| { | |
| "paper_id": "2021", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T14:32:01.215850Z" | |
| }, | |
| "title": "NorDial: A Preliminary Corpus of Written Norwegian Dialect Use", | |
| "authors": [ | |
| { | |
| "first": "Jeremy", | |
| "middle": [], | |
| "last": "Barnes", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "jeremycb@uio.no" | |
| }, | |
| { | |
| "first": "Petter", | |
| "middle": [], | |
| "last": "Maehlum", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "petterma@uio.no" | |
| }, | |
| { | |
| "first": "Samia", | |
| "middle": [], | |
| "last": "Touileb", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "samiat@uio.no" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "Norway has a large amount of dialectal variation, as well as a general tolerance to its use in the public sphere. There are, however, few available resources to study this variation and its change over time and in more informal areas, e.g. on social media. In this paper, we propose a first step to creating a corpus of dialectal variation of written Norwegian. We collect a small corpus of tweets and manually annotate them as Bokm\u00e5l, Nynorsk, any dialect, or a mix. We further perform preliminary experiments with state-of-the-art models, as well as an analysis of the data to expand this corpus in the future. Finally, we make the annotations and models available for future work. 'jaei g\u00e5r', 'e g\u00e5'.", | |
| "pdf_parse": { | |
| "paper_id": "2021", | |
| "_pdf_hash": "", | |
| "abstract": [ | |
| { | |
| "text": "Norway has a large amount of dialectal variation, as well as a general tolerance to its use in the public sphere. There are, however, few available resources to study this variation and its change over time and in more informal areas, e.g. on social media. In this paper, we propose a first step to creating a corpus of dialectal variation of written Norwegian. We collect a small corpus of tweets and manually annotate them as Bokm\u00e5l, Nynorsk, any dialect, or a mix. We further perform preliminary experiments with state-of-the-art models, as well as an analysis of the data to expand this corpus in the future. Finally, we make the annotations and models available for future work. 'jaei g\u00e5r', 'e g\u00e5'.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Abstract", | |
| "sec_num": null | |
| } | |
| ], | |
| "body_text": [ | |
| { | |
| "text": "Norway has a large tolerance towards dialectal variation (Bull et al., 2018) and, as such, one can find examples of dialectal use in many areas of the public sphere, including politics, news media, and social media. Although there has been much variation in writing Norwegian, since the debut of Nynorsk in the 1850's, the acceptance of dialect use in certain settings is relatively new. The official language policy after World War 2 was to include forms belonging to all layers of society into the written norms, and a \"dialect wave\" has been going on since the 1970's (Bull et al., 2018, 235-238) .", | |
| "cite_spans": [ | |
| { | |
| "start": 57, | |
| "end": 76, | |
| "text": "(Bull et al., 2018)", | |
| "ref_id": "BIBREF4" | |
| }, | |
| { | |
| "start": 571, | |
| "end": 599, | |
| "text": "(Bull et al., 2018, 235-238)", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "From 1980 to 1983 there was an ongoing project called Den f\u00f8rste lese-og skriveopplaering p\u00e5 dialekt 'The first training in reading and writing in dialect' (Bull, 1985) , where primary school students were allowed to use their own dialect in school, with Tove Bull as project leader. Bull et al. * The authors have equal contribution.", | |
| "cite_spans": [ | |
| { | |
| "start": 156, | |
| "end": 168, | |
| "text": "(Bull, 1985)", | |
| "ref_id": "BIBREF3" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "(2018) also point out that later interest in writing in dialect in media such as e-mail and text messages can be seen as an extension of the interest in dialectal writing in the 1980s (Bull et al., 2018, 239) . They also note that the tendency has been the strongest in the county of Tr\u00f8ndelag initially, but later spreading to other parts of the country, also spreading among adults.", | |
| "cite_spans": [ | |
| { | |
| "start": 184, | |
| "end": 208, | |
| "text": "(Bull et al., 2018, 239)", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "At the same time, there are two official main writing systems, i.e. Bokm\u00e5l and Nynorsk, which offer prescriptive rules for how to write the spoken variants. This leads to a situation where people who typically use their dialect when speaking often revert to one of the written standards when writing. However, despite there being only two official writing systems, there is considerable variation within each system, as the result of years of language policies. Today we can find both 'radical' and 'conservative' versions of each writing system, where the radical ones try to bridge the gap between the two norms, while the conservative versions attempt to preserve differences. However, it is still natural that these standards have a regularizing effect on the written varieties of people who normally speak their dialect in most situations (Gal, 2017) . As such, it would be interesting to know to what degree dialect users deviate from these established norms and use dialect traits when writing informal texts, e.g. on social media. This could also provide evidence of the vitality of certain dialectal traits.", | |
| "cite_spans": [ | |
| { | |
| "start": 844, | |
| "end": 855, | |
| "text": "(Gal, 2017)", | |
| "ref_id": "BIBREF6" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "In this paper, we propose a first step towards creating a corpus of written dialectal Norwegian by identifying the best methods to collect, clean, and annotate tweets into Bokm\u00e5l, Nynorsk, or dialectal Norwegian. We concentrate on geolects, rather than sociolects, as we observe these are easier to collect on Twitter, i.e. the traits that identify a geolect are more likely to be written than those that identify a sociolect. This is a necessary simplification, as dialect users rarely write with full phonetic awareness, making it impossible to find dialect traits that lie mainly in the phonology. As such, our corpus relies more on lexical and clear phonetic traits to determine whether a tweet is written in a dialect.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "We collect a corpus of 1,073 tweets which are manually annotated as Bokm\u00e5l, Nynorsk, Dialect, or Mixed and perform a first set of experiments to classify tweets as containing dialectal traits using state-of-the-art methods. We find that fine-tuning a Norwegian BERT model (NB-BERT) leads to the best results. We perform an analysis of the data to find useful features for searching for tweets in the future, confirming several linguistic observations of common dialectal traits and find that certain dialectal traits (those from Tr\u00f8ndelag) are more likely to be written, suggesting that since their traits strongly diverge from Bokm\u00e5l and Nynorsk, they are more likely to deviate from the established norms when composing tweets. Finally, we release the annotations and dialect prediction models for future research. 1", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "The importance of incorporating language variation into natural language processing approaches has gained visibility in recent years. The VarDial workshop series deals with computational methods and language resources for closely related languages, language varieties, and dialects and have offered shared tasks on language variety identification for Romanian, German, Uralic languages (Zampieri et al., 2019) , among others. Similarly, there have been shared tasks on Arabic dialect identification (Bouamor et al., 2019; Abdul-Mageed et al., 2020) . To our knowledge, however, there are no available written dialect identification corpora for Norwegian.", | |
| "cite_spans": [ | |
| { | |
| "start": 386, | |
| "end": 409, | |
| "text": "(Zampieri et al., 2019)", | |
| "ref_id": "BIBREF16" | |
| }, | |
| { | |
| "start": 499, | |
| "end": 521, | |
| "text": "(Bouamor et al., 2019;", | |
| "ref_id": "BIBREF2" | |
| }, | |
| { | |
| "start": 522, | |
| "end": 548, | |
| "text": "Abdul-Mageed et al., 2020)", | |
| "ref_id": "BIBREF0" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Related Work", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "Many successful approaches to dialect identification use linear models (e.g. Support Vector Machines, Multinomial Naive Bayes) with word and character n-gram features (Wu et al., 2019; Jauhiainen et al., 2019a) , while neural approaches often perform poorly (Zampieri et al., 2019 ) (see Jauhiainen et al. (2019b) for a full discussion). More recent uses of pretrained language models based on transformer architectures (Devlin et al., 2019) , however, have shown promise (Bernier-Colborne et al., 2019).", | |
| "cite_spans": [ | |
| { | |
| "start": 167, | |
| "end": 184, | |
| "text": "(Wu et al., 2019;", | |
| "ref_id": "BIBREF15" | |
| }, | |
| { | |
| "start": 185, | |
| "end": 210, | |
| "text": "Jauhiainen et al., 2019a)", | |
| "ref_id": "BIBREF7" | |
| }, | |
| { | |
| "start": 258, | |
| "end": 280, | |
| "text": "(Zampieri et al., 2019", | |
| "ref_id": "BIBREF16" | |
| }, | |
| { | |
| "start": 420, | |
| "end": 441, | |
| "text": "(Devlin et al., 2019)", | |
| "ref_id": "BIBREF5" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Related Work", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "Corpus-related work on Norwegian dialects has mainly focused on spoken varieties. There are two larger corpora available for Norwegian: the newer Nordic Dialect Corpus (Johannessen et al., 2009) , which contains spoken data from several Nordic languages, and the Language Infrastructure made Accessible (LIA) Corpus, which in addition to Norwegian also contains S\u00e1mi language clips. 2 There is also the Talk of Norway Corpus (Lapponi et al., 2018), which contains transcriptions of parliamentary speeches in many language varieties. While they contain rich dialectal information, this information is not kept in writing, as they are normalized to Bokm\u00e5l and Nynorsk. These resources are useful for working with speech technology and questions about Norwegian dialects as they are spoken, but they are likely not sufficient to answer research questions about how dialects are expressed when written. The transcriptions in these corpora also differ from written dialect sources in the sense that they are in a way truer representations of the dialects in question. In writing dialect representations tend to focus more on a few core words, even if the actual phonetic realization of certain words could have been marked in writing.", | |
| "cite_spans": [ | |
| { | |
| "start": 168, | |
| "end": 194, | |
| "text": "(Johannessen et al., 2009)", | |
| "ref_id": "BIBREF9" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Related Work", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "In this first round of annotations, we search for tweets containing Bokm\u00e5l, Nynorsk, and Dialect terms (See Appendix A), discarding tweets that are shorter than 10 tokens. The terms were collected by gathering frequency bigram lists from the Nordic Dialect Corpus (Johannessen et al., 2009) from the written representation of the dialectal varieties.", | |
| "cite_spans": [ | |
| { | |
| "start": 264, | |
| "end": 290, | |
| "text": "(Johannessen et al., 2009)", | |
| "ref_id": "BIBREF9" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Data collection", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "Two native speakers annotated these tweets with four labels: Bokm\u00e5l, Nynorsk, Dialect, and Mixed. The Mixed class refers to tweets where there is a clear separation of dialectal and nondialectal texts, e.g. reported speech in Bokm\u00e5l with comments in Dialect. This class can be very problematic for our classification task, as the content can be a mix of all the other three classes. We nevertheless keep it, as it still reflects one of the written representations of Norwegian.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Data collection", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "In Example 1, we show two phrases from the Nordic Dialect Corpus, from a speaker in Ballangen, Nordland county. We show it in dialectal (1) (a) AE ha l\u00f8sst\u00e5 faer dit. AE har l\u00f8sst\u00e5 g\u00e5 p\u00e5 skole daer. (b) Jeg har lyst\u00e5 fare dit. Jeg har lyst\u00e5 g\u00e5 p\u00e5 skole der. (c) Eg har lyst\u00e5 fara dit. Eg har lyst\u00e5 g\u00e5 p\u00e5 skule der. (d) AE ha l\u00f8sst\u00e5 faer dit. Jeg har lyst\u00e5 g\u00e5 p\u00e5 skole der. (e) I want to go there. I want to go to school there.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Data collection", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "The two annotators doubly annotated a subset of the data in order to assess inter annotator agreement. On a subset of 126 tweets, they achieved a Cohen's Kappa score of 0.76, which corresponds to substantial agreement. Given the strong agreement on this subset, we did not require double annotations for the remaining tweets. Table 1 shows the final distribution of tweets in the training, development, and test splits. Bokm\u00e5l tweets are the most common, followed by Dialect and Nynorsk, and as can be seen, Mixed represents a smaller subset of the data.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 326, | |
| "end": 333, | |
| "text": "Table 1", | |
| "ref_id": "TABREF1" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Data collection", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "Certain traits made the annotation difficult. Many tweets, especially those written in dialect, are informal, and therefore contain more slang and spelling mistakes. For example, jeg 'I' can be misspelled as eg, which if found in a non-Nynorsk setting could indicate dialectal variation. Spelling mistakes should not interfere with dialect identification, but as some tweets can contain as little as one token that serve to identify the language variety as dialectal, this can cause problems. Some dialects are also quite similar to either Bokm\u00e5l or Bokm\u00e5l-Dialect Nynorsk-Dialect 'e' 288.7 'e' 131.8 'ae'", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Data collection", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "188.0 'ae' 92.5 'ska' 55.0 'ska' 23.9 'hu' 36.6 'ei' 18.9 'te' 28.9 'berre' 14.5 ('ae', 'e') 27.5 'hu' 14.4 'ka'", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Data collection", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "22.0 'heilt' 13.8 'mae' 21.6 ('ae', 'e') 13.2 'g\u00e5r'", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Data collection", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "19.9 'meir' 12.3 'va'", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Data collection", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "12.4 'mae' 11.9 Table 2 : Top 10 features and \u03c7 2 values between Bokm\u00e5l -Dialect tweets and Nynorsk -Dialect.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 16, | |
| "end": 23, | |
| "text": "Table 2", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Data collection", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "Nynorsk, and speakers might switch between them when speaking or writing. Similarly, certain elements can be indicative of either a geolect or a sociolect, e.g. the pronoun dem 'they' as the third person plural subject pronoun (de in Bokm\u00e5l and Nynorsk), which in a rural setting might be typical for an East Norwegian dialect, while in an urban setting might be a strong sociolectal indicator. Tweets with similar problems are annotated in favor of the dialect class. Additionally, there is the problem of internal variation. A tweet can belong to a radical or conservative variety of standardized Norwegian, e.g. Riksm\u00e5l, and thereby not be dialectal. However, this distinction can be difficult to make if a writer uses forms that are now removed from the main standards (Bokm\u00e5l and Nynorsk), and therefore become more marked, such as sprog instead of spr\u00e5k 'language'.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Data collection", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "To find the most salient written dialect traits compared to Bokm\u00e5l and Nynorsk, we perform a \u03c7 2 test (Pearson, 1900) on the occurrence of unigrams, bigrams, and trigrams pairwise between Bokm\u00e5l and Dialect, and then Nynorsk and Dialect and set p = 0.5. The most salient features (see Table 2 ) are mainly unigrams that contain dialect features, e.g. ae 'I', e 'am/is/are', ska 'shall/will', te 'to', mae 'me', fr\u00e5 'from', although there are also two statistically significant bigrams, e.g. ae e 'I am', ae ska 'I will'. We notice that many of these features likely correspond to Tr\u00f8ndersk and Nord-norsk variants. Similar features from other dialects (i, jae, je 'I') are not currently found in the corpus. This may reflect the natural usage, but it is also possible that the original search query should be improved. Example 2 shows an example of a Dialect tweet (the English translation is 'Now you know how I've felt for a few years') where the dialectal words have been highlighted in red and marked words , which are not necessarily dialectal, but which often help with classification, have been highlighted in green.", | |
| "cite_spans": [ | |
| { | |
| "start": 102, | |
| "end": 117, | |
| "text": "(Pearson, 1900)", | |
| "ref_id": "BIBREF14" | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 285, | |
| "end": 292, | |
| "text": "Table 2", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Dialectal traits", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "(2) N\u00e5 vet du\u00e5ssen ae har hatt det i noen\u00e5r", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Dialectal traits", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "We propose baseline experiments on a 80/10/10 split for training, development and testing and use a Multinomial Naive Bayes (MNB) and a linear SVM. As features, we use tf-idf word and character (1-5) n-gram features, with a minimum document frequency of 5 for words, and 2 for characters. We use MNB with alpha=0.01, and SVM with hinge loss and regularization of 0.5 and use grid search to identify the best combination of parameters and features.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Experiments", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "We also compare two Norwegian BERT models: NorBERT 3 (Kutuzov et al., 2021) and NB-BERT 4 (Kummervold et al., 2021), which use the same architecture as BERT base cased (Devlin et al., 2019) . NorBERT uses a 28,600 entry Norwegian-specific sentence piece vocabulary and was jointly trained on 200M sentences in Bokm\u00e5l and Nynorsk, while NB-BERT uses the vocabulary from multilingual BERT and is trained on 18 billion tokens from a variety of sources 5 , including historical texts, which presumably contain more examples of written dialect. We use the huggingface transformers implementation and feed the final '[CLS]' embedding to a linear layer, followed by a softmax for classification. The only hyperparameter we optimize is the number of training epochs. We use weight decay on all parameters except for the bias and layer norms and set the learning rate for AdamW (Loshchilov and Hutter, 2019) to 1e\u22125 and set all other hyperparameters to default settings. We train the model for 20 epochs, and keep the model that achieves the best macro F 1 on the dev set. Table 3 shows the results for all models. MNB is the weakest model on both dev and test on all metrics. Despite the fact that it usually gives good results for dialect identification, it is quite clear that it does not fit our dataset. We think that this might mainly be due to the large vocabulary overlap between the classes, especially in the Mixed class. SVM has the best precision on test (0.86), while recall is lower (0.67). NB-BERT has the best recall on both dev and test (0.90/0.78), best precision on dev (0.89), and is the best overall model on test F 1 (0.79), followed by NorBERT.", | |
| "cite_spans": [ | |
| { | |
| "start": 53, | |
| "end": 75, | |
| "text": "(Kutuzov et al., 2021)", | |
| "ref_id": "BIBREF11" | |
| }, | |
| { | |
| "start": 168, | |
| "end": 189, | |
| "text": "(Devlin et al., 2019)", | |
| "ref_id": "BIBREF5" | |
| }, | |
| { | |
| "start": 869, | |
| "end": 898, | |
| "text": "(Loshchilov and Hutter, 2019)", | |
| "ref_id": "BIBREF13" | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 1064, | |
| "end": 1071, | |
| "text": "Table 3", | |
| "ref_id": "TABREF3" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Experiments", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "6 Error analysis Figure 1 shows a confusion matrix of NB-BERT's predictions on the test data. The main three categories (Bokm\u00e5l, Nynorsk, and Dialect) are generally well predicted, while Mixed is currently the hardest category to predict. This is expected, as the Mixed class comprises all of the three other forms. The model has a tendency to predict Nynorsk or Mixed for Dialect and struggles with Mixed, predicting either Bokm\u00e5l or Dialect. The same observations apply to NorBERT, MNB, and SVM classifiers.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 17, | |
| "end": 25, | |
| "text": "Figure 1", | |
| "ref_id": "FIGREF0" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Experiments", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "Given that our main interest lies in the ability to predict future Dialect tweets, we compute precision, recall, and F 1 on only this label. The NB-BERT model achieves 0.82, 0.91, and 0.86, respectively while NorBERT follows with 0.84, 0.77, and 0.81. The SVM model achieves 0.80, 0.69, and 0.74 respectively, while MNB obtains slightly less scores with respectively 0.77, 0.66, and 0.71. This suggests that future experiments should consider using NB-BERT.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Experiments", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "In this paper we have described our first annotation effort to create a corpus of dialectal variation in written Norwegian. In the future, we plan to use our trained models to expand the corpus in a semi-supervised fashion by refining our searches for tweets with dialectal traits in order to have a larger corpus of dialectal tweets, effectively pursuing a high-precision low-recall path. In parallel, we will begin to download large numbers of tweets and use our trained models to automatically annotate these (low-precision, high-recall). At the same time we plan to perform continuous manual evaluations of small amounts of the data in order to identify a larger variety of dialectal tweets, which we will incorporate into the training data for future models.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Conclusion and Future Work", | |
| "sec_num": "7" | |
| }, | |
| { | |
| "text": "Second, we would like to annotate these dialectal tweets with their specific dialect. To avoid collecting too many tweets from overrepresented dialects, we will first annotate the current dialectal tweets with their dialect, and perform a balanced search to find a similar number of tweets for each dialect.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Conclusion and Future Work", | |
| "sec_num": "7" | |
| }, | |
| { | |
| "text": "Finally, we would like to incorporate texts from different sources which contain rich dialectal variation, as e.g. books, music, poetry.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Conclusion and Future Work", | |
| "sec_num": "7" | |
| }, | |
| { | |
| "text": "Available at https://github.com/jerbarnes/ norwegian_dialect", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "https://www.hf.uio.no/iln/english/research/projects/languageinfrastructure-made-accessible/", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "https://huggingface.co/ltgoslo/ norbert 4 https://huggingface.co/NbAiLab/ nb-bert-base 5 See https://github.com/NBAiLab/notram.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| } | |
| ], | |
| "back_matter": [], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "NADI 2020: The first nuanced Arabic dialect identification shared task", | |
| "authors": [ | |
| { | |
| "first": "Muhammad", | |
| "middle": [], | |
| "last": "Abdul-Mageed", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Chiyu", | |
| "middle": [], | |
| "last": "Zhang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Houda", | |
| "middle": [], | |
| "last": "Bouamor", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Nizar", | |
| "middle": [], | |
| "last": "Habash", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2020, | |
| "venue": "Proceedings of the Fifth Arabic Natural Language Processing Workshop", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "97--110", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Muhammad Abdul-Mageed, Chiyu Zhang, Houda Bouamor, and Nizar Habash. 2020. NADI 2020: The first nuanced Arabic dialect identification shared task. In Proceedings of the Fifth Arabic Nat- ural Language Processing Workshop, pages 97-110, Barcelona, Spain (Online). Association for Compu- tational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "Improving cuneiform language identification with BERT", | |
| "authors": [ | |
| { | |
| "first": "Gabriel", | |
| "middle": [], | |
| "last": "Bernier-Colborne", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Cyril", | |
| "middle": [], | |
| "last": "Goutte", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Serge", | |
| "middle": [], | |
| "last": "L\u00e9ger", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2019, | |
| "venue": "Proceedings of the Sixth Workshop on NLP for Similar Languages, Varieties and Dialects", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "17--25", | |
| "other_ids": { | |
| "DOI": [ | |
| "10.18653/v1/W19-1402" | |
| ] | |
| }, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Gabriel Bernier-Colborne, Cyril Goutte, and Serge L\u00e9ger. 2019. Improving cuneiform language iden- tification with BERT. In Proceedings of the Sixth Workshop on NLP for Similar Languages, Varieties and Dialects, pages 17-25, Ann Arbor, Michigan. Association for Computational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "The MADAR shared task on Arabic finegrained dialect identification", | |
| "authors": [ | |
| { | |
| "first": "Houda", | |
| "middle": [], | |
| "last": "Bouamor", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Sabit", | |
| "middle": [], | |
| "last": "Hassan", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Nizar", | |
| "middle": [], | |
| "last": "Habash", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2019, | |
| "venue": "Proceedings of the Fourth Arabic Natural Language Processing Workshop", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "199--207", | |
| "other_ids": { | |
| "DOI": [ | |
| "10.18653/v1/W19-4622" | |
| ] | |
| }, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Houda Bouamor, Sabit Hassan, and Nizar Habash. 2019. The MADAR shared task on Arabic fine- grained dialect identification. In Proceedings of the Fourth Arabic Natural Language Processing Work- shop, pages 199-207, Florence, Italy. Association for Computational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "Lesing og barns talem\u00e5l", | |
| "authors": [ | |
| { | |
| "first": "Tove", | |
| "middle": [], | |
| "last": "Bull", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1985, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Tove Bull. 1985. Lesing og barns talem\u00e5l. Novus, Oslo.", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "Norsk spr\u00e5khistorie", | |
| "authors": [ | |
| { | |
| "first": "Tove", | |
| "middle": [], | |
| "last": "Bull", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Espen", | |
| "middle": [], | |
| "last": "Karlsen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Eli", | |
| "middle": [], | |
| "last": "Raanes", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Rolf", | |
| "middle": [], | |
| "last": "Theil", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2018, | |
| "venue": "", | |
| "volume": "3", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Tove Bull, Espen Karlsen, Eli Raanes, and Rolf Theil. 2018. Norsk spr\u00e5khistorie, volume 3. Novus, Oslo.", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "BERT: Pre-training of deep bidirectional transformers for language understanding", | |
| "authors": [ | |
| { | |
| "first": "Jacob", | |
| "middle": [], | |
| "last": "Devlin", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ming-Wei", | |
| "middle": [], | |
| "last": "Chang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Kenton", | |
| "middle": [], | |
| "last": "Lee", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Kristina", | |
| "middle": [], | |
| "last": "Toutanova", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2019, | |
| "venue": "Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies", | |
| "volume": "1", | |
| "issue": "", | |
| "pages": "4171--4186", | |
| "other_ids": { | |
| "DOI": [ | |
| "10.18653/v1/N19-1423" | |
| ] | |
| }, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of deep bidirectional transformers for language under- standing. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 4171-4186, Minneapolis, Minnesota. Associ- ation for Computational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "Visions and revisions of minority languages: Standardization and its dilemmas", | |
| "authors": [ | |
| { | |
| "first": "Susan", | |
| "middle": [], | |
| "last": "Gal", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2017, | |
| "venue": "Standardizing Minority Languages: Competing Ideologies of Authority and Authenticity in the Global Periphery", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "222--242", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Susan Gal. 2017. Visions and revisions of minority languages: Standardization and its dilemmas. In Pia Lane, James Costa, and Haley de Korne, editors, Standardizing Minority Languages: Competing Ide- ologies of Authority and Authenticity in the Global Periphery, pages 222-242. Routledge.", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "title": "Discriminating between Mandarin Chinese and Swiss-German varieties using adaptive language models", | |
| "authors": [ | |
| { | |
| "first": "Tommi", | |
| "middle": [], | |
| "last": "Jauhiainen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Krister", | |
| "middle": [], | |
| "last": "Lind\u00e9n", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Heidi", | |
| "middle": [], | |
| "last": "Jauhiainen", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2019, | |
| "venue": "Proceedings of the Sixth Workshop on NLP for Similar Languages, Varieties and Dialects", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "178--187", | |
| "other_ids": { | |
| "DOI": [ | |
| "10.18653/v1/W19-1419" | |
| ] | |
| }, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Tommi Jauhiainen, Krister Lind\u00e9n, and Heidi Jauhi- ainen. 2019a. Discriminating between Mandarin Chinese and Swiss-German varieties using adaptive language models. In Proceedings of the Sixth Work- shop on NLP for Similar Languages, Varieties and Dialects, pages 178-187, Ann Arbor, Michigan. As- sociation for Computational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "Language model adaptation for language and dialect identification of text", | |
| "authors": [ | |
| { | |
| "first": "Tommi", | |
| "middle": [], | |
| "last": "Jauhiainen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Krister", | |
| "middle": [], | |
| "last": "Lind\u00e9n", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Heidi", | |
| "middle": [], | |
| "last": "Jauhiainen", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2019, | |
| "venue": "Natural Language Engineering", | |
| "volume": "25", | |
| "issue": "5", | |
| "pages": "561--583", | |
| "other_ids": { | |
| "DOI": [ | |
| "10.1017/S135132491900038X" | |
| ] | |
| }, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Tommi Jauhiainen, Krister Lind\u00e9n, and Heidi Jauhi- ainen. 2019b. Language model adaptation for lan- guage and dialect identification of text. Natural Language Engineering, 25(5):561-583.", | |
| "links": null | |
| }, | |
| "BIBREF9": { | |
| "ref_id": "b9", | |
| "title": "The nordic dialect corpusan advanced research tool", | |
| "authors": [ | |
| { | |
| "first": "Janne", | |
| "middle": [], | |
| "last": "Bondi Johannessen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Joel", | |
| "middle": [ | |
| "James" | |
| ], | |
| "last": "Priestley", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Kristin", | |
| "middle": [], | |
| "last": "Hagen", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2009, | |
| "venue": "Proceedings of the 17th Nordic Conference of Computational Linguistics", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "73--80", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Janne Bondi Johannessen, Joel James Priestley, Kristin Hagen, Tor Anders\u00c5farli, and \u00d8ystein Alexan- der Vangsnes. 2009. The nordic dialect corpus- an advanced research tool. In Proceedings of the 17th Nordic Conference of Computational Linguis- tics (NODALIDA 2009), pages 73-80, Odense, Den- mark. Northern European Association for Language Technology (NEALT).", | |
| "links": null | |
| }, | |
| "BIBREF10": { | |
| "ref_id": "b10", | |
| "title": "Operationalizing a national digital library: The case for a norwegian transformer model", | |
| "authors": [ | |
| { | |
| "first": "Javier", | |
| "middle": [], | |
| "last": "Per Egil Kummervold", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Rosa", | |
| "middle": [], | |
| "last": "De La", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2021, | |
| "venue": "Proceedings of the 23rd Nordic Conference on Computational Linguistics", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Per Egil Kummervold, Javier de la Rosa, Freddy Wet- jen, and Svein Arne Brygfjeld. 2021. Operational- izing a national digital library: The case for a nor- wegian transformer model. In Proceedings of the 23rd Nordic Conference on Computational Linguis- tics (NoDaLiDa 2021).", | |
| "links": null | |
| }, | |
| "BIBREF11": { | |
| "ref_id": "b11", | |
| "title": "Large-scale contextualised language modelling for norwegian", | |
| "authors": [ | |
| { | |
| "first": "Andrey", | |
| "middle": [], | |
| "last": "Kutuzov", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jeremy", | |
| "middle": [], | |
| "last": "Barnes", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Erik", | |
| "middle": [], | |
| "last": "Velldal", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Lilja", | |
| "middle": [], | |
| "last": "\u00d8vrelid", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Stephan", | |
| "middle": [], | |
| "last": "Oepen", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2021, | |
| "venue": "Proceedings of the 23rd Nordic Conference on Computational Linguistics", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Andrey Kutuzov, Jeremy Barnes, Erik Velldal, Lilja \u00d8vrelid, and Stephan Oepen. 2021. Large-scale contextualised language modelling for norwegian. In Proceedings of the 23rd Nordic Conference on Computational Linguistics (NoDaLiDa 2021).", | |
| "links": null | |
| }, | |
| "BIBREF12": { | |
| "ref_id": "b12", | |
| "title": "The talk of norway: a richly annotated corpus of the norwegian parliament, 1998-2016. Language Resources and Evaluation", | |
| "authors": [ | |
| { | |
| "first": "Emanuele", | |
| "middle": [], | |
| "last": "Lapponi", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Martin", | |
| "middle": [], | |
| "last": "S\u00f8yland", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Erik", | |
| "middle": [], | |
| "last": "Velldal", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Stephan", | |
| "middle": [], | |
| "last": "Oepen", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2018, | |
| "venue": "", | |
| "volume": "52", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": { | |
| "DOI": [ | |
| "10.1007/s10579-018-9411-5" | |
| ] | |
| }, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Emanuele Lapponi, Martin S\u00f8yland, Erik Velldal, and Stephan Oepen. 2018. The talk of norway: a richly annotated corpus of the norwegian parlia- ment, 1998-2016. Language Resources and Eval- uation, 52.", | |
| "links": null | |
| }, | |
| "BIBREF13": { | |
| "ref_id": "b13", | |
| "title": "Decoupled weight decay regularization", | |
| "authors": [ | |
| { | |
| "first": "Ilya", | |
| "middle": [], | |
| "last": "Loshchilov", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Frank", | |
| "middle": [], | |
| "last": "Hutter", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2019, | |
| "venue": "International Conference on Learning Representations", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Ilya Loshchilov and Frank Hutter. 2019. Decoupled weight decay regularization. In International Con- ference on Learning Representations.", | |
| "links": null | |
| }, | |
| "BIBREF14": { | |
| "ref_id": "b14", | |
| "title": "X. on the criterion that a given system of deviations from the probable in the case of a correlated system of variables is such that it can be reasonably supposed to have arisen from random sampling", | |
| "authors": [ | |
| { | |
| "first": "Karl", | |
| "middle": [], | |
| "last": "Pearson", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1900, | |
| "venue": "The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science", | |
| "volume": "50", | |
| "issue": "302", | |
| "pages": "157--175", | |
| "other_ids": { | |
| "DOI": [ | |
| "10.1080/14786440009463897" | |
| ] | |
| }, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Karl Pearson. 1900. X. on the criterion that a given system of deviations from the probable in the case of a correlated system of variables is such that it can be reasonably supposed to have arisen from random sampling. The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Sci- ence, 50(302):157-175.", | |
| "links": null | |
| }, | |
| "BIBREF15": { | |
| "ref_id": "b15", | |
| "title": "Language discrimination and transfer learning for similar languages: Experiments with feature combinations and adaptation", | |
| "authors": [ | |
| { | |
| "first": "Nianheng", | |
| "middle": [], | |
| "last": "Wu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Eric", | |
| "middle": [], | |
| "last": "Demattos", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Kwok Him So", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Pin-Zhen Chen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "\u00c7\u00f6ltekin", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2019, | |
| "venue": "Proceedings of the Sixth Workshop on NLP for Similar Languages, Varieties and Dialects", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "54--63", | |
| "other_ids": { | |
| "DOI": [ | |
| "10.18653/v1/W19-1406" | |
| ] | |
| }, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Nianheng Wu, Eric DeMattos, Kwok Him So, Pin-zhen Chen, and \u00c7 agr\u0131 \u00c7\u00f6ltekin. 2019. Language discrim- ination and transfer learning for similar languages: Experiments with feature combinations and adapta- tion. In Proceedings of the Sixth Workshop on NLP for Similar Languages, Varieties and Dialects, pages 54-63, Ann Arbor, Michigan. Association for Com- putational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF16": { | |
| "ref_id": "b16", | |
| "title": "A report on the third VarDial evaluation campaign", | |
| "authors": [ | |
| { | |
| "first": "Marcos", | |
| "middle": [], | |
| "last": "Zampieri", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Shervin", | |
| "middle": [], | |
| "last": "Malmasi", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Yves", | |
| "middle": [], | |
| "last": "Scherrer", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Tanja", | |
| "middle": [], | |
| "last": "Samard\u017ei\u0107", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Francis", | |
| "middle": [], | |
| "last": "Tyers", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Miikka", | |
| "middle": [], | |
| "last": "Silfverberg", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Natalia", | |
| "middle": [], | |
| "last": "Klyueva", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Tung-Le", | |
| "middle": [], | |
| "last": "Pan", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Chu-Ren", | |
| "middle": [], | |
| "last": "Huang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Radu", | |
| "middle": [ | |
| "Tudor" | |
| ], | |
| "last": "Ionescu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Andrei", | |
| "middle": [ | |
| "M" | |
| ], | |
| "last": "Butnaru", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Tommi", | |
| "middle": [], | |
| "last": "Jauhiainen", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2019, | |
| "venue": "Proceedings of the Sixth Workshop on NLP for Similar Languages, Varieties and Dialects", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "1--16", | |
| "other_ids": { | |
| "DOI": [ | |
| "10.18653/v1/W19-1401" | |
| ] | |
| }, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Marcos Zampieri, Shervin Malmasi, Yves Scherrer, Tanja Samard\u017ei\u0107, Francis Tyers, Miikka Silfver- berg, Natalia Klyueva, Tung-Le Pan, Chu-Ren Huang, Radu Tudor Ionescu, Andrei M. Butnaru, and Tommi Jauhiainen. 2019. A report on the third VarDial evaluation campaign. In Proceedings of the Sixth Workshop on NLP for Similar Languages, Va- rieties and Dialects, pages 1-16, Ann Arbor, Michi- gan. Association for Computational Linguistics.", | |
| "links": null | |
| }, | |
| "BIBREF17": { | |
| "ref_id": "b17", | |
| "title": "daei bli', 'daemm bli', 'daem bli', 'di blir', 'demm bli', 'dem bli', 'daemm bi', 'daem bi', 'd\u00f8mm bli', 'd\u00f8m bli', 'd\u00f8mm bi', 'd\u00f8m bi', 'di har', 'di ha', 'daemm ha', 'daem ha', 'daemm har', 'daem har', 'daei he', 'demm har', 'dem har', 'demm ha', 'dem ha', 'daei ha', 'di he", | |
| "authors": [ | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "De G\u00e5r", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "De Blir", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "De Har", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "Nynorsk Terms: 'eg Har", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "De G\u00e5r", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "De Blir", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "De Har", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "Ho Gaar", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "Ho Blir", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "Ho Har", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "Ho Er", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "Ha", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "Di G\u00e5r ; E Bli", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "'", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": null, | |
| "venue": "A Appendix Bokm\u00e5l terms: 'jeg har", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "A Appendix Bokm\u00e5l terms: 'jeg har', 'de g\u00e5r', 'jeg skal', 'jeg blir', 'de skal', 'jeg er', 'de blir', 'de har', 'de er', 'dere g\u00e5r', 'dere skal', 'dere blir', 'dere har', 'dere er', 'hun g\u00e5r', 'hun skal', 'hun blir', 'hun har', 'hun er', 'jeg g\u00e5r'. Nynorsk terms: 'eg har', 'dei g\u00e5r', 'eg skal', 'eg blir', 'dei skal', 'eg er', 'dei blir', 'dei har', 'dei er', 'de g\u00e5r', 'dykk g\u00e5r','de skal','dykk skal','de blir','dykk blir','de har','dykk har','de er','dykk er', 'ho gaar', 'ho skal', 'ho blir', 'ho har', 'ho er', 'eg g\u00e5r'. Dialect terms: 'e ha', 'ae ha', 'ae har', 'e har', 'jae ha', 'eg har', 'eg ha', 'je ha', 'jae har', 'di g\u00e5r', 'demm g\u00e5r', 'dem g\u00e5r', 'daemm g\u00e5r', 'daem g\u00e5r', 'daei g\u00e5r', 'demm g\u00e5', 'dem g\u00e5', 'di g\u00e5r', 'domm g\u00e5', 'dom g\u00e5', 'd\u00f8mm g\u00e5r', 'd\u00f8m g\u00e5r', 'daemm g\u00e5', 'daem g\u00e5', 'e ska', 'ae ska', 'jae ska', 'eg ska', 'je ska', 'i ska', 'ei ska', 'jaei ska', 'je skae', 'e bli', 'ae bli', 'jae bli', 'e bi', 'ae blir', 'ae bi', 'je bli', 'e blir', 'i bli', 'di ska', 'daemm ska', 'daem ska', 'daei ska', 'demm ska', 'dem ska', 'domm ska', 'dom ska', 'd\u00f8mm ska', 'd\u00f8m ska', 'dae ska', 'domm ska', 'dom ska', 'aemm ska', 'aem ska', 'eg e', 'ae e', 'e e', 'jae ae', 'e ae', 'jae aer', 'je ae', 'i e', 'aeg e', 'di bi', 'di bli', 'daei bli', 'daemm bli', 'daem bli', 'di blir', 'demm bli', 'dem bli', 'daemm bi', 'daem bi', 'd\u00f8mm bli', 'd\u00f8m bli', 'd\u00f8mm bi', 'd\u00f8m bi', 'di har', 'di ha', 'daemm ha', 'daem ha', 'daemm har', 'daem har', 'daei he', 'demm har', 'dem har', 'demm ha', 'dem ha', 'daei ha', 'di he', 'daemm e', 'daem e', 'di e', 'daei e', 'demm e', 'dem e', 'di ae', 'd\u00f8mm ae', 'd\u00f8m ae', 'demm ae', 'dem ae', 'dei e', 'daei ae', 'd\u00e5kk g\u00e5r', 'd\u00e5kke g\u00e5r', 'd\u00e5kke g\u00e5', 'de g\u00e5r', 'd\u00e5kk ska', 'dere ska', 'd\u00e5kker ska', 'd\u00e5kke ska', 'di ska', 'de ska', '\u00e5kk ska', 'r\u00f8kk ska', 'd\u00f8kker ska', 'd\u00f8kk bli', 'd\u00e5kker bi', 'd\u00e5kke bli', 'd\u00e5kker har', 'd\u00e5kker ha', 'dere ha', 'd\u00e5kk ha', 'de har', 'd\u00e5kk har', 'dere har', 'de ha', 'd\u00f8kk ha', 'd\u00e5kker e', 'd\u00e5kk e', 'd\u00e5kke e', 'di e', 'dere aer', 'd\u00e5kk ae', 'de e', '\u00f8kk e', 'd\u00f8kk ae', 'ho g\u00e5r', 'hu g\u00e5r', 'ho jenng', 'ho gjenng', 'u g\u00e5r', 'o g\u00e5r', 'ho jaenng', 'ho gjaenng', 'ho jenngg', 'ho gjen- ngg', 'ho jennge', 'ho gjennge', 'ho g\u00e5', 'ho ska', 'hu ska', 'a ska', 'u ska', 'o ska', 'hu skar', 'honn ska', 'ho sjka', 'haenne ska', 'ho bli', 'ho bi', 'o bli', 'ho blir', 'hu bli', 'hu bler', 'hu bi', 'ho bir', 'a blir', 'ho ha', 'ho har', 'ho he', 'hu har', 'hu ha', 'hu he', 'o har', 'o ha', 'hu e', 'ho e', 'hu e', 'ho ae', 'hu ae', 'o e', 'hu aer', 'u e', 'ho aer', 'ho er', 'e g\u00e5r', 'ae g\u00e5r', 'eg g\u00e5r', 'jae g\u00e5', 'jae g\u00e5r', 'ae g\u00e5',", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "FIGREF0": { | |
| "num": null, | |
| "type_str": "figure", | |
| "uris": null, | |
| "text": "Confusion matrix of NB-BERT on Bokm\u00e5l (BK), Nynorsk (NN), Dialect (DI), and Mixed (MIX)." | |
| }, | |
| "TABREF1": { | |
| "num": null, | |
| "content": "<table><tr><td>form (a) and the Bokm\u00e5l (b) transcription, but with</td></tr><tr><td>added punctuation marks. To exemplify the two</td></tr><tr><td>other categories we have manually translated it to</td></tr><tr><td>Nynorsk (c) and added a mixed version (d), as well</td></tr><tr><td>as an English translation (e) for reader comprehen-</td></tr><tr><td>sion.</td></tr></table>", | |
| "text": "Data statistics for the corpus, including number of tweets per split.", | |
| "html": null, | |
| "type_str": "table" | |
| }, | |
| "TABREF3": { | |
| "num": null, | |
| "content": "<table><tr><td>BK</td><td>36</td><td>0</td><td>0</td><td>2</td></tr><tr><td>NN</td><td>0</td><td>28</td><td>3</td><td>0</td></tr><tr><td>DI</td><td>2</td><td>1</td><td>32</td><td>0</td></tr><tr><td>MIX</td><td>0</td><td>0</td><td>4</td><td>2</td></tr><tr><td/><td>BK</td><td>NN</td><td>DI</td><td>MIX</td></tr></table>", | |
| "text": "Precision, recall, and macro F 1 for each model, on the dev and test sets.", | |
| "html": null, | |
| "type_str": "table" | |
| } | |
| } | |
| } | |
| } |