{
"paper_id": "2021",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T03:14:46.369584Z"
},
"title": "Quantitative Evaluation of Alternative Translations in a Corpus of Highly Dissimilar Finnish Paraphrases",
"authors": [
{
"first": "Li-Hsin",
"middle": [],
"last": "Chang",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Turku",
"location": {
"country": "Finland"
}
},
"email": "lhchan@utu.fi"
},
{
"first": "Sampo",
"middle": [],
"last": "Pyysalo",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Turku",
"location": {
"country": "Finland"
}
},
"email": ""
},
{
"first": "Jenna",
"middle": [],
"last": "Kanerva",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Turku",
"location": {
"country": "Finland"
}
},
"email": ""
},
{
"first": "Filip",
"middle": [],
"last": "Ginter",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Turku",
"location": {
"country": "Finland"
}
},
"email": ""
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "In this paper, we present a quantitative evaluation of differences between alternative translations in a large recently released Finnish paraphrase corpus focusing in particular on non-trivial variation in translation. We combine a series of automatic steps detecting systematic variation with manual analysis to reveal regularities and identify categories of translation differences. We find the paraphrase corpus to contain highly non-trivial translation variants difficult to recognize through automatic approaches.",
"pdf_parse": {
"paper_id": "2021",
"_pdf_hash": "",
"abstract": [
{
"text": "In this paper, we present a quantitative evaluation of differences between alternative translations in a large recently released Finnish paraphrase corpus focusing in particular on non-trivial variation in translation. We combine a series of automatic steps detecting systematic variation with manual analysis to reveal regularities and identify categories of translation differences. We find the paraphrase corpus to contain highly non-trivial translation variants difficult to recognize through automatic approaches.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "The study of translation language for Finnish has largely focused on individual linguistic features. The debate on the existence of translation universals sparked the well-developed research line of comparing translated and original language. Examples of such studies include the comparison of nonfinite structures in translated and original Finnish (Puurtinen, 2003; Eskola, 2004) , and investigation of subject changes in translations using a French-Finnish parallel corpus (Huotari, 2021) . Variation in alternative translations is less studied. Paloposki and Koskinen (2004) qualitatively compare the degree of domestication in language use in Finnish first translations and retranslations. While this study is done qualitatively, several paraphrase corpora with translated language have been released more recently, enabling research from a quantitative prospective. Such corpora include Opusparcus (Creutz, 2018) and TaPaCo (Scherrer, 2020) , both constructed automatically using language pivoting and containing Finnish subsets.",
"cite_spans": [
{
"start": 350,
"end": 367,
"text": "(Puurtinen, 2003;",
"ref_id": "BIBREF12"
},
{
"start": 368,
"end": 381,
"text": "Eskola, 2004)",
"ref_id": "BIBREF2"
},
{
"start": 476,
"end": 491,
"text": "(Huotari, 2021)",
"ref_id": "BIBREF3"
},
{
"start": 549,
"end": 578,
"text": "Paloposki and Koskinen (2004)",
"ref_id": "BIBREF11"
},
{
"start": 904,
"end": 918,
"text": "(Creutz, 2018)",
"ref_id": "BIBREF1"
},
{
"start": 930,
"end": 946,
"text": "(Scherrer, 2020)",
"ref_id": "BIBREF14"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Recently, the Turku Paraphrase Corpus has become available (Kanerva et al., 2021) , consisting of paraphrase pairs, of which the vast major-ity are manually selected from the OpenSubtitles 1 dataset. The construction of the paraphrase corpus capitalizes on the fact that many movies and TV shows have multiple independently produced translations. The selection is carried out manually, comparing side-by-side the two lexically maximally distant subtitle versions for each movie or TV show and selecting instances of paraphrases. Upon selection, the candidate pairs are assigned to a category such as paraphrase in any context or paraphrase in this context but not universally, etc. The Turku paraphrase corpus is substantial in size, with 45,000 manually extracted, naturally occurring paraphrase pairs (a paraphrase pair henceforth refers to two segments of text, each about a sentence long or slightly longer), and a further 7,900 pairs created by editing an extracted pair so as to obtain a fully context-independent paraphrase.",
"cite_spans": [
{
"start": 59,
"end": 81,
"text": "(Kanerva et al., 2021)",
"ref_id": "BIBREF4"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Due to the way in which it was constructed, the corpus is directly applicable to the study of translation language and in particular to the analysis of variation in translation. The unique value of the corpus for this purpose is that it consists mostly of fully manually selected translation variants focused on lexically and structurally dissimilar pairs. These are very difficult to extract automatically: automatic methods can reliably identify only simple variation, while lexically and structurally substantially different pairs are very difficult to automatically distinguish from non-paraphrases, i.e. phrases that are not alternative translations.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In this paper, we will characterize the paraphrase corpus in terms of translation language, focusing especially on the types of variation (e.g. synonym usage, redundancy or verbosity) occurring in the data. Our aim is to establish whether the corpus can be of utility to translation language modelling and machine translation system evaluation. To this end, we will focus on two main ques-tions: (a) how easily could the translation pairs be extracted automatically, and (b) what are the main types of variation exhibited by the pairs.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The full corpus includes 45,000 naturally occurring paraphrases and 7,900 pairs obtained by rewriting a previously extracted example. The source of these paraphrases is in the vast majority of cases alternative translations of subtitles, with a small section originating from news headings. To construct a lexically and structurally diverse paraphrase corpus, the annotators were instructed to only select non-trivial paraphrase candidates, avoiding simple, uninteresting changes such as minor differences in inflection and word order. 2 For the analysis in this paper, we use the training section of the corpus, restricting further exclusively to examples originating from Open-Subtitles. This gives 34,561 naturally occurring paraphrase pairs and 5,445 rewritten paraphrases. Each naturally occurring paraphrase pair in the corpus have a numerical label manually assigned by an annotator from the following set: 4: universally paraphrase regardless of context, 3: paraphrase in the given context but not universally, 2: related but not paraphrase. Additionally, those annotated as 4 can be assigned one or several flags which sub-categorize different types of paraphrases: > or <: universal paraphrase in one direction but not the other, s: substantial difference in style, i: meaning-affecting difference restricted to a small number of morphosyntactic features. By contrast to the original paraphrases, the rewrites are always full, universally valid paraphrases, i.e. label 4. The rewriting process strives to change as little of the original sentences as possible: these include simple fixes such as word or phrase deletion, addition or re-placement with a synonym or changing an inflection, while more complicated changes are avoided. The rewrites are thus an efficient way to obtain full paraphrases in terms of corpus creation. The label distribution of the Turku paraphrase corpus subset used for later analysis is shown in Table 1 .",
"cite_spans": [],
"ref_spans": [
{
"start": 1934,
"end": 1941,
"text": "Table 1",
"ref_id": "TABREF1"
}
],
"eq_spans": [],
"section": "Corpus statistics and pre-processing",
"sec_num": "2"
},
{
"text": "For the purpose of the subsequent analysis, we parse the paraphrases using the Turku Neural Parser Pipeline (Kanerva et al., 2018 (Kanerva et al., , 2020 , a state-of-the-art parser producing POS and mor- phological tags, word lemmas, as well as dependency trees in the Universal Dependencies scheme (Nivre et al., 2016) . We use the model trained on UD Finnish-TDT v2.7 corpus, which utilizes the pre-trained FinBERT language model in tagging and dependency parsing (Virtanen et al., 2019). 3 3 Analysis of variation",
"cite_spans": [
{
"start": 108,
"end": 129,
"text": "(Kanerva et al., 2018",
"ref_id": "BIBREF5"
},
{
"start": 130,
"end": 153,
"text": "(Kanerva et al., , 2020",
"ref_id": "BIBREF6"
},
{
"start": 300,
"end": 320,
"text": "(Nivre et al., 2016)",
"ref_id": "BIBREF9"
},
{
"start": 467,
"end": 493,
"text": "(Virtanen et al., 2019). 3",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Corpus statistics and pre-processing",
"sec_num": "2"
},
{
"text": "To investigate and categorize the paraphrase pairs by the form of variation, we calculate the difference in the set of lemmas (i.e. insertions/deletions of lemma, henceforth lemma indels) for each pair, excluding punctuation characters from the analysis. Figure 1 shows the distribution of the number of lemma indels for all universal paraphrases showed in We next investigate paraphrase pairs that can be accounted for by automatic synonym substitutions. We combine two resources to build a synonym dictionary for lemmas. The first resource is Word2Vec embeddings (Mikolov et al., 2013) for lemmas trained from Suomi24 discussion fora texts 4 . For each lemma, we take at most 15 closest lemmas in the vector space as synonyms using the gensim library (\u0158eh\u016f\u0159ek and Sojka, 2010). In addition, we supplement our synonym dictionary with Finnish WordNet (Lind\u00e9n and Niemi, 2014) using the NLTK library (Bird et al., 2009) . Out of the 14,878 pairs of paraphrases with lemma indels, 951 pairs (\u223c6%) have all of their lemma indels accounted by synonyms. An additional 7370 pairs (\u223c49%) have lemma indels partially accounted by synonyms. The synonym dictionary only takes into account one-to-one synonyms. As a consequence, one-to-many synonyms and phrasal paraphrases are not included. Table 2 shows the lemmas that are most overrepresented among the inserted or deleted words relative to their overall frequency. We find emphasizers (e.g. tosi (really)), particles (e.g. kyll\u00e4 (truly)), auxiliary verbs, other functional words, and a small number of very common synonym pairs among the most frequently varying words.",
"cite_spans": [
{
"start": 565,
"end": 587,
"text": "(Mikolov et al., 2013)",
"ref_id": "BIBREF8"
},
{
"start": 851,
"end": 875,
"text": "(Lind\u00e9n and Niemi, 2014)",
"ref_id": "BIBREF7"
},
{
"start": 899,
"end": 918,
"text": "(Bird et al., 2009)",
"ref_id": "BIBREF0"
}
],
"ref_spans": [
{
"start": 255,
"end": 263,
"text": "Figure 1",
"ref_id": "FIGREF0"
},
{
"start": 1281,
"end": 1288,
"text": "Table 2",
"ref_id": "TABREF3"
}
],
"eq_spans": [],
"section": "Automatic categorization",
"sec_num": "3.1"
},
{
"text": "To further focus on meaningful variation, we disregard all words with a dependency relation deemed functional in the Content-Word Labeled Attachment Score (CLAS) (Nivre and Fang, 2017) , which is developed to evaluate dependency parsing with focus on content-bearing words. 5 After disregarding these functional words, we are able to account for the variation in a further 82 paraphrase pairs. All of the above mentioned findings are summarized in Table 3 . As the variation in 13,608 pairs (i.e. full 90% of the data) is not accountable by using the above automatic categories, we characterize these manually.",
"cite_spans": [
{
"start": 162,
"end": 184,
"text": "(Nivre and Fang, 2017)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [
{
"start": 448,
"end": 455,
"text": "Table 3",
"ref_id": "TABREF5"
}
],
"eq_spans": [],
"section": "Automatic categorization",
"sec_num": "3.1"
},
{
"text": "In the manual categorization, we sample 100 paraphrase pairs among those paraphrases where the variation is not fully explainable using the automatic metrics defined above. Each paraphrase pair is annotated in terms of 8 different variation categories: word-to-word, word-to-phrase and phraseto-phrase synonyms indicating a straightforward single word synonym replacement, a single word replaced with a synonymous phrase, or a phrase replaced with a synonymous phrase, redundancy or verbosity for including additional words not strictly essential for the meaning, explicit pronouns for explicitly including pronouns visible otherwise in the verb inflection, emphasizer for including additional emphasis words (such as very), figurative language/idioms, and uncertainty or hedging where both statements express hedging with different markers. For each paraphrase pair a set of categories explaining the variation is annotated. In Table 4 Table 4 : Manual analysis results plot the frequency of each category, showing the straightforward single word synonym replacement being by far the most frequent category, occurring in 61% of the paraphrase pairs. However, albeit word-to-word replacement being frequent, it rarely accounts for the whole variation in the pair. Only 12% of the paraphrases include wordto-word synonyms as sole variation category, other instances occurring in combination with at least one additional variation category.",
"cite_spans": [],
"ref_spans": [
{
"start": 929,
"end": 936,
"text": "Table 4",
"ref_id": null
},
{
"start": 937,
"end": 944,
"text": "Table 4",
"ref_id": null
}
],
"eq_spans": [],
"section": "Manual categorization",
"sec_num": "3.2"
},
{
"text": "We measure the proportion of non-elementary variation in the alternative translations in terms of percentage of text (in terms of alphanumeric characters) in the manually extracted paraphrase pairs, out of the total amount of the source material that the annotators processed. The proportion is 15.8%, meaning that approximately every sixth line was considered to be dissimilar in an interesting manner by the annotators, enough to be included in the paraphrase corpus. The remaining 84% of the text is reported by the corpus creators to be for the most part elementary variation, text without correspondence in the other subtitle version, conflicting erroneous translations, and rarely pairs that are meaningless without deep understanding of their broader context.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Amount of Non-elementary Variation",
"sec_num": "3.3"
},
{
"text": "To establish the proportion of the manually extracted paraphrase pairs that could be identified through their source text, as well as to establish the feasibility of automatically aligning the paraphrase pairs with their English source, we use the OpenSubtitles section of the OPUS machine translation dataset and identify those pairs in our dataset that have at least one common English source segment in the English-Finnish OpenSubtitles section of OPUS. We normalize both Finnish and English texts by lowercasing and dropping all non-alphanumeric characters so as to maximize the recall. Such language pivoting is a common technique for mining cases of translation variation. Language pivoting targets candidates, where the same source-language segment is translated into two different target-language segments, using a corpus of aligned bilingual document pairs. The candidates are typically further filtered by various means to remove spurious alignments and other pairs which are not equivalent in meaning, despite sharing the same aligned source-language segment.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Language pivoting",
"sec_num": "3.4"
},
{
"text": "We find that 2,136 pairs were matched, a mere 6% of all categories of paraphrase in the corpus (barring rewrites). Full 94% of the paraphrase pairs cannot be reached through simple language pivoting at least on the level of full segments. Further, while the average length of texts found through pivoting is 3.8 tokens, the average length of texts in the data is 8.4 tokens. The pivoting thus unsurprisingly biases towards short segments, that are more likely to be appropriately aligned and identified. Clearly, in order to align the paraphrase pairs with their (mostly English) source, a manual annotation step will be necessary.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Language pivoting",
"sec_num": "3.4"
},
{
"text": "In this paper, we have presented a quantitative analysis of a large, manually extracted paraphrase dataset from the point of view of translation language, and especially its non-elementary variation. Our findings are two-fold. Firstly, we demonstrated that in the case of OpenSubtitles -a very widely used corpus in machine translation -the proportion of non-elementary variation in alternate translations is relatively small, at 16% of the text. Secondly, we have shown that the paraphrase corpus contains highly non-trivial translation variants that are difficult to account for through simple heuristics and can thus serve for further study in translation language without biasing the results towards simpler examples that can be identified automatically.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Discussion, Conclusions and Future Work",
"sec_num": "4"
},
{
"text": "The corpus in its current form can serve as a resource for evaluating robustness of different evaluation metrics. Quora Question Pairs (QQP) 6 and the QQP subset of Paraphrase Adversaries from Word Scrambling (PAWS) (Zhang et al., 2019) have been used to evaluate the robustness of machine translation and image captioning metrics (Zhang et al., 2020) . QQP is a collection of question headings from the Quora forum labeled as either duplicate or not, while PAWS is an adversarial dataset automatically generated from QQP and Wikipedia to contain highly lexically similar paraphrases and non-paraphrases. Based on our findings, the Turku paraphrase corpus serves as an interesting resource to be used in a similar manner to evaluate metric robustness. An obvious direction for future work is to align, through a combination of heuristics and manual annotation, the paraphrase pairs with their English source. This would result in a test set suitable for evaluation of machine translation systems in terms of their rephrasing ability, as well as for research on MT system evaluation methodology in presence of substantial rephrasing.",
"cite_spans": [
{
"start": 216,
"end": 236,
"text": "(Zhang et al., 2019)",
"ref_id": "BIBREF17"
},
{
"start": 331,
"end": 351,
"text": "(Zhang et al., 2020)",
"ref_id": "BIBREF16"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Discussion, Conclusions and Future Work",
"sec_num": "4"
},
{
"text": "http://www.opensubtitles.org",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "Finnish has relatively free word order and reordering can be trivially detected automatically.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "Model available at https://turkunlp.org/ Turku-neural-parser-pipeline/models.html",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "dl.turkunlp.org/finnish-embeddings/ finnish_s24_skgram_lemmas.bin",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "These dependency relations are aux (auxiliary), aux:pass (passive auxiliary), case (pre/postposition), cc (coordinating conjunction), clf (classifier), cop (copula), det (determiner), mark (marker), punct (punctuation), cc:preconj (preconjunct), and cop:own (copula in possessive clauses).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "data.quora.com/First-Quora-Dataset-\\ Release-Question-Pairs",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "The research presented in this paper was partially supported by the European Language Grid project through its open call for pilot projects. The European Language Grid project has received funding from the European Union's Horizon 2020 Research and Innovation programme under Grant Agreement no. 825627 (ELG). The research was also supported by the Academy of Finland and the DigiCampus project. Computational resources were provided by CSC -the Finnish IT Center for Science. We thank Veronika Laippala for her advice from a linguistic point of view.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgments",
"sec_num": null
},
{
"text": "A Example instances of manual analysis categories Wordword Table 5 : Examples of manual analysis categories. English translations in Table 6 . [...] It seems that the kids [...] Table 6 : Examples of manual analysis categories, best-effort translation to English.",
"cite_spans": [
{
"start": 143,
"end": 148,
"text": "[...]",
"ref_id": null
}
],
"ref_spans": [
{
"start": 59,
"end": 66,
"text": "Table 5",
"ref_id": null
},
{
"start": 133,
"end": 140,
"text": "Table 6",
"ref_id": null
},
{
"start": 178,
"end": 185,
"text": "Table 6",
"ref_id": null
}
],
"eq_spans": [],
"section": "annex",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Natural Language Processing with Python",
"authors": [
{
"first": "Steven",
"middle": [],
"last": "Bird",
"suffix": ""
},
{
"first": "Ewan",
"middle": [],
"last": "Klein",
"suffix": ""
},
{
"first": "Edward",
"middle": [],
"last": "Loper",
"suffix": ""
}
],
"year": 2009,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Steven Bird, Ewan Klein, and Edward Loper. 2009. Natural Language Processing with Python. O'Reilly Media, Inc.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Open Subtitles paraphrase corpus for six languages",
"authors": [
{
"first": "Mathias",
"middle": [],
"last": "Creutz",
"suffix": ""
}
],
"year": 2018,
"venue": "Proceedings of the Eleventh International Conference on Language Resources and Evaluation",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Mathias Creutz. 2018. Open Subtitles paraphrase corpus for six languages. In Proceedings of the Eleventh International Conference on Language Re- sources and Evaluation (LREC 2018).",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Untypical frequencies in translated language: A corpus-based study on a literary corpus of translated and non-translated Finnish",
"authors": [
{
"first": "Sari",
"middle": [
"Eskola"
],
"last": "",
"suffix": ""
}
],
"year": 2004,
"venue": "Translation Universals: Do they exist?",
"volume": "",
"issue": "",
"pages": "83--99",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Sari Eskola. 2004. Untypical frequencies in translated language: A corpus-based study on a literary corpus of translated and non-translated Finnish. In Anna Mauranen and Pekka Kujam\u00e4ki, editors, Translation Universals: Do they exist?, pages 83 -99. John Ben- jamins Publishing Company.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Effet du prototype sur le changement de sujet en traduction :\u00c9tude d'un corpus bidirectionnel litt\u00e9raire fran\u00e7ais\u2194finnois",
"authors": [
{
"first": "L\u00e9a",
"middle": [],
"last": "Huotari",
"suffix": ""
}
],
"year": 2021,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "L\u00e9a Huotari. 2021. Effet du prototype sur le change- ment de sujet en traduction :\u00c9tude d'un corpus bidi- rectionnel litt\u00e9raire fran\u00e7ais\u2194finnois. Ph.D. thesis, University of Helsinki.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Finnish paraphrase corpus",
"authors": [
{
"first": "Jenna",
"middle": [],
"last": "Kanerva",
"suffix": ""
},
{
"first": "Filip",
"middle": [],
"last": "Ginter",
"suffix": ""
},
{
"first": "Li-Hsin",
"middle": [],
"last": "Chang",
"suffix": ""
},
{
"first": "Iiro",
"middle": [],
"last": "Rastas",
"suffix": ""
},
{
"first": "Valtteri",
"middle": [],
"last": "Skantsi",
"suffix": ""
},
{
"first": "Jemina",
"middle": [],
"last": "Kilpel\u00e4inen",
"suffix": ""
},
{
"first": "Hanna-Mari",
"middle": [],
"last": "Kupari",
"suffix": ""
},
{
"first": "Jenna",
"middle": [],
"last": "Saarni",
"suffix": ""
},
{
"first": "Maija",
"middle": [],
"last": "Sev\u00f3n",
"suffix": ""
},
{
"first": "Otto",
"middle": [],
"last": "Tarkka",
"suffix": ""
}
],
"year": 2021,
"venue": "Proceedings of the 23nd Nordic Conference on Computational Linguistics",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jenna Kanerva, Filip Ginter, Li-Hsin Chang, Iiro Ras- tas, Valtteri Skantsi, Jemina Kilpel\u00e4inen, Hanna- Mari Kupari, Jenna Saarni, Maija Sev\u00f3n, and Otto Tarkka. 2021. Finnish paraphrase corpus. In Pro- ceedings of the 23nd Nordic Conference on Compu- tational Linguistics (NoDaLiDa 2021).",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Turku neural parser pipeline: An end-to-end system for the conll 2018 shared task",
"authors": [
{
"first": "Jenna",
"middle": [],
"last": "Kanerva",
"suffix": ""
},
{
"first": "Filip",
"middle": [],
"last": "Ginter",
"suffix": ""
},
{
"first": "Niko",
"middle": [],
"last": "Miekka",
"suffix": ""
},
{
"first": "Akseli",
"middle": [],
"last": "Leino",
"suffix": ""
},
{
"first": "Tapio",
"middle": [],
"last": "Salakoski",
"suffix": ""
}
],
"year": 2018,
"venue": "Proceedings of the CoNLL 2018 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jenna Kanerva, Filip Ginter, Niko Miekka, Akseli Leino, and Tapio Salakoski. 2018. Turku neural parser pipeline: An end-to-end system for the conll 2018 shared task. In Proceedings of the CoNLL 2018 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies. Association for Computational Linguistics.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Universal Lemmatizer: A sequence to sequence model for lemmatizing Universal Dependencies treebanks",
"authors": [
{
"first": "Jenna",
"middle": [],
"last": "Kanerva",
"suffix": ""
},
{
"first": "Filip",
"middle": [],
"last": "Ginter",
"suffix": ""
},
{
"first": "Tapio",
"middle": [],
"last": "Salakoski",
"suffix": ""
}
],
"year": 2020,
"venue": "Natural Language Engineering",
"volume": "",
"issue": "",
"pages": "1--30",
"other_ids": {
"DOI": [
"10.1017/S1351324920000224"
]
},
"num": null,
"urls": [],
"raw_text": "Jenna Kanerva, Filip Ginter, and Tapio Salakoski. 2020. Universal Lemmatizer: A sequence to se- quence model for lemmatizing Universal Depen- dencies treebanks. Natural Language Engineering, pages 1-30.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Is it possible to create a very large wordnet in 100 days? An evaluation. Language Resources and Evaluation",
"authors": [
{
"first": "Krister",
"middle": [],
"last": "Lind\u00e9n",
"suffix": ""
},
{
"first": "Jyrki",
"middle": [],
"last": "Niemi",
"suffix": ""
}
],
"year": 2014,
"venue": "",
"volume": "48",
"issue": "",
"pages": "191--201",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Krister Lind\u00e9n and Jyrki Niemi. 2014. Is it possible to create a very large wordnet in 100 days? An evaluation. Language Resources and Evaluation, 48(2):191-201.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Efficient estimation of word representations in vector space",
"authors": [
{
"first": "Tomas",
"middle": [],
"last": "Mikolov",
"suffix": ""
},
{
"first": "Kai",
"middle": [],
"last": "Chen",
"suffix": ""
},
{
"first": "Greg",
"middle": [],
"last": "Corrado",
"suffix": ""
},
{
"first": "Jeffrey",
"middle": [],
"last": "Dean",
"suffix": ""
}
],
"year": 2013,
"venue": "Proceedings of Workshop at ICLR",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Tomas Mikolov, Kai Chen, Greg Corrado, and Jeffrey Dean. 2013. Efficient estimation of word represen- tations in vector space. In Proceedings of Workshop at ICLR.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Universal dependencies v1: A multilingual treebank collection",
"authors": [
{
"first": "Joakim",
"middle": [],
"last": "Nivre",
"suffix": ""
},
{
"first": "Marie-Catherine",
"middle": [],
"last": "De Marneffe",
"suffix": ""
},
{
"first": "Filip",
"middle": [],
"last": "Ginter",
"suffix": ""
},
{
"first": "Yoav",
"middle": [],
"last": "Goldberg",
"suffix": ""
},
{
"first": "Jan",
"middle": [],
"last": "Hajic",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Christopher",
"suffix": ""
},
{
"first": "Ryan",
"middle": [],
"last": "Manning",
"suffix": ""
},
{
"first": "Slav",
"middle": [],
"last": "Mcdonald",
"suffix": ""
},
{
"first": "Sampo",
"middle": [],
"last": "Petrov",
"suffix": ""
},
{
"first": "Natalia",
"middle": [],
"last": "Pyysalo",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Silveira",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16)",
"volume": "",
"issue": "",
"pages": "1659--1666",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Joakim Nivre, Marie-Catherine De Marneffe, Filip Ginter, Yoav Goldberg, Jan Hajic, Christopher D Manning, Ryan McDonald, Slav Petrov, Sampo Pyysalo, Natalia Silveira, et al. 2016. Universal dependencies v1: A multilingual treebank collec- tion. In Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16), pages 1659-1666.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Universal Dependency evaluation",
"authors": [
{
"first": "Joakim",
"middle": [],
"last": "Nivre",
"suffix": ""
},
{
"first": "Chiao-Ting",
"middle": [],
"last": "Fang",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the NoDaLiDa 2017 Workshop on Universal Dependencies",
"volume": "",
"issue": "",
"pages": "86--95",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Joakim Nivre and Chiao-Ting Fang. 2017. Univer- sal Dependency evaluation. In Proceedings of the NoDaLiDa 2017 Workshop on Universal Dependen- cies (UDW 2017), pages 86-95, Gothenburg, Swe- den.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "A thousand and one translations: Revisiting retranslation",
"authors": [
{
"first": "Outi",
"middle": [],
"last": "Paloposki",
"suffix": ""
},
{
"first": "Kaisa",
"middle": [],
"last": "Koskinen",
"suffix": ""
}
],
"year": 2001,
"venue": "Claims, Changes and Challenges in Translation Studies: Selected Contributions from the EST Congress",
"volume": "",
"issue": "",
"pages": "27--38",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Outi Paloposki and Kaisa Koskinen. 2004. A thousand and one translations: Revisiting retranslation. In Gyde Hansen, Kirsten Malmkjaer, and Daniel Gile, editors, Claims, Changes and Challenges in Trans- lation Studies: Selected Contributions from the EST Congress, Copenhagen 2001, pages 27 -38. John Benjamins Publishing Company.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Nonfinite constructions in Finnish children's literature: Features of translationese contradicting translation universals?",
"authors": [
{
"first": "",
"middle": [],
"last": "Tiina Puurtinen",
"suffix": ""
}
],
"year": 2003,
"venue": "Corpus-based Approaches to Contrastive Linguistics and Translation Studies",
"volume": "",
"issue": "",
"pages": "141--154",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Tiina Puurtinen. 2003. Nonfinite constructions in Finnish children's literature: Features of transla- tionese contradicting translation universals? In Syl- viane Granger, Jacques Lerot, and Stephanie Petch- Tyson, editors, Corpus-based Approaches to Con- trastive Linguistics and Translation Studies, pages 141 -154. Brill.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Software Framework for Topic Modelling with Large Corpora",
"authors": [
{
"first": "Petr",
"middle": [],
"last": "Radim\u0159eh\u016f\u0159ek",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Sojka",
"suffix": ""
}
],
"year": 2010,
"venue": "Proceedings of the LREC 2010 Workshop on New Challenges for NLP Frameworks",
"volume": "",
"issue": "",
"pages": "45--50",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Radim\u0158eh\u016f\u0159ek and Petr Sojka. 2010. Software Frame- work for Topic Modelling with Large Corpora. In Proceedings of the LREC 2010 Workshop on New Challenges for NLP Frameworks, pages 45-50.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "TaPaCo: A corpus of sentential paraphrases for 73 languages",
"authors": [
{
"first": "Yves",
"middle": [],
"last": "Scherrer",
"suffix": ""
}
],
"year": 2020,
"venue": "Proceedings of The 12th Language Resources and Evaluation Conference (LREC 2020)",
"volume": "",
"issue": "",
"pages": "6868--6873",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Yves Scherrer. 2020. TaPaCo: A corpus of sentential paraphrases for 73 languages. In Proceedings of The 12th Language Resources and Evaluation Confer- ence (LREC 2020), pages 6868-6873.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Filip Ginter, and Sampo Pyysalo. 2019. Multilingual is not enough: BERT for Finnish",
"authors": [
{
"first": "Antti",
"middle": [],
"last": "Virtanen",
"suffix": ""
},
{
"first": "Jenna",
"middle": [],
"last": "Kanerva",
"suffix": ""
},
{
"first": "Rami",
"middle": [],
"last": "Ilo",
"suffix": ""
},
{
"first": "Jouni",
"middle": [],
"last": "Luoma",
"suffix": ""
},
{
"first": "Juhani",
"middle": [],
"last": "Luotolahti",
"suffix": ""
},
{
"first": "Tapio",
"middle": [],
"last": "Salakoski",
"suffix": ""
}
],
"year": null,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1912.07076"
]
},
"num": null,
"urls": [],
"raw_text": "Antti Virtanen, Jenna Kanerva, Rami Ilo, Jouni Lu- oma, Juhani Luotolahti, Tapio Salakoski, Filip Gin- ter, and Sampo Pyysalo. 2019. Multilingual is not enough: BERT for Finnish. arXiv preprint arXiv:1912.07076.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "BERTScore: Evaluating text generation with BERT",
"authors": [
{
"first": "Tianyi",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "Varsha",
"middle": [],
"last": "Kishore",
"suffix": ""
},
{
"first": "Felix",
"middle": [],
"last": "Wu",
"suffix": ""
},
{
"first": "Kilian",
"middle": [
"Q"
],
"last": "Weinberger",
"suffix": ""
},
{
"first": "Yoav",
"middle": [],
"last": "Artzi",
"suffix": ""
}
],
"year": 2020,
"venue": "International Conference on Learning Representations",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q. Weinberger, and Yoav Artzi. 2020. BERTScore: Evaluating text generation with BERT. In Interna- tional Conference on Learning Representations.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "PAWS: Paraphrase adversaries from word scrambling",
"authors": [
{
"first": "Yuan",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "Jason",
"middle": [],
"last": "Baldridge",
"suffix": ""
},
{
"first": "Luheng",
"middle": [],
"last": "He",
"suffix": ""
}
],
"year": 2019,
"venue": "Vasta ammuttu Ammuttu hiljattain Olen pist\u00e4m\u00e4tt\u00f6m\u00e4n hygieeninen. Olen moitteettoman hygieeninen. Etk\u00f6 mennyt poliisin luo? Et mennyt poliisin puheille",
"volume": "1",
"issue": "",
"pages": "1298--1308",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Yuan Zhang, Jason Baldridge, and Luheng He. 2019. PAWS: Paraphrase adversaries from word scram- bling. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Tech- nologies, Volume 1 (Long and Short Papers), pages 1298-1308. Vasta ammuttu Ammuttu hiljattain Olen pist\u00e4m\u00e4tt\u00f6m\u00e4n hygieeninen. Olen moitteettoman hygieeninen. Etk\u00f6 mennyt poliisin luo? Et mennyt poliisin puheille?",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Wordphrase Anteeksi odotus",
"authors": [],
"year": null,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Word - phrase Anteeksi odotus.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "On niin paljon valinnanvaraa. On niin paljon mist\u00e4 valita. Useimmat teist\u00e4 tiet\u00e4v\u00e4t",
"authors": [
{
"first": "Ett\u00e4",
"middle": [],
"last": "Anteeksi",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Kesti",
"suffix": ""
}
],
"year": null,
"venue": "Suurin osa teist\u00e4 tiet\u00e4\u00e4",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Anteeksi, ett\u00e4 kesti. En edes osaa n\u00e4ytell\u00e4. En edes tied\u00e4 miten n\u00e4ytell\u00e4. On niin paljon valinnanvaraa. On niin paljon mist\u00e4 valita. Useimmat teist\u00e4 tiet\u00e4v\u00e4t [...] Suurin osa teist\u00e4 tiet\u00e4\u00e4, [...]",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "Phrasephrase Andrew ehti ensin. Andrew oli vain nopeampi",
"authors": [],
"year": null,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Phrase - phrase Andrew ehti ensin. Andrew oli vain nopeampi. I\u00e4n my\u00f6t\u00e4 [...] Mit\u00e4 vanhemmaksi tulin, sit\u00e4 [...]",
"links": null
},
"BIBREF21": {
"ref_id": "b21",
"title": "Miksi h\u00e4n tekee niin? Etk\u00f6 ole utelias? Etk\u00f6 halua tiet\u00e4\u00e4 miksi h\u00e4n tekee niin? kuuluuko seuralaisennekin tilin osakkaisiin? Kuuluuko tili my\u00f6s seuralaisellenne? Figurative Olen t\u00e4ysin hereill\u00e4",
"authors": [],
"year": null,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Miksi h\u00e4n tekee niin? Etk\u00f6 ole utelias? Etk\u00f6 halua tiet\u00e4\u00e4 miksi h\u00e4n tekee niin? kuuluuko seuralaisennekin tilin osakkaisiin? Kuuluuko tili my\u00f6s seuralaisellenne? Figurative Olen t\u00e4ysin hereill\u00e4, [...] Olen pirte\u00e4 kuin peipponen [...]",
"links": null
},
"BIBREF22": {
"ref_id": "b22",
"title": "Pid\u00e4 nyt vain p\u00e4\u00e4si kiinni. Teitk\u00f6 sen tasataksesi tilit? Teitk\u00f6 sen p\u00e4\u00e4st\u00e4ksesi tasoihin? Tied\u00e4 h\u00e4nt\u00e4",
"authors": [],
"year": null,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ole nyt vain hiljaa. Pid\u00e4 nyt vain p\u00e4\u00e4si kiinni. Teitk\u00f6 sen tasataksesi tilit? Teitk\u00f6 sen p\u00e4\u00e4st\u00e4ksesi tasoihin? Tied\u00e4 h\u00e4nt\u00e4. En min\u00e4 tied\u00e4.",
"links": null
},
"BIBREF23": {
"ref_id": "b23",
"title": "En tiennyt koko s\u00e4\u00e4nn\u00f6st\u00e4",
"authors": [
{
"first": "",
"middle": [],
"last": "Emph",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Jopa Runoja. Runojakin",
"suffix": ""
}
],
"year": null,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Emph. Jopa runoja. Runojakin. En tiennyt koko s\u00e4\u00e4nn\u00f6st\u00e4. En edes tiennyt s\u00e4\u00e4nn\u00f6st\u00e4.",
"links": null
},
"BIBREF24": {
"ref_id": "b24",
"title": "Mit\u00e4 t\u00e4\u00e4ll\u00e4 tapahtui? Mit\u00e4 ihmett\u00e4 t\u00e4\u00e4ll\u00e4 on tapahtunut?",
"authors": [],
"year": null,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Mit\u00e4 t\u00e4\u00e4ll\u00e4 tapahtui? Mit\u00e4 ihmett\u00e4 t\u00e4\u00e4ll\u00e4 on tapahtunut?",
"links": null
},
"BIBREF25": {
"ref_id": "b25",
"title": "kaiken aina selvemmin",
"authors": [],
"year": null,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": ".] n\u00e4en kaiken aina selvemmin.",
"links": null
},
"BIBREF26": {
"ref_id": "b26",
"title": "Se mit\u00e4 voin kertoa teille, on ett\u00e4",
"authors": [
{
"first": "/",
"middle": [],
"last": "Verbosity",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Redund",
"suffix": ""
}
],
"year": null,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Verbosity/ redund. Voin kertoa teille, ett\u00e4 [...] Se mit\u00e4 voin kertoa teille, on ett\u00e4 [...]",
"links": null
},
"BIBREF27": {
"ref_id": "b27",
"title": "on kuollut! Se ihminen jota etsit on kuollut! Mihin voin laittaa t\u00e4m\u00e4n? Pedille. Minne voin laskea t\u00e4m\u00e4n? Voit laittaa sen s\u00e4ngylle",
"authors": [
{
"first": "Ket\u00e4",
"middle": [],
"last": "Se",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Etsit",
"suffix": ""
}
],
"year": null,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Se, ket\u00e4 etsit, on kuollut! Se ihminen jota etsit on kuollut! Mihin voin laittaa t\u00e4m\u00e4n? Pedille. Minne voin laskea t\u00e4m\u00e4n? Voit laittaa sen s\u00e4ngylle.",
"links": null
},
"BIBREF28": {
"ref_id": "b28",
"title": "Hae ensiapupakkaus vessan kaapista. Hae ensiapupakkaus. Se on vessan kaapissa",
"authors": [],
"year": null,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Hae ensiapupakkaus vessan kaapista. Hae ensiapupakkaus. Se on vessan kaapissa.",
"links": null
},
"BIBREF29": {
"ref_id": "b29",
"title": "ne saattavat kiinnitt\u00e4\u00e4 liikaa huomiota",
"authors": [],
"year": null,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": ".] ett\u00e4 ne saattavat kiinnitt\u00e4\u00e4 liikaa huomiota.",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"text": "Distribution of the number of lemma indels for universal paraphrases labeled 4/4s including rewrites.",
"num": null,
"uris": null,
"type_str": "figure"
},
"TABREF1": {
"text": "Label distribution of paraphrases from the subset of alternative subtitle translations in Turku paraphrase corpus training set.",
"content": "