Unserdeutsch, English, and Tok Pisin
to a considerable extent, the latter depending on the speaker and generally
occurring more sporadically. As the system of GAT 2 minimum transcript
does not include the use of capital letters, capitalization is not an option to
differentiate between German and non-German substantives. As a result,
German-English homographs, such as <mission> for English [ˈmɪʃ(ə)n] and
likewise Unserdeutsch [miˈsion], or <plan> for English [plæn] and likewise
LLM Special Issue 2017
81
Unserdeutsch [plan], do appear. Names are especially relevant here, as
<angela> for English [ˈeɪndʒəla] could also be pronounced like in
Unserdeutsch [ˈɡɛla]. One exceptional rule for two highly frequent
homographs was set up from the beginning: the differentiation between the
English first person singular nominative pronoun I (capital letter) and the
equivalent pronoun in Unserdeutsch i (small letter, from Standard German
ich ‘I’). For a detailed study of code-switching phenomena, however, it
would be inevitable to make additional use of the recording for the
disambiguation of words like those mentioned above. A painless solution,
unless one wants to give up the morphologically-oriented standard
orthography, is the use of a comment tier for the notation of the English- or
German-oriented pronunciation of homographs. The problem rarely appears
with Tok Pisin because of its strict phonologically oriented orthography,
which shows very few homographic equivalents to English or German words.
Additionally, hybrid morphological structures can occur, such as in the
combination of English verbal stems with German or Tok Pisin suffixes. For
example, in a sentence like er sackim i (‘he sacked me’) it is not apparent
from the transcript alone whether the stem of the verb sackim (English to
sack + Tok Pisin suffix {-im}) is pronounced like in English (with [ɛ]) or
perhaps like in German (with [a]). Again, a remark in the comment tier can
solve the problem. Alternatively, a more phoneme-oriented transcription (eye
dialect) could have been used in case of an English pronunciation (e.g., in the
mentioned case: <säcken>). The issue becomes even more delicate with
speakers sporadically pronouncing English words in a pseudo-German style:
For example, words like Japanese or Asian are sometimes pronounced as
[japanis] and [asian] etc. In such cases, a remark in the comment tier is
inevitable.
(2) Dealing with homophony
In particular cases, the decision whether an expression derives from English
or Tok Pisin is not easily made. This concerns quasi-homophone word pairs
such as English New Guinea versus Tok Pisin Niugini or English boy versus
Tok Pisin boi. The problem also occurs in cases where it is not recognizable
if a word-final consonant occurs, such as in English New Ireland versus Tok
Pisin Niu Ailan. For various other homophones a decision can be made easily
from the context: in the phrase [lɪtl
̩ˈmaʊntɛn], for example, the prenominal
Documenting Unserdeutsch: A workshop report
82
adjective little (Tok Pisin: liklik) clearly indicates that the noun represents the
English word mountain instead of the quasi-homophone Tok Pisin equivalent
maunten. In case of proper nouns, however, the case is not always clear, as
the donor language could be English as well as Tok Pisin (e.g., New Guinea
vs. Niugini). The whole issue is further exacerbated by the existence of a
“Tok Pisin-to-English continuum”, resulting from decreolization processes
currently ongoing in Tok Pisin (cf. Devette-Chee 2011). Regarding verb
forms like leasim, ownim, rentim, servim, takim, teachim etc. which are used
by some speakers, it is sometimes difficult to decide upon their linguistic
status: Should they be dealt with as hybrid constructions (English loan + Tok
Pisin transitive marker {-im}), favouring spellings like those chosen above,
or as words already fully integrated into (urban-acrolectal, anglicized) Tok
Pisin? The latter would suggest spellings like tichim (acrolectal form besides
tisim) and so on. The decisions taken at this stage will influence the results
of subsequent analyses anyhow.
(3) Dealing with fluctuating realizations
The delabialization of the German umlaut vowels [y] [i] and [ø] [ɛ],
being typical for Unserdeutsch, are often realized in a gradually different
manner even by the same speaker. Therefore, an articulatory continuum
emerges, where a binary decision is not always clear to make. The same issue
occurs in basilectal hypercorrect depalatalization [ʃ] [s] (e.g.,
Unserdeutsch partly [s]warze for German [ʃ]warze ‘black people’, [s]wer for
German [ʃ]wer ‘heavy’, [s]tation for German [ʃ]tation), where various
articulatory graduations can be found. In the end, a decision for either the one
or the other sound is unavoidable for transcribers due to their being restricted
to using the alphabet. Only the phonetic transcription of specific passages,
which is still to be done, will take account of a finer mapping. Even more
difficult seems the decision whether or not a simplification such as the
deaffrication [ts] → [s], which is typical for Unserdeutsch, occurs in specific
phonetic environments. In cases of an alveolar nasal or, even worse, an
alveolar plosive preceding the sound in question it is almost impossible to
reach a decision on a purely auditive basis: [ganʦ] (‘very’) with German-
oriented spelling <ganz> or, assuming [gans], spelled <gans>? Similarly,
[gɛʦu] (‘go to’) may be spelled <geht zu> according to German
orthography – or otherwise <geht su>, assuming deaffrication in the onset of
LLM Special Issue 2017
83
the preposition. In such cases, it is only possible to decide by looking at the
pattern occurring in other phonetic environments. When in doubt, the
standard orthography is to be preferred (here with <z>). The same applies to
the potential loss of final consonants, when the consonant in question co-
occurs in the onset of the following word or syllable. This is especially so in
cases of cliticization: [unˈdan] (‘and then’) may be taken as <und dann> or
<un dann>, the latter presupposing a coda cluster simplification.
(4) Dealing with local proper nouns
Various local proper nouns, which are part of the speakers personal history
but are hard to identify for an outsider, are mentioned in the interviews. These
include names of rarely recorded places (e.g., little plantations on the island
New Britain), former local employers, or vernacular names for certain
animals (e.g., particular fish species) or plants. Names for local food are not
always easy to identify either. Most of these difficulties are related to the
countless personal names that are brought up in the interviews. In this respect,
the Facebook group created for the community as part of the project turned
out to be especially useful. Some of the elder speakers did create a Facebook
profile solely for the purpose of joining this group, so now they can easily be
consulted by the project staff any time. In other instances, sons and daughters
of the speakers can be asked for help, who are mostly available on Facebook
anyway. This convenient opportunity prevents time-consuming searches and
serves to clarify remaining terms not yet identified.
8 PRACTICAL ISSUES 3: ASSIGNMENT PROBLEMS
The further processing of the transcribed language data, i.e. the
normalization, lemmatization and annotation of the data, has prompted some
questions. These can be summarized in one main point: How far should we
be oriented towards Standard German grammar? In particular, if we decide
to move away from it, which is highly recommended from at least a
typological point of view, how can we do these (semi-)automatized
processing steps without too much manual correction effort and how can we
make sure that researchers who are not familiar with the structure of
Unserdeutsch and with our conventions will find what they are looking for?
Documenting Unserdeutsch: A workshop report
84
(1) Dealing with normalization issues
Our guiding principle for the normalization process with a few justified
exceptions is that we only do phonological and no morphological
normalization. That means we trace back all Unserdeutsch words to their
Standard German equivalents, thereby reversing substitutions and deletions
of Standard German sounds, but we do not reconstitute Standard German
inflectional paradigms, which have been eliminated in Unserdeutsch. Verbs
for example, which are generally uninflected in Unserdeutsch, thus do not
receive inflected normalized forms.
3
Conversely, of course, we preserve all
inflected forms, which do occur in the data (most often used by meso- and
acrolectal speakers or with high-frequency lexemes). Another aspect is the
capitalization of nouns in Standard German. Since the POS tagger, which is
trained on German word material, relies on capitalization in assigning word
classes, we adopt the capitalized forms recommended by the program.
Toponyms (like Vunapope, Rabaul) often have to be corrected by hand, as
with Unserdeutsch morphosyntax, the tagger does not recognize them as
proper nouns. Another source of error are all German-English homographs
in the transcripts (like mission, station): The algorithm assumes that these are
German words and thus capitalizes them regardless of the cotext (or even the
pronunciation). Most errors, however, occur with the automatic
normalization of short word forms (with only two or three letters), be they in
English, Tok Pisin or Unserdeutsch. The algorithm does not know any of
these languages; it only has the lexical entries for different varieties of spoken
German. Many of these short words now happen to coincide with reduced
regional word forms of spoken German, and consequently, the tagger
recognizes” them and assigns the wrong normalized form (e.g., UD ma ‘do
→ *mal [correct: mach], UD nu ‘only *nun [correct: nur], UD son
‘already’*so ein [correct: schon] – or EN no *noch; EN we *wenn;
EN is *ist). The only possible solution here is the manual correction of all
errors. In the mid-term, such errors could be avoided or reduced by overriding
existing entries in the normalization lexicon with entries derived from the
Unserdeutsch data.
3
However, we assign the SG lemma to all UD verb forms during the process of lemmati-
zation for reasons of searchability (this concerns verbs, whose uninflected default form
deviates from the SG infinitive).
LLM Special Issue 2017
85
(2) Dealing with annotation issues
Basically, there are four main points that have to be dealt with in the process
of the POS tagging. The first issue is the assignment of tags to foreign-
language lexemes (mostly English, to a lesser extent Tok Pisin and some
single words in other languages such as Japanese or Kuanua). The STTS
provides the tag FM (foreign-language material) to avoid the problem of word
classification in foreign languages on the basis of a tagset, which has been
developed for another language (and, above all, it would be unnecessary
work to tag the foreign material, since the researcher is interested in
Unserdeutsch). What we do is to extend the FM-tag with a language-specific
abbreviation (based on the ISO 639 codes), resulting in FM-EN (English), FM-
TP (Tok Pisin) etc. This way later on it will become easy to exclude foreign-
language material from searches or to, for example, look for the proportion
of Tok Pisin lexemes in Unserdeutsch.
4
The second issue that comes along with the POS tagging is that we have
to modify the STTS tagset for the annotation of Unserdeutsch in some
respects: a) Introduction of new categories: This concerns the word class
“plural word”, resulting in a tag PL, which is assigned to the plural word alle
all in prenominal position. But since alle may also bear the original
Standard German meaning allin the same position (and then functions as
an indefinite pronoun), it seems difficult to create a rule for the tagger. Thus,
this has to be done by hand. (b) Elimination of existing categories: Some
grammatical distinctions provided by the tagger are not reasonably applicable
to Unserdeutsch. One example is the distinction between finite forms (VVFIN)
and infinite forms (VVINF) of verbs, since there is – at least among basilectal
speakers no paradigm of verbal inflection in Unserdeutsch. Therefore, we
will eliminate this distinction everywhere, where it is not relevant. (c)
Divergent interpretations: in some cases, the tagger recognizes forms and
constructions correctly (at least in light of a grammar of spoken German),
but, based on the Unserdeutsch grammar, we have to choose another way of
4
The closed classes of function words and discourse elements from English and Tok Pisin
that are fully integrated into Unserdeutsch, i.e. occur frequently across speakers, are, how-
ever, tagged according to their grammatical function, as far as they are embedded in Un-
serdeutsch cotext. This particularly refers to pragmatic elements from Tok Pisin (orait,
maski, nogat and the question tag a) as well as junctions from English (e.g., whether,
cause). In this way we can assume that future subsequent grammatical analyses will be
facilitated.
Documenting Unserdeutsch: A workshop report
86
analyzing them. An example is the tagging of am in progressive or habitual
constructions, which in Standard German is a contraction (portmanteau) of
the preposition an and the inflected form dem of the definite article. In
Unserdeutsch, that analysis is in no way transparent and am has a much more
specific function (i.e. marking aspectual use of verbs), so we cannot use the
STTS tag APPRART (preposition + article) for it. Another example would be
the pronoun die, which functions as the 3PL personal pronoun in
Unserdeutsch (SG: sie), but which the tagger can only recognize as a
demonstrative pronoun, its function in spoken German.
The third issue in the process of annotating Unserdeutsch is the lack of
some features that the tagger requires to assign word classes correctly, based
on its rule knowledge from spoken German in Europe. This mainly results
from the following conditions: (a) the broad absence of inflectional elements
in Unserdeutsch, and (b) special word order rules in Unserdeutsch, especially
the absence of a formal distinction between sentence types in Unserdeutsch
(e.g., resulting in the problem of recognizing some subjunctives correctly).
Finally, the fourth issue is, obviously, the occurrence of words that are
unknown to the tagger because of their language-specific form, such as the
gender-indefinite article de or the polyfunctional lexeme fi. In each of the
cases mentioned above, there are three possibilities: (a) manual correction,
(b) the creation of a new rule for the tagger, or (c) the training of the tagger
by means of a manually annotated Unserdeutsch test corpus to enable it to
decide based on likelihood factors. Option (a) of course shows the lowest
susceptibility to errors, but is, on the other hand, far more time-consuming.
Option (b) only seems reasonable if a clear rule can be formulated without
too many exceptions. Option (c) again is a question of cost-benefit ratios, as
it takes time to create the test corpus, and afterwards, a manual correction
process still seems to be inevitable. For each case, this decision has to be
taken individually.
9 OUTLOOK
Upon completion of the Unserdeutsch corpus, it will provide the basis for the
linguistic description of the last remaining undocumented (partly) Germanic
language. Therefore, a funding for a follow-up project will need to be applied
LLM Special Issue 2017
87
for at the German Research Foundation (DFG) after having completed the
corpus development. A detailed reference grammar for Unserdeutsch should
be written as well within this future phase. The unexpected high interest in
Germany and beyond – including Australia and Papua New Guinea – obliges
us to present results soon; not to mention the relevance of Unserdeutsch for
international research far beyond German linguistics. Unserdeutsch provides
a rich source for research in language contact and language change, for
evolutionary linguistics and sociolinguistics, as well as for creole studies,
linguistic typology and other linguistic subdisciplines. The chance to get a
profound insight into the history and structure of the only German-based
creole language was made possible first and foremost thanks to the
cooperativeness and commitment of the last remaining Unserdeutsch
speakers.
ABBREVIATIONS
1PL first person plural
1SG first person singular
2SG second person singular
3PL third person plural
APPRART preposition + article
ART article
ATTR attributive
AUX auxiliary
DEF definite
EN English
FM foreign-lang. material
FUT future
REFERENCES
Austin, Peter. 2014. Language documentation in the 21st century.
JournaLIPP 3. 57–71.
Documenting Unserdeutsch: A workshop report
88
Austin, Peter & Lenore Grenoble. 2007. Current trends in language
documentation. In Peter Austin (ed.), Language Documentation and
Description, Vol. 4, 12–25. London: SOAS.
Devette-Chee, Kilala. 2011. Decreolization of Tok Pisin: Is there a Tok Pisin-
to-English continuum? Language and Linguistics in Melanesia 29. 95–
103.
DFG. 2013. Informationen zu rechtlichen Aspekten bei der Handhabung von
Sprachkorpora. URL: <http://bit.ly/1PG4Gq6>; Date accessed:
14.11.2017.
Dwyer, Arienne. 2006. Ethics and practicalities of cooperative fieldwork and
analysis. In Jost Gippert, Nikolaus Himmelmann & Ulrike Mosel (eds.),
Essentials of Language Documentation, 31–66. Berlin: de Gruyter.
Ehlich, Konrad & Jochen Rehbein. 1976. Halbinterpretative
Arbeitstranskriptionen (HIAT). Linguistische Berichte 45. 21–41.
Gippert, Jost, Nikolaus Himmelmann & Ulrike Mosel (eds.). 2006.
Essentials of Language Documentation. Berlin: de Gruyter.
Gumperz, John. 1982. Discourse Strategies. Cambridge: Cambridge
University Press.
Himmelmann, Nikolaus. 2006. Language documentation: What is it and what
is it good for? In Jost Gippert, Nikolaus Himmelmann & Ulrike Mosel
(eds.), Essentials of Language Documentation,1–30. Berlin: de Gruyter.
Labov, William 1972. Sociolinguistic Patterns. Philadelphia: University of
Pennsylvania Press.
Maitz, Péter & Werner König, & Craig Alan Volker. 2016. Unserdeutsch
(Rabaul Creole German). Dokumentation einer stark gefährdeten
Kreolsprache in Papua-Neuguinea. Zeitschrift für Germanistische
Linguistik 44. 93–96.
Maitz, Péter & Craig Alan Volker. 2017. Documenting Unserdeutsch:
Reversing colonial amnesia. Journal of Pidgin and Creole Languages
32. 365397.
Maitz, Péter & Craig Alan Volker (forthc.). Unserdeutsch (Rabaul Creole
German). In Hans Boas, Ana Deumert, Mark L. Louden & ter Maitz
(eds.), Varieties of German Worldwide. Oxford: Oxford University
Press.
LLM Special Issue 2017
89
Schiller, Anne, Simone Teufel & Christine Stöckert. 1999. Guidelines für das
Tagging deutscher Textcorpora mit STTS (kleines und großes Tagset).
URL: <http://www.sfs.uni-tuebingen.de/resources/stts-1999.pdf>;
Date accessed: 14.11.2017.
Schmidt, Thomas. 2012. EXMARaLDA and the FOLK tools – two toolsets
for transcribing and annotating spoken language. In Nocoletta Calzolari
et al. (eds.), Proceedings of the Eighth Eonference on International
Language Resources and Evaluation (LREC’12), 236–240. Istanbul:
European Language Resources Association (ELRA).
Schmidt, Thomas. 2014. The Database for Spoken German – DGD2. In
Nocoletta Calzolari et al. (eds.), Proceedings of the Ninth Conference
on International Language Resources and Evaluation (LREC’14),
1451–1457. Reykjavik: European Language Resources Association
(ELRA).
Schmidt, Thomas & Kai Wörner. 2014. EXMARaLDA. In Jacques Durand,
Ulrike Gut, & Gjert Kristofferson (eds.), Oxford Handbook of Corpus
Phonology, 402–419. Oxford: Oxford University Press.
Schmidt, Thomas, Wilfried Schütte & Jenny Winterscheid. 2015. cGat.
Konventionen für das computergestützte Transkribieren in Anlehnung
an das Gesprächsanalytische Transkriptionssystem 2 (GAT 2). AGD,
IDS Mannheim. URL: <http://bit.ly/2gJbapP>; Date accessed:
14.11.2017.
Selting, Margret et al. 2009. Gesprächsanalytisches Transkriptionssystem 2
(GAT 2). Gesprächsforschung. Online-Zeitschrift zur verbalen
Interaktion 10. 353–402.
Thomason, Sarah. 2015. Endangered Languages: An Introduction.
Cambridge: Cambridge University Press.
UNESCO. 2003. Language Vitality and Endangerment. Document submitted
to the International Expert Meeting on UNESCO Programme
Safeguarding of Endangered Languages. Paris, 10.–12. März 2003.
URL: <http://bit.ly/1GDzwra>; Date accessed: 14.11.2017.
Volker, Craig Alan. 1982. An Introduction to Rabaul Creole German
(Unserdeutsch). MLitSt thesis, University of Queensland.
Documenting Unserdeutsch: A workshop report
90
Whalen, Douglas. 2004. How the study of endangered languages will
revolutionize linguistics. In Piet van Sterkenburg (ed.), Linguistics
Today: Facing a Greater Challenge, 321–342. Amsterdam &
Philadelphia: Benjamins.
Westpfahl, Swantje & Thomas Schmidt. 2013. POS für(s) FOLK: Part of
Speech Tagging des Forschungs- und Lehrkorpus Gesprochenes
Deutsch. In Heike Zinsmeister, Ulrich Heid & Kathrin Beck, (eds.), Das
Stuttgart-Tübingen Wortarten-Tagset: Standard und Perspektiven =
Journal for Language Technology and Computational Linguistics
28(1), 139–153. Regensburg: Gesellschaft für Sprachtechnologie und
Computerlinguistik (GSCL).
Westpfahl, Swantje. 2014. STTS 2.0? Improving the tagset for the part-of-
speech-tagging of German spoken data. In Lori Levin & Manfred Stede
(eds.), Proceedings of LAW VIII – The 8th Linguistic Annotation
Workshop, 1–10. Dublin, Ireland: Association for Computational
Linguistics and Dublin City University.
Westpfahl, Swantje, Thomas Schmidt, Jasmin Jonietz & Anton Borlinghaus.
2017. STTS 2.0. Guidelines für die Annotation von POS-Tags für
Transkripte gesprochener Sprache in Anlehnung an das Stuttgart
bingen Tagset (STTS). Version 1, 1. März 2017. URL: <https://ids-
pub.bsz-bw.de/frontdoor/index/index/docId/6063>; Date accessed:
14.11.2017.
Article
This paper outlines an innovative corpus‐based method for establishing word class using as a case study the verbal status of the locative copula in Cameroon Pidgin English. The corpus‐based methodology outlined here uses a combination of distributional, sociolinguistic, and frequency analysis to establish that the Cameroon Pidgin English locative copula, deiy, is a verb and in doing so outlines a blueprint for establishing word class in under described languages. Pidgin creole languages are often under described and identifying the word class of multifunctional words can be particularly challenging. The methodology presented here can resolve such ambiguity by basing the classification on the language in which the word occurs rather than on typological generalisations. This corpus‐based methodology has wide ranging applicability and could be especially useful in establishing word class in lesser described world English varieties and pidgin creole languages.
Article
Full-text available
Unserdeutsch, also known as Rabaul Creole German, is the only known German-lexifier creole. This critically endangered language has its origins in an orphanage in German New Guinea for mixed-race children, where Standard German was taught by mission personnel. Unserdeutsch was creolised in one generation, and became the in-group language of a small mixed-race community. It is now spoken by around 100 elderly speakers, nearly all immigrants to Australia. The current project is only the second documentation based on actual fieldwork and has a specific focus on the use and vitality of the language as used by the last generation of speakers. It has the aim of producing an Unserdeutsch corpus that will facilitate both future linguistic research and contact with the language for the descendants of Unserdeutsch speakers. Preliminary findings show variation among speakers along a continuum from heavily creolised basilect to an almost European German acrolect. Most of the lexicon is derived from German, while a number of basilectal grammatical constructions are the result of the loss of marked features in German and possible imperfect second language learning as well as relexification of Tok Pisin, the presumed substrate language.
Conference Paper
Full-text available
This paper presents two toolsets for transcribing and annotating spoken language: the EXMARaLDA system, developed at the University of Hamburg, and the FOLK tools, developed at the Institute for the German Language in Mannheim. Both systems are targeted at users interested in the analysis of spontaneous, multi-party discourse. Their main user community is situated in conversation analysis, pragmatics, sociolinguistics and related fields. The paper gives an overview of the individual tools of the two systems – the Partitur-Editor, a tool for multi-level annotation of audio or video recordings, the Corpus Manager, a tool for creating and administering corpus metadata, EXAKT, a query and analysis tool for spoken language corpora, FOLKER, a transcription editor optimized for speed and efficiency of transcription, and OrthoNormal, a tool for orthographical normalization of transcription data. It concludes with some thoughts about the integration of these tools into the larger tool landscape.
Conference Paper
Full-text available
The Database for Spoken German (Datenbank für Gesprochenes Deutsch, DGD2, http://dgd.ids-mannheim.de) is the central platform for publishing and disseminating spoken language corpora from the Archive of Spoken German (Archiv für Gesprochenes Deutsch, AGD, http://agd.ids-mannheim.de) at the Institute for the German Language in Mannheim. The corpora contained in the DGD2 come from a variety of sources, some of them in-house projects, some of them external projects. Most of the corpora were originally intended either for research into the (dialectal) variation of German or for studies in conversation analysis and related fields. The AGD has taken over the task of permanently archiving these resources and making them available for reuse to the research community. To date, the DGD2 offers access to 19 different corpora, totalling around 9000 speech events, 2500 hours of audio recordings or 8 million transcribed words. This paper gives an overview of the data made available via the DGD2, of the technical basis for its implementation, and of the most important functionalities it offers. The paper concludes with information about the users of the database and future plans for its development.
Conference Paper
Full-text available
Part-of-speech tagging (POS-tagging) of spoken data requires different means of annotation than POS-tagging of written and edited texts. In order to capture the features of German spoken language, a distinct tagset is needed to respond to the kinds of elements which only occur in speech. In order to create such a coherent tagset the most prominent phenomena of spoken language need to be analyzed, especially with respect to how they differ from written language. First evaluations have shown that the most prominent cause (over 50%) of errors in the existing automatized POS-tagging of transcripts of spoken German with the Stuttgart Tubingen Tagset (STTS) and the treetagger was the inaccurate interpretation of speech particles. One reason for this is that this class of words is virtually absent from the current STTS. This paper proposes a recategorization of the STTS in the field of speech particles based on distributional factors rather than semantics. The ultimate aim is to create a comprehensive reference corpus of spoken German data for the global research community. It is imperative that all phenomena are reliably recorded in future part-of-speech tag labels.
Book
Language documentation is a rapidly emerging new field in linguistics which is concerned with the methods, tools and theoretical underpinnings for compiling a representative and lasting multipurpose record of a natural language. This volume presents in-depth introductions to major aspects of language documentation, including overviews on fieldwork ethics and data processing, guidelines for the basic annotation of digitally-stored multimedia corpora and a discussion on how to build and maintain a language archive. It combines theoretical and practical considerations and makes specific suggestions for the most common problems encountered in language documentation. Key features: textbook introduction to Language Documentation considers all common problems.