Voice archives in Arabic dialectology: the case of the southern Tunisian recordings in the Berliner Lautarchiv

ABSTRACT Some of the earliest voice recordings of Arabic made for linguistic purposes date from World War I and were made by German authorities who recorded prisoners of war in the Halbmondlager camp outside of Berlin. This study analyzes two voice records in particular, which are labelled ‘Tripolitanisch-Arabisch (Tunesien)’ and stem from what is now southern Tunisia. This study seeks to historicise the scientific context of these voice records as well as interpret the linguistic data preserved by them in the light of Arabic dialectology. It also raises questions about the reliability of both voice recordings and printed linguistic data from the time.


Introduction
On March 2, 1918, a young prisoner of war named Sliman from 'Gaṣ r Mednīn' in North Africa stood in front of a gramophone funnel in Halbmondlager, a prison camp on the outskirts of Berlin, to recite a short fairy tale. The sound of his voice was etched onto a shellac disc for future study by linguists, who, taking advantage of the numerous languages spoken by prisoners in the camp, had set up a large-scale, if impromptu, recording project. One week later, on March 9, Sliman returned to the gramophone, where he recited a list of words and numbers, likewise recorded onto a shellac disc. Sliman's fate after these sessions is unknown. Perhaps he died somewhere in Europe, or perhaps he was able eventually to return home to Gaṣ r Mednīn. But his voice has remained, imprinted within the crevices of a fragile shellac disc and stored in the cabinets of an archive in Berlin, for over a century.

The Lautarchiv recordings
Following the outbreak of World War 1, German linguist Wilhelm Doegen (1877Doegen ( -1967, had 'the idea of using the involuntary stay of the prisoners of war held in Germany for sound recordings of speech,' and to this end he petitioned the Prussian War Ministry, and then the Ministry of Culture and Education, to create a large-scale project (Doegen 1925, 9). The Royal Prussian Phonographic Commission was founded in Berlin in 1915 with this very goal, and Doegen at its head. Its task was to 'systematically record on sound discs the languages, the music, and the sounds of all the peoples residing in German prisoner-of-war camps according to methodological principles and in relation to accompanying texts' (Doegen 1925, 10). 1 As the title of Doegen's volume of studies on the recordings suggests, he envisioned pioneering a totally new kind of ethnology ('eine neue Völkerkunde'). By the end of the war only three years later, an impressive number of recordings had been made. On the music side, Carl Stumpf (1848Stumpf ( -1936 and his team recorded over 1,000 wax phonograph cylinders of music by 1918. In 1920, these were handed over to the Berlin Phonogramm-Archiv, which had itself been founded in 1900 to record the music of 'non-European' peoples (Ziegler 2006). On the spoken language side, a team of linguists under the direction of Doegen produced some 1,650 shellac discs of sound recordings between 1915 and 1918. These became the basis of the Lautabteilung of the Preußische Staatsbibliothek in Berlin and today are stored in the Lautarchiv of the Humboldt-Universität zu Berlin.
Located in the village of Wünsdorf outside of Berlin, the prisoner-of-war camp known as Halbmondlager ('Half Moon Camp') was chosen as a particularly promising site for the work of the Phonographic Commission, as it was populated by Asian and African conscripts to colonial armies who were in German captivity. The initial purpose of gathering these disparate peoples together, in the Halbmondlager as well as in the Weinberglager in nearby Zossen, was not simply to detain them as enemy troops: the captives were to be indoctrinated through German-conceived extremist Islamic instruction with the ultimate aim of persuading them, once returned home, to rebel against their colonial rulers, be they the English in southern Asia or French in northern Africa (Höpp 1997).
Since prisoners in the camps spoke a number of Asian and African languages, the Phonographic Commission naturally included an 'Orientalist', Hubert Grimme (1864Grimme ( -1942. Grimme, who served as translator for Arabicspeaking prisoners, was present, in addition to Doegen, for most of the recording sessions of North African prisoners, which yielded about 100 shellac discs with texts in 'Algerisch-Arabisch', 'Marokkanisch-Arabisch', 'Berbersprache', 'Maltesisch', and other language varieties. The comparative dialectological intention of some of the recordings is clear: a number of the discs recorded in March 1918 contain the same wordlists, primarily 'Zahlen' and 'Wörter', in different varieties. 2 In addition, the same speakers also provided a variety of folktales, songs, poems, and even Quran recitations. In comparison, the recordings made in 1916 largely seem to be freely chosen or spontaneously composed texts. For example, in a recording session in May 1916, Sadok Ben Rachid, a prisoner from Monastir, recited a poem of his own composition in which he recalled his experience fighting in the trenches of Europe. 3 This extremely early collection of Arabic and Berber recordings, possibly the first recordings of either language ever made for linguistic purposes, however, was hardly used. Grimme contributed a chapter, entitled 'Die Farbigen von Nordwestafrika', on the North African recordings to Doegen's 1925 ethnographic volume, focusing on what he called 'Schützengrabenpoesie' of Algerian and Tunisian prisoners (Grimme 1925). In 1928 the Arabist Hans Stumme (1864Stumme ( -1936, professor at Leipzig, transcribed and translated six recordings-five in Tunisian Arabic by Sadok Ben Rachid from Monastir and one in Shawi Berber by Mehiddin Ben Buzid from Algeria-for a short pamphlet published by the Lautabteilung der Preußischen Staatsbibliothek (Stumme 1928). While only very recently have scholars and practitioners in sound studies begun to devote attention to these materials, in particular those recorded by Sadok Ben Rachid, 4 the recordings have otherwise escaped the notice of linguists of Arabic or Berber.
The present article is therefore the first to use one of the Lautarchiv recordings from North Africa for a linguistic study. But as these are recordings made in a prisoner of war camp, proper contextualisation is imperative, as is careful consideration of the research purposes to which they are put. Though they may appear to the linguist to be merely raw data, the recordings are part of an archive which is formed by and dependent on imperial power relations. 5 Moreover, an awareness of just what it is that we are listening to is important. As Hoffmann has pointed out, the recorded voices of the Lautarchiv are characterised by being not only acousmatic, but also by the delocalisation of the speaker: their recited or sung material is separated, in the recorded object, from its social and epistemic contexts, thus complicating or obscuring the meaning of a recorded piece and the position of the speaker (Hoffmann 2015, 77-80). These recorded voices, as 'voice-only characters in the form of archived recordings that stem from projects of colonial knowledge production', are what Hoffmann refers to as colonial acoustic figures, or 'colonial acousmêtres' (ibid., 79). Britta Lange gives three examples of ways in which these recordings can potentially be used: for linguistic purposes, as 'documents for historicizing spoken language and for exploring how dialects were shaped and pronounced nearly one hundred years ago'; for history of science, in reconstructing how 'science' was used in the war and what categories and products of knowledge were relevant; and for cultural history as documents representing the perspective of the colonised serving as soldiers in European armies (Lange 2014, 372). The possible uses for a recording depend to some extent on the nature of that recording. The original intentions behind them mean that specific recordings (e.g. music performances, word-lists, etc.) were intended for specific areas of knowledge. Our study follows the groove of archival intentions, at least for some of the way. And while ours is a linguistic study, we do not foreclose the possible use of this and similar recordings for additional purposes, scientific, creative, or otherwise. Some prisoners, for example, composed texts which represent an important perspective on a part of history for which there is little documentation, especially not directly in the voice of colonised conscripts (Lange 2011;. Bringing out these subaltern voices as part of cultural history is of great importance.
But what about the records which do not contain compositions commenting directly on the war or on imprisonment, but rather contain the productions based on and intended for traditional dialectology-such as that of Sliman? We think such recordings can make a critical contribution to Arabic dialectology, when examined with care not only as voice recordings, but as 'sensitive documents' and cultural-historical artefacts. The two discs recorded by Sliman are the only two from this particular region of North Africa. They represent a unique case, since there are also no contemporary linguistic studies from this region either. As what Lange refers to as 'documents for historicizing spoken language', that is, as linguistic data, this case can help us reconsider older sources which make claims, sometimes even without written data, about the dialects of North Africa in general and Tunisia in particular. In addition, this case can provide diachronic data which, in comparison with current data, can shed valuable light on language change over time. While we can uphold these points about the linguistic value of these early voice recordings, we also must sound a note of ambivalence. Sliman and others were coerced into recording for a European project conceived with the expressed intention to use them as nonconsensual sources for 'modern' linguistic analysis. But most of the records were never used for such, and have lain in the archive for more than a century, only even heard a handful of times. Is it then better, to finally make use of the recordings for that original purpose, or to leave them mostly unheard for yet another century?

The 'Tripolitanisch-Arabisch (Tunesien)' discs
The two discs recorded by Sliman were labelled 'Tripolitanisch-Arabisch (Tunesien)' and numbered PK 1221 and PK 1228. The first disc, PK 1221, contains a 'Märchen' of 3'23" in length, and the second disc, PK 1228, contains 'Zahlen' and 'Wörter' and is recorded on both sides, totalling 3'08" in length. 6 According to the accompanying Personalbögen (personal information sheets, see Image 1) made at the time of recording and signed by Grimme and Doegen, PK 1221 was made at 12:30 on March 2, 1918 and PK 1228 exactly one week later, at 12:30 on March 9, 1918. The exact technique used to guide the speakers through the wordlists and texts is not always clear, but in some cases it is known that an assistant would whisper the words to the speaker, who would repeat them into the gramophone funnel. The Image 1. The Personalbogen (personal information sheet) for Sliman from Gas  r Mednīn, accompanying disc PK 1221 in the Lautarchiv der Humboldt-Universität zu Berlin. Published with permission of the Lautarchiv.
Personalbogen for PK 1228 even adds a note to the entry 'Art der Aufnahme' ('type of recording'): 'vorgeflüstert von dem Fachmann' ('whispered out by the specialist'). 7 On the other hand, PK 1221 containing the 'Märchen' does not have this comment, as the speaker simply seems to have recited or improvised the story in a relatively fluid manner; however, some accounts seem to indicate that the speakers practiced their texts beforehand so as to minimise errors or pauses (Scheer 2010, 306). In neither recording, however, can background sounds or the voices of other persons be heard. Doegen also added a comment about the quality of Sliman's voice, describing it as a 'kräftige dunkle Stimme mit geringer stimmloser Konsonanz'. The Personalbögen record that Sliman was born in 1897 in 'Gaṣ r Mednīn' (an older name for today's Medenine), to parents from the same place. The only other personal information provided is that Sliman had been living in Tunis at the age of 20, was an 'Arab' and 'Muslim', was illiterate but could speak a bit of French, and that he worked on the railroad. Geographically, the two Personalbögen give the location of 'Gaṣ r Mednīn' as on the Tunis-Tripolitanian border, but also add a comment that the general region is 'western Tripoli'. 8 Again, the colonial acousmêtre confronts us: only the recorded voice of Sliman exists, but Sliman as a speaker and his position as a speaker are unknown, and potentially unknowable. 9 In this section, both discs will be transcribed in their entirety and furnished with an English translation. The transcription of both discs has been divided into sections for ease of reference and translation; it should be noted that these do not represent any divisions in the audio recording itself. Likewise, the thematic labels for the transcription of the second disc have been added by us but are not present in the original audio.
He said to him: 'I will only give you money if you go where I went.' Thereafter he went far away on his horse and found a mountain full of silver. He filled up a lot of money from it. 4 tamm sāyir yalga wāh  id is-srāya fīha bint. gatlu: yā wildi, a  š ža  bik l-i hnā?
(The ġūl) said to him: 'Swords or shoulders?' He said to him: 'Shoulders'. They struggled with each other. He seised him, threw him to the ground and cut off his head. 11 gāl-lha: yā l-bint, wēnhum na  sik? gat-lu: na  si bla  dhәm bʕīda.
He said to her: 'Girl, where is your family?' She said to him: 'My family's country is far away.
[Make?] smoke on the multi-storey house, then my family will come.' Her family came and warmly welcomed him. They went to her father's tent and he married her. He said to him: 'My father-in-law, I want to return home.' 13 ʕt āh zōz xdam w-zōz wis fān w-zōz m-il-xēl.
He gave him two female servants, two black ones and two horses. Thereafter he was on the way for ten days. 14 gal-lu: yā ws īf, r āni mutt b-il-ʕt  aš. gal-lu: r ūh  th  awwis ʕa-l-ma.
He went off. When he went off he found water, he found a well. He said to him: 'O servant, can you climb into the well?' 16 gal-lu: yā sīdi, ana mā-nnažžim-š. gal-lu: He said to him: 'My master, I can't.' He said to him: 'I will climb down.' He climbed down, transported up the water and all the people drank. 17 gal-lu: yā ws īf, t  allaʕni! gal-lu: mā-nt  allʕik-š.
He said to him: 'O servant, pull me up!' He said to him: 'I am not pulling you up!' So he stayed in the well. He said: 'My God, can you give me my lot?' 18 hūwa rāgid f-il-līl is  -s  bāh  nād  ̣ yalga xubza gāl: il-h  amdu l-lla  h yā r abb il-ʕālamīn.
After he had been sleeping in the night, he got up in the morning and found a piece of bread and said: 'Thank God, Lord of the worlds.' 19 ža  tu wah  d il-far xa gāt-lu: kēf i nt allʕik a  š taʕt  īni ʕlī b-is-sūdān? (vii) A young girl (viii) came and said to him: 'If I pull you up what will you give to me, ʕAlī b-is-Sūdān?' 20 gal-lha: il th  ibbi xūd  i. gāt-lu: ġammid  ̣ ʕaynēk w-irkib fōg minni.
He said to her: 'Take whatever you want!' She said to him: 'Close your eyes and ride on my back!' 21 ġammid  ̣ ʕaynēh tamm xāriž xar r ža  tu. gāt-lu: He closed his eyes and finally came out (from the well). She brought him out. He said to her: ix 'What do you want me to give to you?' He said to her, she said to him: 'I don't want anything.' 22 gāl-lha: rāhi qudrit Al l āh illi rislitni ba  š i nt allʕik. She said to him: x 'It was God's omnipotence that sent me to bring you out.' 23 gāt-lu: āna r  āni marsūl ʕlēk w-žītik mi-bla  d bʕīda lamkín ha  d  i ws āya minni tāxud  ha mātfar  r  at  -š fīh a . <tone> She said to him: 'I was sent on your behalf and I came from a country far away. But this is an advice from me that you should take and not ignore.' <tone> Notes on the transcription and translation: (i) Note assimilation for wēn ṛ uḥ t.
(ii) A ġūl is a kind of monster or demon that forms part of the traditional folklore in many Arab societies (see Szombathy 2013).
(iii) The speaker starts to stay ʕAlī bin … and then corrects himself to Mḥ ammid Bit-Ta  žir. (iv) The word here cannot be heard clearly.
(v) Perhaps the speaker intended Classical farḥ an šadīdan here. (vi) For kullha, but the h is not audible. (vii) This is not the name the speaker uses in section 9; see (iii) above. (viii) faṛ xa usually designates a 'baby bird', 'a child out of wedlock' or 'a nasty child'. But as was confirmed by a Tunisian native speaker, it can mean 'a young girl' without any negative connotation in Central Tunisia; the same is true for western Libyan dialects as well. (ix) The speaker actually says 'She said to him', and as such uses the masculine forms to address the man. The context, however, definitely calls for the girl being addressed. (x) Here the speaker actually says 'he said to her'.

Disc two
Notes on the transcription and translation: (i) This word does not appear in the sources for South Tunisian dialects, but is documented in the glossary of Takrouna as 'terrain bas et humide en cuvette qui fournit un bon pâturage; endroit où se déverse le trop plein d'un oued' (Marçais & Guîga 1958-61, 3792). (ii) gamṛ a 'moon' seems to be out of place here and is already mentioned above within the word field 'nature'. (iii) In the recording tam i r 'dates' is mentioned, seemingly out of order, after the metals.

Linguistic notes
In the following, the most striking phonological, morphological and syntactic peculiarities of the text and the word lists are described, and the most important lexical characteristics are noted. In Table 2 and in the discussion (Section 5) these features are embedded into the Tunisian dialectal landscape and an attempt will be made to characterise and localise the speaker's dialect in view of the existing literature on Tunisian dialects.

Phonology
1. Vowels: The speaker seems to rely mainly on a system of three short vowels (a, i and u), the allophones of which depend on their consonantal environment. However, in some words, clearly ә is audible. As for the long vowels, he uses ā, a  , 10 ī, ū and the monophthongized vowels ō and ē.

2.
Monophthongization: aw is generally monophthongized as ō; ay as ē: zōz 'two'; fōg 'above'; bēt 'tent'; xēl 'horses'; kēf 'when; if (conjunction)'; but līl 'night' and ʕaynē-k 'your eyes'. 3. g: gal-lu 'he said to him'; gʕad 'he sat, he stayed'; gaṭ ṭ ūs 'cat'; il-gāyla 'noon'. The only occurrences of q in the text and the word lists are tqābid  ̣ hūw w-yā-h 'they struggled with each other'; qahwa 'coffee'; qudṛ a 'omnipotence' and qas i m 'lot', the two latter used in a religious context. 4. ž: žwa  d-u 'his horse'; žā 'he came'; mā-nnažžim-š 'I can't'. 5. Interdentals are retained: fid  ̣ d  ̣ a 'silver'; t  amma 'there is'; t  la  t  a 'three'; tāxud  -ha 'you (m.) take her/it'; ha  d  i 'this (f.)'. 6. There is no vowel change from a > i/u in the verbal and nominal pattern C 1 vC 2 C 3 v(C) as in the other Southern Bedouin dialects (see Table 2). On the contrary, we have forms that are common in the urban dialects as well as in the Central Tunisian dialects: laṭ x-u 'he seized him and threw him on the ground'; šarbu 'they drank' and farḥ u 'they welcomed; they were happy' (but rislit-ni 'she/it sent me'), forms which would be lutx-a or lut u x-a, širbu and firḥ u in the South Tunisian dialect of Douz. 11 An example of a noun is bagṛ a 'cow' which is bugṛ a in other South Tunisian dialects. 7. No Imāla of stressed word-final ā or a: l-i hnā 'here'; mā, ma 'water'; šta 'winter'; žā 'he came'; lga 'he found'; t  la  t  a mya 'three hundred'. This clearly distinguishes our speaker's dialect from both the urban dialects (ma  ) and the South Tunisian Bedouin-type dialects (mē, mī). 8. Generally, also the word-internal Imāla is less strong than in many Tunisian dialects, e.g. wiṣ fān 'black people (pejorative)', which would be wiṣ fa  n in Tunis and in Douz, or ayyām 'days'. 9. No traces of a short a in open unstressed syllables (as is for instance common in Douz): žbal 'mountain'; nsīb-i 'my father-in-law'; xrīf 'autumn'; i rgad 'he fell asleep'. 10. The distribution of short vowels in verbs does not follow the Classical Arabic pattern but is subject to consonantal influence. Examples are the perfect forms tqābid  ̣ hūw w-yāh 'they struggled with each other', ġammid  ̣ 'he closed (his eyes)' (also the imperative which has exactly the same form occurs in the text); and the imperfect form mā-tfaṛ ṛ aṭ -š 'you (sg.m.) don't ignore'. 11. An epenthetic vowel between two word final consonants is not the rule: mutt 'I died'; bint 'girl'; il-xubz 'bread'; but qas i m 'lot'; sam i š 'sun'.
There are no pausal forms in the text that could be dinstinguished from their respective contextual forms; and no vowel harmony caused by the suffix -ha ( fī-ha 'in it/her', and not fa  -ha as for instance in Douz).
There is no genitive marker in the text, and no passive verb forms.

Syntax
Indefinite specifying article: wāḥ id is-srāya 'a multi-storey house', waḥ d it-ta  žir 'a merchant', waḥ d il-faṛ xa 'a young girl'; wāḥ id il-bīr 'a well'. This is another remarkable feature of Sliman's speech, as it is so far not attested in Tunisian dialects. In most Tunisian varieties, indefinite nouns simply occur without the definite article; constructions like wāḥ id fṛ anṣ āwi imply a specific indefinite 'a certain French man', and the noun has no article. 13 Although texts of the TUNOCENT-corpus which could provide counterevidence have not been transcribed yet, the project collaborators from Central Tunisia confirmed that they were unaware of constructions like wāḥ id il-bīr. They are, however, found in Morocco and also in (the Western part of) Algeria. 14 Constructions comparable to wāḥ id + article + noun in Sliman's text are šwayya l-mā 'some water', and yā l-bint 'oh girl', both usually found without the definite article in Tunisian dialects.

Lexis
Among the characteristic lexical items used by Sliman are the following: t  amma 'there is'; l-i hnā 'here'; ya  sir 'very; a lot'; ṛ āḥ -yṛ ūḥ 'to go (off)'; yḥ awwis ʕa-l-mā 'to search for water'; sam i š 'sun'; naww 'rain'; bēd  ̣ a 'egg'; žwa  d 'horse'; faṛ xa 'young girl'. Whereas the existential marker t  amma is typically found in all of Tunisia, and the adverb ya  sir and the noun naww are commonly found in Tunisian Bedouin-type dialects, faṛ xa with the meaning 'young girl' seems to be restricted to Central Tunisian dialects. Some of the other lexical items seem to be more common in Algeria (see Section 5). Finally, there are no French loanwords in the text.

Dialectological analysis
The retention of interdentals and the use of ž described in Section 4.1 are almost pan-Tunisian phonological phenomena common in both sedentary and Bedouin-type dialects. The voiced realisation of q as g, gender distinction with verbs, the conjugation of III-weak verbs (žū as opposed to ža  w in sedentary dialects) and the lengthening of the short vowel of the perfect 3.p.sg.f. in combination with a vocalic suffix (ʕṭ āt-u 'she gave him'), even the adverb ya  sir 'very', clearly mark the dialect under investigation here as of the type traditionally called 'Bedouin' within the dialectal landscape of Tunisia. For discussion of isoglosses separating the two major groups of Bedouintype dialects known so far in Tunisia, the starting point is usually W. Marçais' 1950 article where he divides the Tunisian Bedouin dialects into Hilāli and Sulaymi subgroups. 15 Though this study has some problems, due to its scarce linguistic data and possibly misleading categorizations, to a certain degree, it may indicate what the dialectal landscape of Tunisia might have looked like approximately in the colonial period. 16 On the basis of what W. Marçais proposes, the Medenine region would be located among the so-called 'Sulaymi' dialects that cover all of Southern Tunisia. In W. Marçais' description, the 'Sulaymi' dialects are also spoken by the tribes along the eastern coastline, and in northern Tunisia between the Medjerda River and the Mediterranean Sea. He lists the tribes in these areas as follows: 'les Ouerghemma, les Marâzîg et les gens du Nefzâoua, les 'Akkâra, les Hamârna, les Benî Zîd, les oasis de la région de Gabès, les Mhâdhba, les 'Agârba, les Neffât, les Mthâlîth, les Souâsi, les Oulâd Sa'îd, les Hdîl, les Mog'od et les groupes humains de la Kroumirie' (Marçais 1950, 214). 17 Of these, the Medenine area would be occupied by the Ouerghemma confederation, as a map from the middle of the 19 th century shows, and to be even more precise, by the Touazine tribe. 18 If we assume that the information provided by Marçais is accurate, then the dialect of Medenine in the first half of the 20 th century would be a 'Sulaymi'-type dialect. However, one of the main issues with Marçais' classification of Tunisian dialects is that he provides very little data to demonstrate them, indeed sometimes no data at all, making it difficult on linguistic grounds to accept their validity. But before dealing with these issues in more detail, let us examine what is known about the dialects of southern Tunisia first.
Though there is no publication concerned with the dialect of Medenine itself, we have sufficient knowledge of the varieties spoken in the proximity of Medenine (shown in Map 1 and Table 1) in essentially all the cardinal directions.
These studies vary in the linguistic domains and amount of material presented; some present only transcribed texts while others provide linguistic analysis but mostly in the domains of phonology or morphology. Regarding the potential relationships between these different dialects, Behnstedt (1998, 53f.) mentions some Arabic-speaking groups on the island of Djerba who originate from Medenine or its environs, while P. Marçais and Viré characterise the dialect of a member of the Mahād  ba tribe in the region of Skhira and Mezzouna as close to the dialects spoken in Gabis, Gafsa, and by the Nefzaoua tribe (Marçais and Viré 1981, 369).
As shown in Table 2, all of these dialects are quite consistent in the use of certain features which can be considered as isoglosses for this group. Among them is the vowel change from a > i/u in the verbal and nominal pattern C 1 vC 2 C 3 v(C) (bugṛ a 'cow' vs. bagṛ a), the suffix of the 3p.sg.m. being -a (dāṛa 'his house' vs. dāṛ -u) and the imāla of final ā that is realised as ē or even ī (mē, mī 'water' vs. ma  , mā). As lexical isoglosses we include the verbs 'to go' and 'to search for'. il-H  āmma Marçais andFarès (1931-1933), Cantineau (1951) 3 Gabes Skik (1969), Singer (1980: 271f.) 4 M h ād  ba tribe near Skhira Marçais and Viré (1981) 5 Bil-Xēr tribe in Gafsa Ritt-Benmimoun (2013) 6 Djerba Behnstedt (1998;1999) 7 Zarzis Bouaicha (1993) 8 Tataouine Singer (1996) 9 Mar āzīg tribe in Douz Boris (1951;1958), Ritt-Benmimoun (2011;2014a), also Bin Murād (1999  As is shown in Table 2, the features attested for the dialects surrounding Medenine mostly do not agree with those used by Sliman. He uses forms that are widespread in central and northwestern Tunisia, but not in southern Tunisia around the region of Medenine. The phonological and morphological features used by Sliman seem to be more of the so-called Tunisian Hilāli dialects (H-dialects) that are spoken in Central Tunisia, at least within W. Marçais' categorisation. 19 Sliman uses these forms consistently throughout the recordings, and does not give the impression of pretending to speak another dialect than he actually does. The contemporary linguistic data collected during the TUNOCENT-project in the past few years provides useful comparanda. It shows only bagṛ a for the governorates of Kef, Siliana, Kasserine, and Sidi Bouzid, no final imala for the adverb l-i hnā, l-i hna 'here' and only -u as suffix for the 3.p.sg.m. in the verb rīt-u or šuft-u 'I saw him/it'. The future marker maš (miš) on the other hand, is frequent not only in southeastern Tunisia 20 but is also widespread in the governorates of Beja, Jendouba, Kef,Siliana,Kasserine,and Sidi Bouzid. 21 On the lexical level the verb yḥ awwis with the meaning 'to search for' is not attested in the larger Medenine area, where it means 'to go for a walk, to go on vacation'. However, we find the verb yḥ awwis with the meaning 'to look for' documented for Algeria (Behnstedt & Woidich 2014, Map 346) and it occurs also in our TUNOCENT data for the governorate of Gafsa. ṛ āḥ -yṛ ūḥ 'to go, to leave' is only rarely attested in the TUNOCENT data where primarily the imperative aṛ ṛ āḥ occurs, but perhaps it was more common in Sliman's time, as the entry ṛ āḥ -yrūḥ 'partir; s'en aller' in Marçais and Farès' glossary of the Ḥ āmma dialect shows (1931)(1932)(1933)230). Additionally, the use of bēd  ̣ a for 'egg' is very uncommon for Tunisia as a whole (and within the data for the TUNOCENT-project bēd  ̣ a for 'egg' is not documented at all), but has also been documented for Algeria (Behnstedt and Woidich 2012, Map 245).
The use of a  na and of yāxud  is unclear. a  na is usually considered as urban form (though we also find it in our data in many locations within the governorate of Gafsa), used among others in Tunis, whereas in Central Tunisian dialects forms like na  , na  y and na  ya are used. In the South Tunisian dialects the realisation of the verbs 'to take' and 'to eat' is yāxud  and yākul as opposed to yōxud  and yōkul in Central Tunisian dialects. Finally, as far as syntax is concerned, the use of an indefinite article (e.g. wāḥ id il-bīr 'a well'), attested only for Moroccan and Algerian dialects, leaves us with another big question mark.
After comparing with the existing data for central and southern Tunisian dialects, it seems that the dialect of Sliman points unexpectedly towards the regions of northern Gafsa, Sidi Bouzid, Kasserine, Kef, and Siliana, that is, towards Tunisia's central and northwestern dialects, and, especially on the lexical and syntactic level, even Algerian dialects. What to make of this conclusion?
If we assume that the linguistic situation according to W. Marçais was actually the case in the early 20 th century and that Sliman's dialect was representative of Medenine at the time, then his recordings contradict the expectation that Medenine would have a 'Sulaymi'-type dialect, or to put it more generally, that the dialect of Medenine would be roughly similar to that of its neighbours. But if W. Marçais was only making conjectures about certain regions, and he did not have complete data, then we have no comparative information about the dialect of Medenine in the early 20 th century, and we cannot say whether or not Sliman's dialect is representative of Medenine's dialect at the time. This latter case leads to more open possibilities, which cannot be proven: perhaps Medenine centre itself had a different dialect than the surrounding rural areas; or perhaps different social groups, such as tribes, had different dialects. Of course, another possibility is that Sliman's place of origin was not recorded correctly by the Halbmondlager authorities -but other studies have shown successfully that these places of origin, including some in Tunisia, seem to be quite accurate (Kropotkine 2021). And in any case, we do not have any information about Sliman's tribal affiliation.
There are additional possibilities worth exploring, though. Perhaps Sliman's original dialect changed for some reason, such as living or working in another region of Tunisia. Although we know next to nothing about his life before his internment in Germany, the Personalbögen give his profession as 'Eisenbahner' (railroad worker). This may be significant, as there was never a railroad in Medenine; railways were concentrated in northeastern Tunisia and the farthest south they extended in the 1910s was Gafsa, only reaching Gabes and Tozeur, and not any further south, after 1930 (Brant 1971, 97-136). As a railroad worker, Sliman is likely to have spent time all over Tunisia and thus may have adjusted his dialect to accommodate to more central or northern Tunisian Arabic features. But, Sliman was only about 21 at the time of the recordings; he may not have spent a long period of time in other regions than Medenine, and even though it seems that he stayed in Tunis for a year at least before his internment, his dialect features do not really point towards the dialect of the city of Tunis.
It could also be the case that Sliman adjusted his dialect during his conscripted service in the French army, or his stay of indeterminate length in the Halbmondlager, where he would have interacted with speakers of numerous Algerian and Tunisian dialects. Even though this scenario is possible, one might expect a kind of partial accommodation to these other dialects and a fair amount of variation, but the linguistic features in Sliman's speech are rather stable and consistent. Likewise, one might expect that when relating a traditional piece of oral literature, one would do so in one's original dialect, but we cannot necessarily make this assumption for every speaker. At the same time, we might also question the pronunciation or variety used by the specialist who prompted Sliman as he recited the wordlists, or the scholar who may have helped him write and practice his tale, if this latter situation did occur. About either of these possibilities we have no information. The data provided by Sliman may therefore show a kind of idiolect that has accommodated to various features of other dialects, rather than an 'authentic' Medenine dialect. Indeed, given the paucity of early dialectological sources and their at times questionable reliability, we may never be able to judge this matter adequately.

Conclusions
In sum, this voice recording from 1918-one of a group of the earliest voice recordings of Arabic dialects in existence-presents us with valuable data. But this data is far from straightforward. Even though we are able to show how Sliman's speech, at least as it was at one point in time, differs from later data on nearby dialects of southern Tunisia, we are still left with many open questions.
Despite the obvious advantages of sound recordings over written data, this record shows that they do not provide easy-to-use, reliable snapshots back in time. They cannot be considered without social and historical context. Indeed even with what can be found of relevant context, our analysis has become more, rather than less, complicated. Further work using materials from the Lautarchiv for linguistic purposes must therefore also historicise the recordings and the data they contain, rather than taking them at face value.
The difficulties in analyzing this voice recording also suggest that written sources from the time, where we are additional steps removed from firsthand linguistic data-lacking information about the speakers who were consulted and their social contexts, (and in some cases lacking even substantial transcribed material)-may have similar complications and problems. Likewise, we should continue to rely on those early written sources likewise only with appropriate historicization.
Notes 22. Singer (1996, 31) states that some of the speaker's inconsistencies are due to influence from Standard Arabic and the urban Tunisian varieties.