Transcription Conventions for the Eisenbeiss German Child Language Corpora Sonja Eisenbeiss University of Essex [email protected] Ingrid Sonnenstuhl Düsseldorfer Akademie Essex Research Reports in Linguistics Volume 60 Number 2 17 Jan, 2011 Dept. of Language and Linguistics, University of Essex, Wivenhoe Park, Colchester, Essex, UK, CO4 3SQ http://www.essex.ac.uk/linguistics/publications/errl/ Essex Research Reports in Linguistics present ongoing research activities of the members of the Department of Language and Linguistics. The main purpose of these reports is to provide a quick publication outlet. They have ‘pre-publication status’, and most will subsequently appear in revised form as research articles in professional journals or in edited books. Copyright remains with the author(s) of the reports. Comments are welcome: please communicate directly with the authors. If you have technical problems downloading a paper, or for further information about these reports, please contact the editor: Doug Arnold: [email protected]. Citation Information: Sonja Eisenbeiss and Ingrid Sonnenstuhl. ‘Transcription Conventions for the Eisenbeiss German Child Language Corpora’, Essex Research Reports in Linguistics, Vol. 60.2. Dept. of Language and Linguistics, University of Essex, Colchester, UK, Jan, 2011. http://www.essex.ac.uk/linguistics/publications/errl/errl60-2.pdf Transcription Conventions for the Eisenbeiss German Child Language Corpora Sonja Eisenbeiss1 and Ingrid Sonnenstuhl2 1 University of Essex, 2Düsseldorfer Akademie Abstract In this document, we describe the transcription conventions for the Eisenbeiss German child language corpora. 1 These conventions are based on the so-called CHAT transcription conventions of the Child Language Data Exchange System (MacWhinney 2000; CHILDES: http://childes.psy.cmu.edu/; CHAT: http://childes.psy.cmu.edu/manuals/chat.pdf). We have incorporated modifications and additions of CHAT for German that were suggested by Stephany and Bast (1999; http://childes.psy.cmu.edu/intro/stephany.pdf) and by Heike Behrens (2006; pc.; http://childes.psy.cmu.edu/manuals/07germanic.doc#_Ref131136188). 1. The Eisenbeiss Corpus This corpus, metadata for individual recordings and further information about the participants are available via the online archive of the Max Planck Institute for Psycholinguistics, Nijmegen: http://corpus1.mpi.nl/ds/imdi_browser/ MPI>Acquisition>L1 >Eisenbeiss. The corpus consists of two sub-corpora. The first sub-corpus is the Eisenbeiss Elicitation Corpus, which contains several hundred recordings of three to six year old German and Dutch children The data collection involved semi-structured elicitation games that encouraged children to produce simple noun phrases, noun phrases with one or more adjectives and noun phrases with possessor phrases in a broad range of syntactic contexts (subject, direct and indirect object, adverbial phrases, etc.). See Eisenbeiss 2010, this volume, Eisenbeiss et al. 2009, for some of the elicitation tasks and materials used in the collection of the Eisenbeiss elicitation corpus. The second subcorpus, the so-called L-Family corpus, involves more than 1000 recordings from a two-year observation of a monolingual German family with four children and two adults – the mother of the children and the father of the two younger children. All participants speak dialect-free standard High German; and during the recording period, the two 1 The collection of both corpora was funded by the Max Planck Society and took place within the Acquisition Group of the Max Planck Institute for Psycholinguistics, headed by Prof. Wolfgang Klein. The development of the transcription conventions and some initial transcriptions of these corpora were funded by the Research Promotion Fund and the Research Endowment Fund of the University of Essex. Essex Research Reports in Linguistics Vol. 60.2, Jan 2011 2 adults were involved in higher education and working part-time. An overview of the children is given in Tab.1: Tab.1. Children involved in the L-Family Corpus Child Gender Age Year of birth Day-care School Lenny (C1) Male 5;2-7;8 1993 1997-2000 from 2000 Leon (C2) Male 2;0-4;6 1996 from 1997 - Liam (C3) Male 0-2;5 1999 from 2000 - Luna (C4) female 0-0;4 2001 - - Two types of data have been obtained: (1) spontaneous speech of children, parents, and guests collected during meals and free play, and (2) semi-spontaneous speech from elicitation games targeted at various types of noun-phrases. The elicitation tasks include the tasks used in the Eisenbeiss Case Elicitation corpus (see Eisenbeiss, this volume). See Slobin et al. (in press) and Eisenbeiss et al (2009) for initial publications based on this corpus. 2. Tools Used for Transcription and Annotation Transcriptions can be created and time-linked to the video/audio-stream using the tools provided by CHILDES or the multi-media-annotator ELAN (Wittenburg et al. 2006; http://www.lat-mpi.eu/tools/elan/). We used the latter as it was well supported at the Max Planck Institute, is XML-based, easy to use, and offers a lot of flexibility and excellent functionality for dealing with multi-speaker and multi-media corpora, for instance automatic recognition of silent periods in the recording. Import and export options are available for a range of text formats and software packages. In particular, ELAN allows for import from and export to the CHAT-format of CHILDES. Thus, one can use the so-called CLAN-tools of CHILDES for complex searches and frequency analyses or for semi-automatic morphosyntactic annotation (MacWhinney 2000; http://childes.psy.cmu.edu/manuals/clan.pdf). 2. Transcription We do not transcribe phonetically, but orthographically. As explained in more detail below, some non-target-like forms are standardised to facilitate computerised searches, but any deviations from standard German that are related to grammatical marking are transcribed as they are. In the following, examples are presented in German as this guide is targeted at users who want to transcribe or use German data. Essex Research Reports in Linguistics Vol. 60.2, Jan 2011 3 2.1. Structure and Naming of ELAN (eaf-)files For each speaker the ELAN file contains • one line for the transcription (independent, parent of comment lines, default) • one line for comments (referring to transcription tier) Transcription tiers have the type TRANSCRIPTION (independent, default language English);and utterances are time-linked to the audio/video-stream. Comment tiers have the type COMMENT (referring, symbolic association, default language English) and refer to the respective transcription tier for the speaker. Transcription/annotation files involving these tiers are created and the selection of tiers and their properties are saved as template files; see Tab.2-5 for some examples. Tab.2: L-Family Template (L_Family_Core.etf) Tier Name Participant Transcription Comment first child, Lenny CH1 CH1_Com CH1_Cas second child, Leon CH2 CH2_Com CH2_Cas third child, Liam CH3 CH3_Com CH3_Cas fourth child, Luna CH4 CH4_Com CH4_Cas mother, Natalie MOT MOT_Com MOT_Cas father, Ole FAT FAT_Com FAT_Cas researcher, Sonja SON SON_Com SON_Cas greatgrandmother, (Ma)tilde TIL TIL_Com Coding TIL_Cas Tab.3: Elicitation Corpus Puzzle Task template (Elicitation_Puzzle.etf) Tier Name Participant Transcription Comment Coding Child CH1 CH1_Com CH1_Cas researcher RES RES_Com RES_Cas Essex Research Reports in Linguistics Vol. 60.2, Jan 2011 4 Tab.4: Elicitation Corpus Picture-Pairing Template (Elicitation_picture_pairing_two_children.etf) Tier Name Participant Transcription Comment Coding first child CH1 CH1_Com CH1_Cas second child CH2 CH2_Com CH2_Cas Researcher RES RES_Com RES_Cas Tab.5: Elicitation Corpus Picture-Pairing Template for Recordings with a Researcher (Elicitation_picture_pairing_child_researcher.etf) Tier Name Participant Transcription Comment Coding child CH1 CH1_Com CH1_Cas researcher RES RES_Com RES_Cas The templates can be used to create new eaf-files, reducing the work load and ensuring consistency. Additional tiers for other participants (for instance occasional visitors to the family) can be added and tiers that are not used can be deleted. To create an ELAN file, transcribers need to start ELAN, and click on FILE > NEW and SELECT MEDIA. Then they must select the video-file (.mpg) and the corresponding sound file (.wav). The SELECT TEMPLATE option allows the transcriber to add the required template. If all files required are in the same folder, this should work easily and the resulting eaf. file can be saved immediately. In order to avoid data loss, files are saved frequently and under different names. All eaf-files created with ELAN for the Eisenbeiss L-Family and the Elicitation Corpus have the following structure: <name of the wav/video-file>_<initials of the transcriber>_<two-digit running number of saved copy>.<eaf> For instance, the first saved copy of a transcription file created by Sonja Eisenbeiss on the basis of the video- and audiofiles ased2001Jul19b.mpg and ased2001Jul19b.wav, would be called ased2001Jul19b_SE_01.eaf. The 17 th Essex Research Reports in Linguistics Vol. 60.2, Jan 2011 copy would be called ased2001Jul19b_SE_17.eaf. When a 5 transcript has been checked by a second transcriber, the initials of this transcriber will be added, e.g. ased2001Jul19b_SE_17_IS.eaf.: 2.2. Special Characters and Capitals Only ASCII-characters are used for transcriptions. Any special characters of German are transcribed using ASCII-characters: ä = ae; ö = oe; ü = ue; ß = ss. Capital letters are only used for proper names, i.e. for names of people, animals, locations, etc. (e.g. Lenny). The beginning of sentences as well as nouns are not marked by capitals (e.g. ich habe Lenny in Koeln gesehen.). 2.3 The Spelling of Stems and Bases As we are not specifically interested in the development of phonology and articulatory abilities, stems are slightly standardised to aid later computer searches. In particular, if one or more sounds are altered without any effects on grammatical marking, the standard forms are used in the transcription. For instance, both dunnel and tonnel are transcribed as tunnel if the intended word stem can be identified from the context. (1) Production Transcription guggema, guggemal gucke mal soen schoen tumachen zumachen topf kopf (if the child refers to a head, not a pot) wer, mer wir (if the intended meaning is "we") NOTE: changes of the stem vowel can function as grammatical markers (e.g. umlaut and ablaut). If this is the case, non-target forms of the stem vowel are transcribed as they are produced. The deviations from the target are indicated by an asterisk in square brackets and the target forms are provided in square brackets, after a colon and a blank; see e.g.: (2) Production Target Transcription manner maenner manner [*] [: maenner] er lauft laeuft er lauft [*] [: laeuft] sie helft hilft sie helft [*] [: hilft] huende hunde huende [*] [: hunde] Essex Research Reports in Linguistics Vol. 60.2, Jan 2011 6 Missing syllables or individual sounds are added in round brackets – as long as grammatical marking is not affected: (3) Production Transcription n (eine)n nane (ba)nane ma ma ma(ch) ma(l) no ma no(ch) ma(l) NOTE: if the omission of syllables or individual sounds affects grammatical marking (e.g. ein instead of eine) and leads to non-target-like forms, no round brackets are used. Rather, the deviation from the target is indicated by an asterix in square brackets; and the target form is provided in square brackets, introduced by a colon; see below. 2.4 The Spelling of Grammatical Markers Grammatical markers, i.e. inflectional morphemes and derivational morphemes, are never standardised. Deviations from the target that are not acceptable in colloquial speech, are indicated by an asterisk in square brackets and the target forms are provided in square brackets, introduced by a colon (see Stephany and Bast 1999): (4) Production Transcription ich treffte ihn. ich treffte [*] [: traf] ihn. ich fahre [//] fahrden rad . ich fahre [//] fahrden [*] [: fuhr] rad . ich fallt. ich fallt [*] [: fiel] . da ist eine gelbe haus da ist eine [*] [: ein] gelbe [*] [: gelbes] haus Incomplete, but acceptable colloquial forms are not marked by an asterisk. If it is possible to add the missing material, this is done in round brackets. If not, the (nonreduced) standard form is added in square brackets after a colon and a blank (see Stephany and Bast 1999): Essex Research Reports in Linguistics Vol. 60.2, Jan 2011 7 (5) Production Transcription ich geh ich geh(e) die habn sich was geholt. die hab(e)n sich was geholt. ham sich was geholt . ham [: haben] sich was geholt . 2.5 Compound nouns and other Complex Nouns All components of morphological compounds are separated by '+': (6) bahn+hof butter+brot mecker+tante An underscore can be used to ensure that several nominal elements are analysed as one unit by search-programs, even though these elements form a phrasal combination, not a compound, e.g.: (7) Neuss_Norf Mickey_Mouse Doctor_Mueller Raupe_Nimmersatt null_komma_nix rucki_zucki NOTE: All words are written out: ‘Sankt’ (not ‘St.’), ‘Doktor’ (not ‘Dr.’) (see Stephany and Bast 1999). 2.6. Punctuation and Apostrophies The following punctuation characters can be used: , . ; ? ! The end of each utterance has to be marked by a full stop, a question mark or an exclamation mark (see Stephany and Bast 1999). Commas and semicolons can be used within an utterance. Clitics are separated from preceeding words by a blank and an apostrophy: (8) die schaffen 's schon Essex Research Reports in Linguistics Vol. 60.2, Jan 2011 8 2.7. Doubtful Material and Unintelligible Speech If it is not entirely clear what the speaker actually said, the word or group of words serving as a best guess may be marked by ‘[?]’ (s. Stephany and Bast 1999): (9) die kommt da rauf [?] When it is difficult to choose between two possible transcriptions, the alternative transcription may be enclosed in square brackets: (10) die kommt da rauf [=? raus] Unintelligible single words, parts of utterances or whole utterances are transcribed by ‘xxx’. 2.8. Scoped Symbols Symbols placed in square brackets (‘[ ]’, e.g. [*] or [?]) can refer to single words or to more material. Then, the entire sequence must be surrounded by pointed brackets (‘< >’; s. Stephany and Bast 1999): (11) <ich fahre> [/] ich fahre rad. ich muss <da rauf> [?] ich muss <da rauf> [=? das rauf] 2.9. Onomatopoeic Forms, Variants of Colloquial Forms, Interjections, etc. Onomatopoeic forms are marked by ‘@o’ (s. Stephany and Bast 1999): (22) kikeriki@o macht der. In written German, many colloquial forms, interjections, and swear words can be spelt in different ways. For the sake of consistency, we are using only one spelling version. Interjections such as auweia are spelt as one word as they will not be analysed further. (13) Variants Transcription Kuck mal, guck mal, gucke mal guck(e) mal Auweia, au weia, auweija auweia oweh, o weh oweh o gott, oh gott, ogott, ogott hurra, hurrah, hurra Essex Research Reports in Linguistics Vol. 60.2, Jan 2011 9 ba, bah, ba baeh, baechs baeh hoppla, hoppela, hoppala hoppla tschuess, tschuessi, tschuehuess tschuess bums, bumms, bums prost, proscht, prosit prost o la la, oh lah lah, olala olala aha, ahah aha ach ach och och pst, pscht pst boing boing okay, ok ok jo, ja, ja pfui pfui nee, noe, nee scheiss, scheisse scheiss(e) herrgottnochmal, herrgottnochemal herr gott noch einmal herrgottnoch(ein)mal For standardisation purposes, each transcriber records the current list of all variants and their transcriptions in the document variants_<year in 4 digits>_<month in two digits>_<day in two digits>_ <transcriber initials>.doc (e.g. variants_2007_10_03_SE.doc). 2.10. Interjections and Direct Speech Interjections are marked by ‘@i’, which is added to the interjection (e.g. hm@i, oh@i). The precise form of the interjections is not indicated: Varieties such as ähm, ähem, mmm are all transcribed as ‘hm@i’, while interjections expressing surprise, such as aha, ah, boah, po, ho, och are all rendered by ‘oh@i’. The form ‘hey’ is transcribed as ‘hey@i’ and is not considered an English word. In cases where it is easy to decide whether the interjection ‘hm@i’ has an interrogative, affirmative or negative function (question, agreement, refusal), it may be further specified (s. Stephany and Bast 1999): Essex Research Reports in Linguistics Vol. 60.2, Jan 2011 10 (12) Variants Transcription interrogative interjection hm@ii affirmative interjection (‘yes’) hm@ia negative interjection (‘no’) hm@in Direct speech is marked by a plus and double quotes at the beginning and double quotes at the end: (15) und dann hat Ben gesagt +" du bist doof" und ist rausgerannt 2.11. Omitted Words and Constituents Though the coding of word and constituent omissions is an extremely difficult and unreliable process, it is very helpful for analyses of the feedback that children receive for non-target-like utterances. Hence, we code omissions of words by placing a combination of the zero symbol with the missing word on the transcription tier for the respective speaker. If what is important is not the actual word omitted, but its grammatical category (part-of-speech), or if it is not possible to determine the missing word then a code for the category can follow the zero: • 0art: article omission This code is only used if the lack of an article leads to an unacceptable noun phrase (e.g. da ist haus). Note that a noun without an article is acceptable if there is another determiner (ich will dieses haus), a possessive pronoun (das ist mein haus) or possessive noun (das ist Susis haus), a plural context (da sind huehner) or when the noun is a mass noun (ich will wasser) or a name (da ist Susi). Thus, NO article omission is coded in these circumstances, even when an article would be acceptable in the respective context (e.g. da ist die Susi) . • 0aux: auxiliary omission This code is used if the sentence contains a past participle without an auxiliary and it is not clear which auxiliary is missing (e.g. Max ist/hat geschwommen). If the auxiliary form can be identified, the zero is combined with the respective missing auxiliary form. • 0subj: subject omission This code is used if the subject is missing in a sentence that would require a subject in the target language. I.e., this code is not used in imperatives, even though they lack overt Essex Research Reports in Linguistics Vol. 60.2, Jan 2011 11 subjects. Sometimes it cannot be determined where the missing subject would occur. This is not important, as long as 0subj is positioned somewhere in the sentence. • 0do: omission of direct object This code is only used if the sentence is clearly unacceptable without the direct object (e.g. ich bewache jetzt aber or ich hab angemalt). It is not used if the sentence is acceptable without a direct object (e.g. ich esse gerade). • 0io: omission of indirect object This code is only used if the sentence is ungrammatical without the indirect object. It is not used if the omission would be acceptable (e.g. ich geb jetzt mal during a card game). • 0prep: omission of preposition This code for a missing preposion is only used if it is clear that a preposition is missing, but it is not possible to determine which specific preposition is missing. When the preposition can be determined, zero is combined with this preposition. (14) Production Transcription da ist haus da ist 0art haus ich weggegangen ich 0bin weggegangen Max geschwommen Max 0aux geschwommen da singt jetzt da singt 0subj jetzt ich hab angemalt ich hab 0do angemalt ich geb die blume ich geb 0io die blume ich will mama ich will 0zur mama ich gehe schule ich gehe 0p 0art schule (NOTE: while one can assume that a preposition and an article are missing, one cannot determine the exact preposition and article as the target utterance might be in die or zur schule) 2.12. Pauses, Retracing, Interruptions and Completions Utterance-internal unfilled pauses are marked by ‘#’ if they last for at least 2 seconds; if they last for at least 8 seconds, this is indicated by ‘##’. Filled Pauses (fp) are transcribed by ‘eh@fp’. Two filled pauses in a sequence are marked by ‘eh@fp eh@fp’ (s. Stephany and Bast 1999): (16) das ist ## ein # eh@fp huhn Essex Research Reports in Linguistics Vol. 60.2, Jan 2011 12 Repetition without correction is marked by ‘[/]’; repetition with correction is indicated by ‘[//]’. If several words are repeated, they are placed in pointed brackets (‘< >’). Several repetition marks may be used in one and the same utterance: (17) ich fahre [/] fahre rad . <ich fahre> [/] ich fahre rad . <ich fahren> [//] ich eh@fp fahre [//] fahrden [*] [: fuhr] rad . Filled pauses occurring directly in front of the repeated word(s) are placed after the retracing symbol (s. Stephany and Bast 1999): (18) der [/] eh@fp der brief. Uninvited interruptions by other speakers are marked by ‘+/’ at the end of the annotation, followed by clause end punctuation. Self-interruptions are indicated by ‘+//’ (s. Stephany and Bast 1999): (19) MOT wie oft muss ich dir +/ ? CH1 (21) doch. MOT und jetzt nimmst du +//. MOT nee, so nicht! Indicating completions in the transcript is crucial for studies of interactions between the child and other speakers. Hence, self-completion is marked by ‘+,’ and completion by others is indicated by ‘++’ at the beginning of the annotation that contains the completion. The utterance that is interrupted (by the speakers themselves or by other speakers) is marked by ‘+...’ at the end of the annotation. (23) CH1: und dann war da so ein +... MOT: ein was? CH1: + haus. MOT : und dann war da so ein +... CH1: ++ haus. Essex Research Reports in Linguistics Vol. 60.2, Jan 2011 13 2.13. Explanations and Comments Brief explanations can be given on the transcription tier for the respective speaker, indicated by [= text]. The comment tiers can be used to provide information that is necessary for the interpretation of the utterances, whether this information can be retrieved from the video itself or not. For instance, references to movies, fictional characters, etc. are explained whenever possible (asking the participants in the recordings might be required for this). Context information must be reliable, whether it is based on information that was provided by the video or by consultations with participants in the recording. (24) CH1 ich will den gelben [=ball] hier. CH1_Com He is pointing to the red ball. There is no yellow ball in the room. Acknowledgements We would like to thank the University of Essex Research Promotion Fund and Research Endowment Fund for supporting our work on child language corpora, which was the basis for this paper. We would also like to thank the Max Planck Society, Wolfgang Klein, and the Technical Group at the Max Planck Institute for Psycholinguistics in Nijmegen for their support in the development of the Eisenbeiss corpora. References Behrens, H. (2006). The input-output relationship in first language acquisition. Language and Cognitive Processes, 21, 2-24. Clahsen, H., Eisenbeiss, S., and Penke, M. 1996. Lexical Learning in early syntactic development. In: Harald Clahsen (ed.), Generative perspectives on language acquisition. empirical findings, theoretical considerations and crosslinguistic comparisons. Amsterdam: John Benjamins, 129-159. Clahsen, H., Eisenbeiss, S., and Vainikka, A. 1994. The seeds of structure. A syntactic analysis of the acquisition of case marking. In: Hoekstra, T. and Schwartz, B.D. (eds.). Language acquisition studies in generative grammar. Amsterdam: John Benjamins, 85-118. Eisenbeiss, S. 1994. Kasus und Wortstellungsvariation im deutschen Mittelfeld. Theoretische Überlegungen und Untersuchungen zum Erstspracherwerb. In: Haftka, B. (ed.), Was Essex Research Reports in Linguistics Vol. 60.2, Jan 2011 14 determiniert Wortstellungsvariation? Studien zu einem Interaktionsfeld von Grammatik, Pragmatik und Sprachtypologie. Opladen: Westdeutscher Verlag, 277298. Eisenbeiss, S. 2003. Merkmalsgesteuerter Grammatikerwerb. Dissertation University of Düsseldorf. (downloadable from http://deposit.ddb.de/cgi- bin/dokserv?idn=97646330x&dok_var=d1&dok_ext=pdf&filename=97646330x.pdf) Eisenbeiss, S. 2010. Production methods in language acquisition research. In: Blom, E. and Unsworth, S. (eds.) Experimental methods. Amsterdam: John Benjamins, 11-34. Eisenbeiss, S., Bartke, S., and Clahsen, H. 2005/2006. Structural and lexical case in child German: evidence from language-impaired and typically-developing children. Language Acquisition 13:3-32. Eisenbeiss, S., Matsuo, A., Sonnenstuhl, I. 2009. Learning to encode possession. In: McGregor, W.B. (ed.): The expression of possession. Berlin: deGruyter, 143-211. MacWhinney, B. 2000) The CHILDES project: Tools for analyzing talk. Third Edition. Mahwah, NJ: Lawrence Erlbaum Associates. Slobin, D.I., Bowerman, M., Brown, P., Eisenbeiss, S., Narasimhan, P. (in press). Putting things in places: developmental consequences of linguistic typology. To appear in: Bohnemeyer, J. and E. Pederson, Event representation in language: Encoding events at the language-cognition interface. Cambridge: Cambridge University Press. Stephany, U. and Bast, C. 1999. Working With The Childes Tools: Transcription, Coding And Analysis. (downloadable from http://childes.psy.cmu.edu/intro/stephany.pdf) Wittenburg, P., Brugman, H., Russel, A., Klassmann, A. and Sloetjes, H. 2006. ELAN: a Professional Framework for Multimodality Research. In: Proceedings of LREC 2006, Fifth International Conference on Language Resources http://www.lat-mpi.eu/papers/papers-2006/elan-paper-final.pdf Essex Research Reports in Linguistics Vol. 60.2, Jan 2011 and Evaluation.