The CoNLL–SIGMORPHON 2018 Shared Task: Universal Morphological Reinflection

The CoNLL–SIGMORPHON 2018 Shared Task: Universal Morphological Reinflection

Abstract

The CoNLL–SIGMORPHON 2018 shared task on supervised learning of morphological generation featured data sets from 103 typologically diverse languages. Apart from extending the number of languages involved in earlier supervised tasks of generating inflected forms, this year the shared task also featured a new second task which asked participants to inflect words in sentential context, similar to a cloze task. This second task featured seven languages. Task 1 received 27 submissions and task 2 received 6 submissions. Both tasks featured a low, medium, and high data condition. Nearly all submissions featured a neural component and built on highly-ranked systems from the earlier 2017 shared task. In the inflection task (task 1), 41 of the 52 languages present in last year’s inflection task showed improvement by the best systems in the low-resource setting. The cloze task (task 2) proved to be difficult, and few submissions managed to consistently improve upon both a simple neural baseline system and a lemma-repeating baseline.

\aclfinalcopy

1 Introduction

Some of a word’s syntactic and semantic properties are expressed on the word form through a process termed morphological inflection. For example, each English count noun has both singular and plural forms (robot/robots, process/processes), known as the inflected forms of the noun. Some languages display little inflection, while others possess a proliferation of forms. A Polish verb can have nearly 100 inflected forms and an Archi verb has thousands Kibrik (1998).

Natural language processing systems must be able to analyze and generate these inflected forms. Fortunately, inflected forms tend to be systematically related to one another. This is why English speakers can usually predict the singular form from the plural and vice versa, even for words they have never seen before: given a novel noun wug, an English speaker knows that the plural is wugs.

We conducted a competition on generating inflected forms. This “shared task” consisted of two separate scenarios. In Task 1, participating systems must inflect word forms based on labeled examples. In English, an example of inflection is the conversion of a citation form1 run to its present participle, running. The system is provided with the source form and the morphosyntactic description (MSD) of the target form, and must generate the actual target form. Task 2 is a harder version of Task 1, where the system must infer the appropriate MSD from a sentential context. This is essentially a cloze task, asking participants to provide the correct form of a lemma in context.

2 Tasks and Evaluation

2.1 Task 1: Inflection

{adjustbox}

width= Lang Lemma Inflection Inflected form en hug V;PST hugged spark V;V.PTCP;PRS sparking es liberar V;IND;FUT;2;SG liberarás descomponer V;NEG;IMP;2;PL no descompongáis de aufbauen V;IND;PRS;2;SG baust auf Ärztin N;DAT;PL Ärztinnen

Table 1: Example training data from task 1. Each training example maps a lemma and inflection to an inflected form, The inflection is a bundle of morphosyntactic features. Note that inflected forms (and lemmata) can encompass multiple words. In the test data, the last column (the inflected form) must be predicted by the system.

The first task was identical to sub-task 1 from the CoNLL–SIGMORPHON 2017 shared task Cotterell et al. (2017), but the language selection was extended from 52 languages to 103. The data sets for the overlapping languages between 2017 and 2018 were also resampled and are not identical. The task consists of morphological generation with sparse training data, something that can be practically useful for MT and other downstream tasks in NLP. Here, participants were given examples of inflected forms as shown in Table 1. Each test example asked participants to produce some other inflected form when given a lemma and a bundle of morphosyntactic features as input.

The training data was sparse in the sense that it included only a few inflected forms from each lemma. That is, as in human L1 learning, the learner does not necessarily observe any complete paradigms in a language where the paradigms are large (e.g., dozens of inflected forms per lemma).2

Key points:

  1. The task is inflection: Given an input lemma and desired output tags, participants had to generate the correct output inflected form (a string).

  2. The supervised training data consisted of individual forms (see Table 1) that were sparsely sampled from a large number of paradigms.

  3. Forms that are empirically more frequent were more likely to appear in both training and test data (see § 3 for details).

  4. Systems were evaluated after training on (low), (medium), and (high) lemma/MSD/inflected form triplets.

2.2 Task 2: Inflection in Context

The cloze test is a common exercise in an L2 instruction setting. In the cloze test, a number of words are deleted from a text and students are required to fill in the gaps with contextually plausible forms, often working from the knowledge about which lemma should be inflected. The second task of the morphology shared task presents two variations of this traditional cloze test in two tracks specifically aimed at data-driven morphology learning.

Solving a cloze test well requires integration of many types of evidence beyond the pure capacity to inflect a word on demand. Since our training sets were gathered from actual textual resources, a good solver that accurately determines the most plausible form must implicitly combine knowledge of morphology, morphosyntax, semantics, and pragmatics. Potentially, even textual register and genre may affect the choice of correct form. Hence, the task is both intrinsically interesting from a linguistic point of view and carries potential to support many downstream NLP applications.

{adjustbox}

width=0.9

Figure 1: Test examples for tracks 1 and 2 in the cloze task. The objective is to inflect the target lemma dog in a contextually appropriate form, which in this case is dogs. Competitors observe context word forms, their lemmata and MSDs in track 1, whereas they only observe the context word forms in track 2.

As shown in Figure 1, both tracks supply the lemma of the omitted target word form and ask the competitors to inflect the lemma in a contextually appropriate way. In the first track, the competitors additionally see the lemmata and MSDs for all context words, whereas in the second track only the context words are available. In contrast to task 1, the MSD for the target lemma is never observed in either the first or the second track. This means that successful inflection requires the competitors to identify relevant contextual cues.

{adjustbox}

width=

Figure 2: Training examples for tracks 1 and 2 in the cloze task. Track 1 supplies a full morphosyntactically annotated corpus as training data, whereas track 2 only supplies lemmata for a number of selected training tokens. Remaining tokens lack annotation altogether.

As training data, the first track supplies a full morphosyntactically annotated corpus of sentences: every token is annotated with a lemma and MSD as shown in Figure 2. In the second track, the training data identifies a number of target tokens. Lemmata are supplied for these tokens but the remaining tokens receive no MSD annotation.

Similarly to task 1, both tracks in task 2 provide three different training data settings providing varying amounts of data: low (ca.  tokens), medium (ca.  tokens) and high (ca.  tokens). The token counts refer to the total number of tokens in the training sets. In the first track, this allows competitors to train their systems on all available tokens. In the second track, however, only a number of tokens supply the input lemma as explained above. Thus, the effective number of training examples is smaller in the second track than in the first track. In both tracks, competitors were restricted to using only the provided training sets. For example, semi-supervised training using external data was forbidden.

Key points:

  1. The task is inflection in context. Given an input lemma in sentential context, participants generate the correct inflected output form.

  2. Two degrees of supervision are provided. In track 1, participants see context word forms and their lemmata, as well as their MSDs. In track 2, participants only witness context word forms.

  3. The supervised training data, the development data, and the test data consist of sampled sentences from Universal Dependencies (UD) treebanks Nivre et al. (2017) together with UD-provided lemmata as well as MSDs, which were converted to the UniMorph format, in track 1.

3 Data

3.1 Data for Task 1

Languages

The data for the shared task was highly multilingual, comprising 103 unique languages. Of these, 52 were shared with the 2017 shared task Cotterell et al. (2017). As with all but 5 of the 2017 languages (Khaling, Kurmanji Kurdish, Sorani Kurdish, Haida, and Basque), the 34 remaining 2018 languages were sourced from the English edition of Wiktionary, a large multi-lingual crowd-sourced dictionary containing morphological paradigms for many lemmata.3

The shared task language set is genealogically diverse, including languages from 20 language stocks. Although the majority of the languages are Indo-European, we also include two language isolates (Haida and Basque) along with languages from Athabaskan (Navajo), Kartvelian (Georgian), Quechua, Semitic (Arabic, Hebrew), Sino-Tibetan (Khaling), Turkic (Turkish), and Uralic (Estonian, Finnish, Hungarian, and Northern Sami) language families. The shared task language set is also diverse in terms of morphological structure, with languages which use primarily prefixes (Navajo), suffixes (Quechua and Turkish), and a mix, with Spanish exhibiting internal vowel variations along with suffixes and Georgian using both infixes and suffixes. The language set also exhibits features such as templatic morphology (Arabic, Hebrew), vowel harmony (Turkish, Finnish, Hungarian), and consonant harmony (Navajo) which require systems to learn non-local alternations. Finally, the resource level of the languages in the shared task set varies greatly, from major world languages (e.g. Arabic, English, French, Spanish, Russian) to languages with few speakers (e.g. Haida, Khaling). Typologically, the majority of the languages are agglutinating or fusional, with three polysynthetic languages; Haida, Greenlandic, and Navajo.4

Data Format

For each language, the basic data consists of triples of the form (lemma, feature bundle, inflected form), as in Table 1. The first feature in the bundle always specifies the core part of speech (e.g., verb).

All features in the bundle are coded according to the UniMorph Schema, a cross-linguistically consistent universal morphological feature set Sylak-Glassman et al. (2015a, b).

Extraction from Wiktionary

For each of the Wiktionary languages, Wiktionary provides a number of tables, each of which specifies the full inflectional paradigm for a particular lemma. These tables were extracted using a template annotation procedure described in Kirov et al. (2018).

Within a language, different paradigms may have different shapes. To prepare the shared task data, each language’s parsed tables from Wiktionary were grouped according to their tabular structure and number of cells. Each group represents a different type of paradigm (e.g., verb). We used only groups with a large number of lemmata, relative to the number of lemmata available for the language as a whole. For each group, we associated a feature bundle with each cell position in the table, by manually replacing the prose labels describing grammatical features (e.g.  “accusative case”) with UniMorph features (e.g. ACC). This allowed us to extract triples as described in the previous section. The dataset produced by this process was sampled to create appropriately-sized data for the shared task, as described in § 3.1.5 The dataset sizes by language are given in Table 2 and Table 3.

Sampling the Train-Dev-Test Splits.

From each language’s collection of paradigms, we sampled the training, development, and test sets as follows.6

Our first step was to construct probability distributions over the (lemma, feature bundle, inflected form) triples in our full dataset. For each triple, we counted how many tokens the inflected form has in the February 2017 dump of Wikipedia for that language. To distribute the counts of an observed form over all the triples that have this token as its form, we use the syncretism resolution method of Cotterell et al. (2018), training a neural network on unambiguous forms to estimate the distribution over all, even ambiguous, forms. We then sampled 12,000 triples without replacement from this distribution. The first 100 were taken as the low-resource training set for sub-task 1, the first 1,000 as the medium-resource training set, and the first 10,000 as the high-resource training set. Note that these training sets are nested, and that the highest-count triples tend to appear in the smaller training sets.

The final 2,000 triples were randomly shuffled and then split in half to obtain development and test sets of 1,000 forms each. The final shuffling was performed to ensure that the development set is similar to the test set. By contrast, the development and test sets tend to contain lower-count triples than the training set.7 Note that for languages that do not have enough triples for this process, we settle for omitting the higher-resource training regimes and scale down the other sizes. Details for all languages are found in Tables 3 and 2.

{adjustbox}

width=.9 Language Family Lemmata / Forms High Medium Low Dev Test Adyghe Caucasian 1666 / 20475 1664/10000 760/1000 99/100 763/1000 749/1000 Albanian Indo-European 589 / 33483 588/10000 375/1000 84/100 377/1000 373/1000 Arabic Semitic 4134 / 140003 3204/10000 832/1000 99/100 807/1000 813/1000 Armenian Indo-European 7033 / 338750 4658/10000 903/1000 98/100 880/1000 900/1000 Asturian Romance 436 / 29797 432/10000 361/1000 90/100 368/1000 365/1000 Azeri Iranian 340 / 8004 340/6488 290/1000 79/100 73/100 81/100 Bashkir Turkic 1084 / 12168 1084/10000 662/1000 94/100 657/1000 651/1000 Basque Isolate 45 / 12663 45/10000 42/1000 24/100 41/1000 43/1000 Belarusian Slavic 1027 / 16113 1027/10000 616/1000 98/100 628/1000 630/1000 Bengali Indo-Aryan 136 / 4443 136/4243 134/1000 65/100 65/100 68/100 Breton Celtic 44 / 2294 44/1983 44/1000 40/100 38/100 39/100 Bulgarian Slavic 2468 / 55730 2133/10000 716/1000 98/100 742/1000 744/1000 Catalan Romance 1547 / 81576 1545/10000 746/1000 95/100 738/1000 738/1000 Classical-Syriac Semitic 160 / 3652 160/2396 160/1000 74/100 70/100 73/100 Cornish Celtic 9 / 469 9/346 9/100 9/50 9/50 Crimean-Tatar Turkic 1230 / 7514 1230/7314 704/1000 94/100 95/100 95/100 Czech Slavic 5125 / 134527 3908/10000 848/1000 97/100 848/1000 849/1000 Danish Germanic 3193 / 25508 3137/10000 877/1000 100/100 866/1000 853/1000 Dutch Germanic 4993 / 55467 4161/10000 913/1000 100/100 898/1000 894/1000 English Germanic 22765 / 120004 8367/10000 989/1000 100/100 985/1000 984/1000 Estonian Uralic 886 / 38215 886/10000 587/1000 94/100 553/1000 577/1000 Faroese Germanic 3077 / 45474 2959/10000 857/1000 99/100 852/1000 865/1000 Finnish Uralic 57642 / 2490377 8643/10000 985/1000 100/100 983/1000 987/1000 French Romance 7535 / 367732 5592/10000 936/1000 98/100 948/1000 941/1000 Friulian Romance 168 / 8071 168/7871 168/1000 76/100 79/100 79/100 Galician Romance 486 / 36801 486/10000 421/1000 91/100 421/1000 423/1000 Georgian Kartvelian 3782 / 74412 3537/10000 861/1000 100/100 872/1000 874/1000 German Germanic 15060 / 179339 6797/10000 961/1000 100/100 945/1000 962/1000 Greek Hellenic 10581 / 186663 5130/10000 897/1000 98/100 915/1000 908/1000 Greenlandic Inuit 23 / 368 23/268 23/100 21/50 21/50 Haida Isolate 41 / 7040 41/6840 41/1000 40/100 34/100 38/100 Hebrew Semitic 510 / 13818 510/10000 470/1000 95/100 431/1000 453/1000 Hindi Indo-Aryan 258 / 54438 258/10000 252/1000 85/100 254/1000 255/1000 Hungarian Uralic 13989 / 503042 7123/10000 963/1000 100/100 973/1000 978/1000 Icelandic Germanic 4775 / 76945 4115/10000 894/1000 100/100 898/1000 906/1000 Ingrian Uralic 50 / 1099 50/999 45/100 30/50 31/50 Irish Celtic 7464 / 107298 5040/10000 906/1000 99/100 913/1000 893/1000 Italian Romance 10009 / 509574 6389/10000 948/1000 100/100 942/1000 944/1000 Kabardian Caucasian 250 / 3092 250/2892 246/1000 81/100 82/100 81/100 Kannada Dravidian 159 / 6402 159/4383 147/1000 54/100 53/100 59/100 Karelian Uralic 20 / 682 20/582 20/100 17/50 18/50 Kashubian Slavic 37 / 509 37/402 34/100 27/50 28/50 Kazakh Turkic 26 / 357 26/257 26/100 22/50 25/50 Khakas Turkic 75 / 1200 52/732 44/100 31/50 32/50 Khaling Sino-TIbetan 591 / 156097 584/10000 426/1000 92/100 411/1000 422/1000 Kurmanji Iranian 15083 / 216370 7046/10000 945/1000 100/100 949/1000 958/1000 Ladin Romance 180 / 7656 180/7456 179/1000 80/100 81/100 75/100 Latin Romance 17214 / 509182 6517/10000 943/1000 100/100 939/1000 945/1000

Table 2: Total number of lemmata and forms available for sampling, and number of distinct lemmata and forms present in each data condition in Task 1. Data permitting, there were 10000,1000, and 100 forms in the High, Medium, and Low conditions, respectively, and 1000 forms in each Dev and Test set.
{adjustbox}

width=.9 Language Family Lemmata / Forms High Medium Low Dev Test Latvian Baltic 7548 / 136998 5268/10000 930/1000 99/100 922/1000 923/1000 Lithuanian Baltic 1458 / 34130 1443/10000 632/1000 96/100 664/1000 639/1000 Livonian Uralic 203 / 3987 203/3787 203/1000 71/100 70/100 70/100 Lower-Sorbian Slavic 994 / 20121 993/10000 616/1000 96/100 621/1000 631/1000 Macedonian Slavic 10313 / 168057 6107/10000 951/1000 99/100 943/1000 956/1000 Maltese Semitic 112 / 3584 112/1560 112/1000 68/100 71/100 69/100 Mapudungun Araucanian 26 / 783 26/602 26/100 22/50 23/50 Middle-French Romance 603 / 36970 603/10000 480/1000 92/100 491/1000 505/1000 Middle-High-German Germanic 29 / 708 29/594 27/100 19/50 22/50 Middle-Low-German Germanic 52 / 1513 52/988 43/100 30/50 34/50 Murrinhpatha Australian 29 / 1110 29/973 28/100 24/50 24/50 Navajo Athabaskan 674 / 12354 674/10000 489/1000 92/100 491/1000 494/1000 Neapolitan Romance 40 / 1808 40/1568 40/1000 36/100 38/100 37/100 Norman Romance 5 / 280 5/180 5/100 5/50 5/50 North-Frisian Germanic 51 / 3204 51/2256 51/1000 42/100 43/100 44/100 Northern-Sami Uralic 2103 / 62677 1977/10000 750/1000 97/100 717/1000 730/1000 Norwegian-Bokmaal Germanic 5527 / 19238 5041/10000 925/1000 100/100 928/1000 930/1000 Norwegian-Nynorsk Germanic 4689 / 16563 4420/10000 922/1000 99/100 903/1000 912/1000 Occitan Romance 174 / 8316 174/8116 173/1000 76/100 81/100 75/100 Old-Armenian Indo-European 4300 / 93085 3413/10000 837/1000 100/100 802/1000 822/1000 Old-Church-Slavonic Slavic 152 / 4148 152/2961 151/1000 78/100 70/100 76/100 Old-English Germanic 1867 / 42425 1795/10000 688/1000 96/100 708/1000 701/1000 Old-French Romance 1700 / 123374 1666/10000 745/1000 96/100 769/1000 722/1000 Old-Irish Celtic 49 / 1078 49/851 38/100 27/50 26/50 Old-Saxon Germanic 863 / 22287 861/10000 514/1000 85/100 535/1000 494/1000 Pashto Iranian 395 / 6945 395/6340 289/1000 82/100 77/100 78/100 Persian Iranian 273 / 37128 273/10000 269/1000 82/100 268/1000 267/1000 Polish Slavic 10185 / 201024 5922/10000 935/1000 99/100 938/1000 942/1000 Portuguese Romance 4001 / 305961 3657/10000 905/1000 98/100 868/1000 865/1000 Quechua Quechuan 1006 / 180004 957/10000 515/1000 91/100 492/1000 506/1000 Romanian Romance 4405 / 80266 3351/10000 858/1000 99/100 854/1000 828/1000 Russian Slavic 28068 / 473481 8241/10000 973/1000 100/100 985/1000 977/1000 Sanskrit Indo-Aryan 917 / 33847 917/10000 548/1000 91/100 585/1000 558/1000 Scottish-Gaelic Celtic 73 / 781 73/681 57/100 36/50 39/50 Serbo-Croatian Slavic 24419 / 840799 6726/10000 963/1000 99/100 965/1000 945/1000 Slovak Slavic 1046 / 14796 1046/10000 625/1000 95/100 590/1000 633/1000 Slovene Slavic 2535 / 60110 2368/10000 757/1000 99/100 760/1000 793/1000 Sorani Iranian 274 / 22990 263/10000 197/1000 74/100 198/1000 199/1000 Spanish Romance 5460 / 383390 4621/10000 906/1000 99/100 902/1000 922/1000 Swahili Bantu 100 / 10092 100/8800 88/1000 49/100 50/100 42/100 Swedish Germanic 10552 / 78407 6508/10000 952/1000 100/100 954/1000 970/1000 Tatar Turkic 1283 / 7832 1283/7632 736/1000 98/100 95/100 95/100 Telugu Dravidian 127 / 1548 18/61 16/50 16/50 Tibetan Sino-Tibetan 65 / 353 63/158 56/100 38/50 38/50 Turkish Turkic 3579 / 275460 2876/10000 821/1000 98/100 849/1000 840/1000 Turkmen Turkic 68 / 810 68/710 51/100 35/50 35/50 Ukrainian Slavic 1493 / 20904 1491/10000 722/1000 99/100 745/1000 736/1000 Urdu Indo-Aryan 182 / 12572 182/10000 113/1000 53/100 105/1000 107/1000 Uzbek Turkic 15 / 1260 15/1060 15/1000 15/100 15/100 15/100 Venetian Romance 368 / 18227 368/10000 339/1000 88/100 341/1000 340/1000 Votic Uralic 55 / 1430 55/1230 55/1000 50/100 48/100 47/100 Welsh Celtic 183 / 10641 183/10000 181/1000 78/100 76/100 80/100 West-Frisian Germanic 85 / 1429 85/1078 85/1000 53/100 62/100 61/100 Yiddish Germanic 803 / 7986 803/7356 581/1000 96/100 90/100 93/100 Zulu Bantu 566 / 39607 566/10000 450/1000 90/100 449/1000 443/1000

Table 3: Total number of lemmata and forms available for sampling, and number of distinct lemmata and forms present in each data condition in Task 1. Data permitting, there were 10000,1000, and 100 forms in the High, Medium, and Low conditions, respectively, and 1000 forms in each Dev and Test set.

3.2 Data for Task 2

All task 2 data sets are based on Universal Dependencies (UD) v2 treebanks Nivre et al. (2017). We used the data sets aimed for the 2017 CoNLL shared task on Multilingual Dependency Parsing Zeman et al. (2017) because those were available before the official UD v2 data sets.8 For contextual inflection data sets, we retained only word forms, lemmata, part-of-speech tags and morphosyntactic feature descriptions. Dependency trees were discarded along with all other annotations present in the treebanks.

Task 2 submissions are evaluated with regard to two distinct criteria: (1) the ability of the system to reconstruct the original word form in the UD test set and (2) the ability of the system to find a contextually plausible form even if the form differs from the original one. Evaluation on plausible forms is based on manually identifying the set of contextually plausible forms for each test example. Because of the need for manual annotation, task 2 covers a more limited set of languages than task 1. In total, there are seven languages: English, Finnish, French, German, Russian, Spanish and Swedish. Token counts for the training, development and test sets are given in Table 4.

{adjustbox}

width= Language Train Dev Test low medium high English 1,009 10,016 100,031 22,509 22,765 Finnish 1,001 10,009 100,003 16,543 15,452 French 1,016 10,004 100,001 28,304 14,992 German 1,005 10,001 079,439 03,752 22,903 Russian 1,003 10,020 075,964 11,292 27,935 Spanish 1,017 10,035 100,000 35,209 27,807 Swedish 1,007 10,009 066,645 07,999 20,808

Table 4: Token counts of the training, development and test sets for task 2.
Language Dev Test
English 2,489 993
Finnish 1,881 787
French 1,655 491
German 0333 989
Russian 1,181 996
Spanish 2,268 713
Swedish 0573 940
Table 5: Counts of target lemmata to be inflected in the development and test sets for task 2.

Data Conversion

Some of the UD treebanks required slight modifications in order to be suitable for reinflection. In the Finnish data sets, lemmata for compound words included morpheme boundaries, for example muisti#kapasiteetti ‘memory capacity’. The morpheme boundary symbols were deleted. In the Russian treebanks, all lemmata were written completely in upper case letters. These were converted to lower case.9

Manual annotation

To produce the complete list of “plausible forms” annotators were given complete UniMorph inflection tables for the center lemma for each sentence and were asked to check off all forms that are “grammatically plausible” in the particular context. For example, given an original sentence We saw the dog, the form dogs would be contextually plausible and would be annotated into the test set. For pro-drop languages and short sentences, it is sometimes the case that all or most indicative, conditional, and future forms of a verb are acceptable when the subject is omitted and agreement is unknown. For example, consider the Spanish sentence from the test data:


{adjustbox}

width= la mejor de Primera ser to be ‘the’ ‘best’ ‘of’ ‘premier (league)’


Obviously, almost any person, tense, and aspect of the verb ‘to be’ will be appropriate for this limited context (sería ‘I would be’, fue ‘he/she/it was’, eres ‘you are’, …). Of course, depending on the genre of the text, some would be highly implausible, but the annotation intends to capture morphosyntactic rather than semantic and pragmatic felicity.

We had one annotator for each test set, with the exception of French, in which, due to practical difficulties in finding a native speaker annotator, we did not annotate the plausible forms and instead used the original sentences.

When forming the final test sets, all test examples with more than contextually plausible word form alternatives were filtered out. This was done because a large number of plausible word forms was deemed to raise the risk of annotation errors. A threshold of plausible forms was chosen because it means that all languages have test sets greater than examples. The test set for French is smaller but this is not due to manual annotations.

{adjustbox}

width=0.85

Figure 3: A morphosyntactically annotated sentence from the original UD treebank for English and the result of an automatic conversion into the UniMorph annotation schema.

Sampling examples

The data sets for each language are based on UD treebanks for the given language. We preserved UD splits into training, development and test data.

For each UD treebank, we first formed sets of training, development and test candidate sentences. A sentence was a candidate for the shared task data set if it contained a token found in the UniMorph resource for the relevant language; or more precisely, a token whose word form, lemma and MSD occur in a same UniMorph inflection table.

We limited target tokens to tokens present in the UniMorph resource in order to facilitate manual annotation of data sets. In particular, we limited the set of possible target MSDs to MSDs which occur in the Unimorph resource. This was necessary to avoid a prohibitively large number of contextually plausible inflections in certain languages. For example, Finnish includes a number of clitics (ko/kä, kin, han/hän, pa/pä, s, kaan/kään) which can be appended relatively freely to word forms. Combinations of clitics are also possible. This easily leads to hundreds of word forms which can be contextually plausible. Restricting the MSDs of a possible output form to the more limited set of MSDs occurring in the UniMorph resource made the selection of plausible forms far more manageable from an annotation perspective.

Training data sets were formed from candidate sentences simply by sampling a suitable number of sentences from the candidate sets in order to achieve the desired token counts , , and for the low, medium, and high data settings, respectively. For German and Russian, all candidate sentences were used in the high data setting, although this was not sufficient to create a training set of tokens. The training sets for German and Russian are, therefore, smaller than those for the other languages. For the development sets, we used all available candidate sentences for all of the languages.

For the test data, we first formed a set of candidate sentences so that the combined number of target tokens in the test sets was 1,000.10 Target tokens in these initial test sets were then manually annotated with additional contextually plausible word forms.

MSD conversion

Sampling of training, development and test examples was based on comparing UD word forms, lemmata and MSDs to equivalents in UniMorph paradigms. Therefore, it was necessary to convert the morphosyntactic annotation in the UD data sets into UniMorph morphosyntactic annotation. We used deterministic tag conversion rules to accomplish this. An example of a source UD sentence and a target UniMorph sentence is shown in Figure 3.

Since the selection of languages in task 2 is small and we do not attempt to correct annotation errors in the UD source materials, conversion between UD and UniMorph morphosyntactic descriptions is generally straightforward.11 However, UD descriptions are more fine-grained than their UniMorph equivalents. For example, UD denotes lexical features such as noun gender which are inherent features of a lexeme possessed by all of its word forms. Such inherent features are missing from UniMorph which exclusively annotates inflectional morphology McCarthy et al. (2018). Therefore, UD features which lack correspondents in the UniMorph tagging schema were simply dropped during conversion.

4 Baselines

4.1 Task 1 Baseline

The baseline system provided for task 1 was based on the observation that, for a large number of languages, producing an inflected form from an input citation form can often be done by memorizing the suffix changes that occur in doing so, assuming enough examples are seen Liu and Mao (2016). For example, in witnessing a Finnish inflection of the noun koti ‘home’ in the singular elative case as kodista, a number of transformation rules can be extracted that may apply to previously unseen nouns:


    $koti$
    $kodista$  N;IN+ABL;SG

In this example, the following transformation rules are extracted:


$ sta$ i$ ista$
ti$ dista$ oti$ odista$
koti$ kodista$

Such rules are then extracted from each example inflection in the training data. At generation time, the longest matching left hand side of a rule is identified and applied to the citation form. For example, if the Finnish noun luoti ‘bullet’ were to be inflected in the elative (N;IN+ABL;SG) using only the extracted rules given above, the transformation oti$ odista$ would be triggered, producing the output luodista. In case there are multiple candidate rules of equally long left hand sides that all match, ties are broken by frequency—i.e. the rule that has been witnessed most times in the training data applies.

Since languages may also use prefixing as a inflectional strategy, a similar process is applied to any identified prefix changes. Identifying which parts of a change in a word form correspond to a prefix and which are considered suffixes requires alignment of the citation form and the output form, which is performed as a preliminary step. We refer the reader to \newcitecotterell-conll-sigmorphon2017 for a detailed description of the baseline system.

4.2 Task 2 Baseline

Neural Baseline

The neural baseline system is an encoder-decoder reinflection system with attention inspired by \newcitekann-schutze:2016:P16-2. The crucial difference is that the reinflection is conditioned on sentence context. This is accomplished by conditioning the encoder on embeddings of context words in track 2 and context words, their lemmata and their MSDs in track 1.

{adjustbox}

width=0.75

Figure 4: The neural baseline system for track 2 of task 2: A bidirectional LSTM encoder, conditioned on embeddings of the left context word The, right context word are and a whole token embedding of the lemma dog, is used to encode the character sequence (d, o, g) into representation vectors , and . An LSTM decoder with an attention mechanism generates the contextually appropriate output word form dogs. The neural baseline system for track 1 is very similar but the encoder is conditioned on embeddings of the context words, context lemmata and context MSDs.

The neural baseline system takes as input

  1. A lemma ,

  2. a left and right context word form and , respectively.

  3. a left and right context lemma and , respectively (only in track 1) and

  4. a left and right context MSD and , respectively (only in track 1).

The neural baseline system produces an inflected form of the lemma as output.

The input characters are first embedded: . Then, context words ( and ) for both tracks, as well as context lemmata ( and ) and MSDs ( and ) for track 1 are also embedded: , and . The system also a uses the whole token embedding of the input lemma : .

A bidirectional LSTM encoder is used to encode the lemma into representation vectors. In order to condition the encoder on the sentence context of the lemma, the encoder input vector for character is

  1. a concatenation of embeddings for the context word forms, context lemmata, context MSDs, input lemma and input character: for track 1, and

  2. a concatenation of embeddings for the context word forms, input lemma and input character: for track 2.

The input vectors are then encoded into representations by a bidirectional LSTM encoder. Finally, a decoder with additive attention Vaswani et al. (2017) is used for generating the output word form based on the representations .

The baseline system uses 100-dimensional embeddings and the LSTM hidden dimension for both the encoder and decoder is of size 100. Both encoder and decoder LSTM networks are single layer networks. The additive attention network is a 2-layer feed-forward network with hidden dimension 100 and nonlinearity.

The baseline system is trained for 20 epochs in both tracks and under all data settings using Adam Kingma and Ba (2014). During training, 30% dropout is applied on all input and recurrent connections in the encoder and decoder LSTM networks. Whole token embeddings for the input lemma, context word forms, lemmata and MSDs are dropped with a probability of 10%.

Copy Baseline

The second baseline is very straightforward. It simply copies the input lemma into the output. The system is based on the observation that in many languages the lemma form is quite common. In some languages, such as English, this baseline is in fact quite difficult to beat when the training set is small.

{adjustbox}

width= predict MSD subword context context RNN context attention multilingual beam search BME-HAS Ács (2018) COPENHAGEN Kementchedjhieva et al. (2018) CUBoulder Liu et al. (2018) NYU Kann et al. (2018) UZH Makarov and Clematide (2018)

Table 6: Features of Task 2 systems.

5 Results

The CoNLL–SIGMORPHON 2018 shared task received submissions from 15 teams with members from 17 universities or institutes (Table 7). Many of the teams submitted more than one system, yielding a total of 33 unique systems entered—27 for task 1, and 6 for task 2. In addition, baseline systems provided by the organizers for both tasks were also evaluated.

{adjustbox}

width=2 Team Institute(s) System Description Paper AXSEMANTICS AX Semantics \newcitemadsack-EtAl:2018:K18-20 BME/BME-HAS Budapest University of Technology and Economics / Hungarian Academy of Sciences \newciteacs:2018:K18-20 COPENHAGEN University of Copenhagen \newcitekementchedjhieva-bjerva-augenstein:2018:K18-20 CUBoulder University of Colorado, Boulder \newciteliu-EtAl:2018:K18-20 HAMBURG Universität Hamburg \newciteschroder-EtAl:2018:K18-20 IITBHU IIT (BHU) Varanasi / IIIT Hyderabad \newcitesharma-katrapati-sharma:2018:K18-20 IIT-VARANASI Indian Institute of Technology (BHU) Varanasi \newcitejain-singh:2018:K18-20 KUCST University of Copenhagen, Centre for Language Technology \newciteagirrezabal:2018:K18-20 MSU Moscow State University \newcitesorokin:2018:K18-20 NYU New York University \newcitekann-lauly-cho:2018:K18-20 RACAI Romanian Academy \newcitedumitrescu-borocbs:2018:K18-20 TUEBINGEN-OSLO University of Oslo / University of Tübingen \newciterama-ccoltekin:2018:K18-20 UA University of Alberta \newcitenajafi-EtAl:2018:K18-20 UZH University of Zurich \newcitemakarov-clematide:2018:K18-20 WASEDA Waseda University \newcitefam-lepage:2018:K18-20

Table 7: Participating teams, member institutes, and the corresponding system description papers. In the results and the main text, team submissions have an additional integer index to distinguish between multiple submissions by one team. The numbers at each abbreviated team name show whether teams participated in task 1, task 2, or both.

5.1 Task 1 Results

{adjustbox}

width=.95 High Medium Low uzh-01 96.00 / 0.08 86.64 / 0.26 57.18 / 1.00 uzh-02 95.97 / 0.08 86.38 / 0.27 57.21 / 1.02 bme-02 94.66 / 0.11 67.26 / 0.88 2.43 / 6.91 iitbhu-iiith-01 94.43 / 0.11 82.90 / 0.34 49.79 / 1.18 iitbhu-iiith-02 94.43 / 0.11 84.19 / 0.32 52.60 / 1.10 bme-03 93.97 / 0.12 67.36 / 0.75 3.63 / 6.75 bme-01 93.88 / 0.12 67.43 / 0.75 3.74 / 6.72 msu-04 91.87 / 0.23 76.40 / 0.55 31.40 / 2.16 iit-varanasi-01 91.73 / 0.16 70.17 / 0.66 23.33 / 2.40 waseda-01 91.12 / 0.19 77.38 / 0.67 44.09 / 1.68 msu-03 90.52 / 0.25 75.74 / 0.55 25.86 / 2.38 axsemantics-01 84.19 / 0.40 58.00 / 1.10 72.00 / 0.96 msu-02 82.68 / 0.41 69.45 / 0.79 41.61 / 1.86 racai-01 79.93 / 0.43   — / —   — / — hamburg-01 77.53 / 0.44 74.03 / 0.54 40.28 / 1.45 axsemantics-02 74.77 / 0.68 60.00 / 1.03 14.89 / 3.89 msu-01 74.33 / 0.78 64.57 / 0.93   — / — tuebingen-oslo-03 63.05 / 1.15 30.98 / 2.25 1.39 / 5.70 tuebingen-oslo-02 56.60 / 1.34 29.72 / 2.36 4.43 / 5.06 kucst-01 54.37 / 1.57 32.28 / 2.23 2.79 / 5.28 tuebingen-oslo-01 49.52 / 1.67 20.97 / 2.81 0.00 / 7.94 ua-08   — / —   — / — 53.22 / 1.35 ua-05   — / —   — / — 50.53 / 1.34 ua-06   — / —   — / — 49.73 / 1.46 ua-03   — / —   — / — 44.82 / 1.45 ua-02   — / —   — / — 41.61 / 2.47 ua-07   — / —   — / — 39.52 / 1.76 ua-01   — / —   — / — 38.22 / 2.02 ua-04   — / —   — / — 21.25 / 3.43 baseline 77.42 / 0.51 63.53 / 0.90 38.89 / 1.88 oracle-fc 99.87 /     98.27 /     77.23 /     oracle-e 98.90 /     93.74 /     74.88 /    

Table 8: Task 1 results: Per-form accuracy (in percentage points) and average Levenshtein distance from the correct form (in characters), averaged across the 103 languages with all languages weighted equally. The columns represent the different training size conditions. Rows are sorted by accuracy under the “High” condition. Numbers in bold are the best accuracy in their category. Greyed-out cells represent partial submissions that did not provide output for every language, and thus do not have comparable mean scores. The per-language performance of these systems can be found in the Appendix.

The relative system performance is described in Table 8, which show the average per-language accuracy of each system by resource condition. The table reflects the fact that some teams submitted more than one system (e.g. UZH-1 & UZH-2 in the table). Learning curves for each language across conditions are shown in Tables 10 and 9, which indicates the best per-form accuracy achieved by a submitted system. Full results can be found in Appendix A. Newer approaches led to better overall results in 2018 compared to 2017. In the low-resource condition, 41 (80%) of the 52 languages shared across years saw improvement in top system performance.

In the lower data conditions, encoder-decoder models are known to perform worse than the baseline model due to data sparsity. One way to work around this weakness is to learn sequences of edit operations instead of a standard string-to-string transduction, a strategy which was used by teams last year and this year (AX SEMANTICS, UZH, HAMBURG, MSU, RACAI). Another strategy is to create artificial training data that biases the neural model toward copying Kann and Schütze (2017); Bergmanis et al. (2017); Silfverberg et al. (2017); Zhou and Neubig (2017); Nicolai et al. (2017), which was also employed this year (TUEBINGEN-OSLO, WASEDA). Learning edit sequences requires input/output alignment, often as a preliminary step. The UZH submissions, which attained the highest average accuracy on the higher data conditions, built upon ideas in their last year’s submission Makarov et al. (2017), which had used such a separate alignment step followed by the application of an edit sequence. Their 2018 submission included edit distance alignment as part of the training loss function in the model, producing an end-to-end model. Another alternative to the edit sequence model is to use pointer generator networks, introduced by See et al. (2017) for text summarization, which also allow for copying parts of the input. This was employed by IITBHU. BME used a modified attention model that attended to both the lemma sequence and the tag sequence, which worked well in the high data condition, but, being without models of data augmentation or edit sequences, it suffered in the low data setting. In general, systems that included edit sequence generation or data augmentation fared significantly better in the low data settings. The HAMBURG submission attempted to learn similarities between characters based on rendering them visually using a font, with the intent to discover similarities such as those between a and ä, where the former is usually a low back vowel, and the latter a fronted version. Ensembling was also a popular choice to improve system performance. The UA system combined multiple models, both neural and non-neural, and focused on performance in the low data setting.

Even though the top-ranked systems used some form of ensembling to improve performance, different teams relied on different overall approaches. As a result, submissions may contain some amount of complementary information, so that a global ensemble may improve accuracy. As in 2017, we present an upper bound on the possible performance of such an ensemble. Table 8 includes an “Ensemble Oracle” system (oracle-e) that gives the correct answer if any of the submitted systems is correct. The oracle performs significantly better than any one system in both the Medium (10%) and Low (25%) conditions. This suggests that the different strategies used by teams to “bias” their systems in an effort to make up for sparse data lead to substantially different generalization patterns.

As in 2017, we also present a second “Feature Combination” Oracle (oracle-fc) that gives the correct answer for a given test triple iff its feature bundle appeared in training (with any lemma). Thus, oracle-fc provides an upper bound on the performance of systems that treat a feature bundle such as V;SBJV;FUT;3;PL as atomic. In the low-data condition, this upper bound was 77%, meaning that 23% of the test bundles had never been seen in training data. Nonetheless, systems should be able to make some accurate predictions on this 23% by decomposing each test bundle into individual morphological features such as FUT (future) and PL (plural), and generalizing from training examples that involve those features. For example, a particular feature or sub-bundle might be realized as a particular affix. For systems to succeed at this type of generalization, they must treat each individual feature separately, rather than treating feature bundles as holistic. In the medium data condition for some languages, some submissions far surpassed oracle-fc. As in 2017, the most notable example of this is Basque, where oracle-fc produced a 44% accuracy while six of the submitted systems produced an accuracy of 80% or above. Basque is an extreme example with very large paradigms for the few verbs that inflect in the language, so the problem of generalizing correctly to unseen feature combinations is amplified.

{adjustbox}

width=0.9 Task 1 - Part 1 High Medium Low Adyghe 100.00(uzh-2) 94.40(uzh-1) 90.60(ua-8) Albanian 98.90(bme-2) 88.80(iitbhu-iiith-2) 36.40(uzh-1) Arabic 93.70(uzh-1) 79.40(uzh-1) 45.20(uzh-1) Armenian 96.90(bme-2) 92.80(uzh-1) 64.90(uzh-1) Asturian 98.70(uzh-1) 92.40(iitbhu-iiith-2) 74.60(uzh-2) Azeri 100.00(axsemantics-2) 96.00(iitbhu-iiith-2) 65.00(iitbhu-iiith-2) Bashkir 99.90(uzh-2) 97.30(uzh-2) 77.80(iitbhu-iiith-1) Basque 98.90(bme-2) 88.10(iitbhu-iiith-2) 13.30(uzh-1) Belarusian 94.90(uzh-1) 70.40(uzh-1) 33.40(ua-8) Bengali 99.00(bme-3) 99.00(uzh-2) 72.00(uzh-2) Breton 100.00(waseda-1) 96.00(uzh-2) 72.00(uzh-1) Bulgarian 98.30(uzh-2) 83.80(uzh-2) 62.90(ua-8) Catalan 98.90(uzh-2) 92.80(waseda-1) 72.50(ua-8) Classical-syriac 100.00(axsemantics-1) 100.00(axsemantics-2) 96.00(uzh-2) Cornish 70.00(uzh-1) 40.00(ua-4) Crimean-tatar 100.00(iit-varanasi-1) 98.00(uzh-2) 91.00(iitbhu-iiith-2) Czech 94.70(uzh-1) 87.20(uzh-1) 46.50(uzh-2) Danish 95.50(uzh-1) 80.40(uzh-1) 87.70(ua-6) Dutch 97.90(uzh-1) 85.70(uzh-1) 69.30(ua-6) English 97.10(uzh-2) 94.50(uzh-1) 91.80(ua-8) Estonian 98.40(uzh-2) 81.60(uzh-1) 35.20(uzh-1) Faroese 87.10(bme-2) 72.60(uzh-1) 49.80(ua-8) Finnish 95.40(uzh-1) 82.80(uzh-1) 25.70(uzh-1) French 90.40(uzh-2) 80.90(uzh-2) 66.60(uzh-2) Friulian 99.00(axsemantics-2) 97.00(iitbhu-iiith-1) 79.00(uzh-2) Galician 99.50(uzh-1) 90.80(uzh-1) 61.10(uzh-2) Georgian 99.10(uzh-1) 94.00(uzh-2) 88.20(ua-8) German 90.20(uzh-2) 80.10(uzh-1) 67.10(ua-3) Greek 91.70(uzh-1) 75.50(uzh-2) 32.30(uzh-1) Greenlandic 98.00(uzh-2) 80.00(iitbhu-iiith-1) Haida 100.00(axsemantics-2) 94.00(uzh-2) 63.00(uzh-2) Hebrew 99.50(uzh-1) 85.40(uzh-1) 56.70(ua-8) Hindi 100.00(axsemantics-1) 97.60(uzh-2) 78.00(uzh-2) Hungarian 87.20(uzh-1) 74.50(iitbhu-iiith-2) 48.20(ua-8) Icelandic 91.30(uzh-1) 73.80(uzh-1) 56.20(ua-8) Ingrian 92.00(uzh-2) 46.00(iitbhu-iiith-2) Irish 91.50(uzh-2) 77.10(uzh-1) 37.70(uzh-1) Italian 98.00(uzh-2) 95.10(uzh-2) 57.40(uzh-2) Kabardian 100.00(hamburg-1) 100.00(bme-2) 92.00(uzh-1) Kannada 100.00(bme-3) 94.00(uzh-2) 61.00(uzh-1) Karelian 100.00(uzh-2) 94.00(ua-5) Kashubian 88.00(bme-2) 68.00(ua-5) Kazakh 88.00(iitbhu-iiith-2) 86.00(uzh-2) Khakas 98.00(bme-3) 86.00(iitbhu-iiith-2) Khaling 99.70(uzh-1) 86.00(iitbhu-iiith-1) 33.80(ua-8) Kurmanji 94.60(uzh-1) 93.20(uzh-1) 87.40(uzh-2) Ladin 99.00(uzh-2) 95.00(uzh-2) 72.00(uzh-1) Latin 78.90(bme-2) 53.30(uzh-1) 33.10(ua-6)

Table 9: Best per-form accuracy (and corresponding system) by language. First 50 languages.
{adjustbox}

width=0.85 Task 1 - Part 2 High Medium Low Latvian 98.20(uzh-2) 90.60(uzh-1) 57.30(ua-6) Lithuanian 95.50(uzh-2) 63.90(uzh-1) 32.60(ua-6) Livonian 100.00(uzh-2) 82.00(uzh-1) 35.00(ua-8) Lower-sorbian 97.80(uzh-1) 85.10(uzh-1) 54.30(ua-6) Macedonian 97.40(uzh-1) 91.60(uzh-1) 68.80(ua-6) Maltese 97.00(uzh-2) 95.00(uzh-1) 49.00(ua-6) Mapudungun 100.00(uzh-2) 86.00(ua-4) Middle-french 99.30(uzh-2) 94.50(uzh-2) 84.50(uzh-2) Middle-high-german 100.00(uzh-2) 84.00(uzh-2) Middle-low-german 100.00(iitbhu-iiith-1) 54.00(uzh-1) Murrinhpatha 96.00(uzh-2) 38.00(ua-8) Navajo 91.00(bme-2) 54.30(uzh-1) 20.80(uzh-1) Neapolitan 99.00(uzh-2) 99.00(uzh-2) 89.00(uzh-2) Norman 88.00(iitbhu-iiith-1) 66.00(ua-4) North-frisian 96.00(bme-1) 91.00(uzh-1) 45.00(iitbhu-iiith-2) Northern-sami 98.30(uzh-1) 76.10(uzh-1) 35.80(ua-8) Norwegian-bokmaal 92.10(uzh-2) 84.10(uzh-1) 90.10(ua-6) Norwegian-nynorsk 94.90(uzh-2) 67.10(uzh-1) 83.60(ua-8) Occitan 99.00(bme-2) 96.00(waseda-1) 77.00(uzh-2) Old-armenian 90.40(uzh-2) 80.20(uzh-1) 42.00(uzh-2) Old-church-slavonic 97.00(uzh-2) 93.00(uzh-2) 53.00(iitbhu-iiith-2) Old-english 88.70(uzh-1) 65.60(uzh-1) 46.50(ua-8) Old-french 92.40(uzh-1) 79.30(uzh-1) 46.20(uzh-2) Old-irish 40.00(uzh-1) 8.00(baseline) Old-saxon 98.30(uzh-1) 80.90(uzh-2) 46.60(ua-6) Pashto 100.00(waseda-1) 85.00(uzh-1) 48.00(uzh-2) Persian 99.90(bme-2) 93.40(uzh-2) 67.60(uzh-2) Polish 93.40(uzh-2) 82.40(uzh-2) 49.40(ua-6) Portuguese 98.60(uzh-2) 94.80(uzh-2) 75.80(uzh-2) Quechua 99.90(uzh-2) 98.90(uzh-1) 70.20(uzh-2) Romanian 89.00(uzh-2) 77.60(uzh-1) 46.20(uzh-1) Russian 94.40(uzh-2) 86.90(uzh-1) 53.50(uzh-1) Sanskrit 96.50(uzh-1) 85.90(uzh-2) 58.00(uzh-1) Scottish-gaelic 94.00(iitbhu-iiith-1) 74.00(iitbhu-iiith-2) Serbo-croatian 92.40(uzh-2) 86.10(uzh-1) 44.80(ua-3) Slovak 97.10(uzh-1) 78.60(uzh-1) 51.80(uzh-2) Slovene 97.40(uzh-1) 86.20(uzh-1) 58.00(uzh-2) Sorani 90.60(uzh-2) 80.20(iitbhu-iiith-2) 40.10(uzh-1) Spanish 98.10(uzh-2) 92.00(iitbhu-iiith-2) 73.20(ua-8) Swahili 100.00(bme-3) 99.00(uzh-2) 72.00(iitbhu-iiith-2) Swedish 93.30(uzh-1) 79.80(uzh-1) 79.00(ua-8) Tatar 99.00(axsemantics-1) 98.00(uzh-2) 90.00(ua-8) Telugu 96.00(ua-8) Tibetan 56.00(uzh-2) 58.00(iitbhu-iiith-1) Turkish 98.50(uzh-2) 90.70(uzh-1) 39.50(iitbhu-iiith-2) Turkmen 98.00(iitbhu-iiith-1) 90.00(uzh-2) Ukrainian 96.20(uzh-2) 81.40(uzh-1) 57.10(ua-6) Urdu 100.00(iitbhu-iiith-1) 96.80(uzh-2) 72.50(uzh-2) Uzbek 100.00(axsemantics-1) 100.00(axsemantics-2) 92.00(uzh-1) Venetian 99.20(uzh-2) 95.10(uzh-2) 78.80(uzh-2) Votic 90.00(uzh-2) 88.00(uzh-2) 34.00(ua-7) Welsh 95.00(bme-3) 85.00(bme-2) 55.00(uzh-2) West-frisian 99.00(uzh-1) 98.00(uzh-2) 56.00(uzh-1) Yiddish 100.00(uzh-2) 94.00(uzh-2) 87.00(ua-8) Zulu 99.80(uzh-1) 87.30(uzh-2) 33.00(uzh-1)

Table 10: Best per-form accuracy (and corresponding system) by language. Remaining 53 languages.

5.2 Task 2 Results

All systems submitted for task 2 were neural systems. All but one of the systems were encoder-decoder systems reminiscent of \newcitekann-schutze:2016:P16-2. The exception, \newcitemakarov-clematide:2018:K18-20, used a neural transition-based transducer with a designated copy action, which edits the input lemma into an output form. Table 6 details some of the design features in task 2 systems.

Predict MSD systems predicted the MSD of the target word form based on contextual cues and used the MSD to improve performance. The system by \newcitekementchedjhieva-bjerva-augenstein:2018:K18-20 used MSD prediction as an auxiliary task. The system by \newciteliu-EtAl:2018:K18-20 instead converted the contextual reinflection problem into ordinary morphological reinflection. They first predicted the MSD of the target word form based on sentence context and then generated the target word form using the input lemma and the predicted MSD.

Several systems improved upon the context model in the neural baseline system. Three systems (BME-HAS, NYU, and ZHU) used subword context models, for example, character-level models to encode context word forms, lemmata and MSDs. Many systems Ács (2018); Kementchedjhieva et al. (2018); Kann et al. (2018) also used a context RNN for encoding sentence context exceeding the immediate neighboring words. \newcitekann-lauly-cho:2018:K18-20 used context attention which refers to an attention mechanisms directed at contextual information.

The system by \newcitekementchedjhieva-bjerva-augenstein:2018:K18-20 was multilingual in the sense that it combined training data for all task 2 languages. Finally, the system by \newcitemakarov-clematide:2018:K18-20 used beam search for decoding.

Overall performance for all data settings in tracks 1 and 2 of task 2 is described in Table 11. For evaluation with regard to original forms, the evaluation criterion is accuracy; that is, how often a system correctly predicted the original UD form. For evaluation with regard to plausible forms, the evaluation criterion is relaxed accuracy given the set of contextually plausible forms. In other words, we measure how often the prediction was one of the variants in the set of plausible forms.

In track 1, the COPENHAGEN system is the clear winner in the high and medium data settings, whereas the UZH system is the clear winner in the low data setting. In fact, UZH is the only system which can beat the lemma copying baseline COPY-BL in the low setting. In track 2, the COPENHAGEN system and the neural baseline system NEURAL-BL deliver comparable performance in the high data setting. In the medium and low setting, the UZH system is the clear winner. Once again, the UZH system is the only system which can beat the lemma copying baseline COPY-BL in the low setting.

Table 11 shows that the best track 1 system outperforms the best track 2 system for every data setting, meaning that the additional supervision offered by context lemmata and MSDs is useful. Moreover, this effect seems to strengthen with increasing amounts of training data: the difference in performance between the best track 1 and track 2 systems for original forms in the low data setting is 3.8%-points, in the medium setting 7.8%-points, and in the high setting 13.6%-points. A further observation is that it seems to be more difficult to deliver improvements over the neural baseline system NEURAL-BL in the high setting in track 2, where NEURAL-BL in fact is one of the top two systems. This may be a result of the relatively small training sets: even in the high data setting, the training sets only contain approximately tokens.

The results on original and plausible forms show strong agreement. In all but one case, the same systems deliver the strongest performance for both evaluation criteria. The only exception is the Track 2 high setting where COPENHAGEN is the top system with regard to original forms and NEURAL-BL with regard to plausible forms. However, the performance of these systems is very similar. This strong agreement indicates that evaluation on plausible forms might not be necessary.

The best-performing systems for each language, track, and data setting in task 2 are given in Table 12. In track 1, COPENHAGEN achieves the strongest results for most languages in the high and medium data settings, whereas UZH delivers the best performance on all languages in the low setting. In track 2, COPENHAGEN and NEURAL-BL deliver the best performance on an equal number of languages in the high setting, whereas UZH delivers best performance for most languages in the low and medium settings, and COPENHAGEN performs best for the remaining languages.

{adjustbox}

width= Track 1 Track 2 Original Plausible Original Plausible High Medium Low High Medium Low High Medium Low High Medium Low BME-HAS 65.69 45.71 29.34 73.21 51.28 32.98 51.83 36.82 24.71 60.15 43.80 31.18 COPENHAGEN 68.51 56.70 24.40 76.10 63.24 26.24 54.93 45.18 29.38 60.50 51.36 33.77 CUBoulder–1 59.73 46.27 23.16 66.22 52.52 25.59 48.97 38.29 23.76 55.63 43.33 26.83 CUBoulder–2 50.32 42.08 29.86 53.89 46.85 34.85 - - - - - - NYU - - - - - - - - 33.38 - - 38.62 UZH - 53.02 42.42 - 61.02 48.49 - 48.88 38.60 - 55.67 45.09 NEURAL-BL 62.41 44.09 01.85 69.53 48.81 02.63 54.48 38.56 02.19 60.79 46.74 03.11 COPY-BL 36.62 36.62 36.62 42.00 42.00 42.00 36.62 36.62 36.62 42.00 42.00 42.00

Table 11: Overall accuracies (in %-points) for Tracks 1 and 2 in Task 2 for different training data settings. Results are presented separately with regard to the original forms in the UD test data sets and the manually annotated sets of plausible forms. NEURAL-BL refers to the baseline encoder-decoder system and COPY-BL to the “lemma copying” baseline system. Note that the output of the COPY-BL is independent of the training data and therefore results for the high, medium and low data setting are the same.
{adjustbox}

width= Track 1 Original Plausible High Medium Low High Medium Low de 73.21 (BME-HAS) 63.90 (UZH) 60.06 (UZH) 77.55 (BME-HAS) 67.34 (UZH) 62.39 (UZH) en 77.84 (CPH) 68.08 (CBL) 68.08 (UZH) 86.81 (CPH) 76.23 (CPH) 74.02 (UZH) es 56.24 (CPH) 51.33 (CPH) 34.78 (UZH) 67.88 (CPH) 60.59 (CPH) 42.08 (UZH) fi 55.27 (CPH) 35.71 (CPH) 24.90 (UZH) 63.02 (CPH) 43.07 (CPH) 28.97 (UZH) fr 70.67 (CPH) 60.29 (CPH) 35.03 (UZH) - - - ru 77.91 (CPH) 63.05 (CPH) 40.76 (UZH) 81.53 (CPH) 66.57 (CPH) 43.47 (UZH) sv 69.26 (CPH) 57.66 (CPH) 33.30 (UZH) 80.32 (CPH) 67.23 (CPH) 40.00 (UZH) Track 2 Original Plausible High Medium Low High Medium Low de 65.72 (NBL) 60.26 (UZH) 59.15 (UZH) 69.97 (NBL) 64.21 (UZH) 61.38 (UZH) en 71.90 (CPH) 68.08 (UZH) 68.08 (UZH) 79.86 (CPH) 75.63 (CPH) 74.02 (UZH) es 51.05 (NBL) 42.50 (CPH) 32.68 (UZH) 59.19 (NBL) 51.75 (CPH) 37.31 (CPH) fi 34.82 (NBL) 27.06 (UZH) 24.40 (UZH) 41.17 (NBL) 31.89 (UZH) 28.21 (UZH) fr 61.51 (CPH) 45.62 (CPH) 29.53 (CPH) - - - ru 56.73 (BME-HAS) 54.02 (UZH) 28.11 (UZH) 60.04 (BME-HAS) 56.53 (UZH) 30.42 (UZH) sv 55.96 (CPH) 47.87 (UZH) 32.77 (UZH) 66.06 (CPH) 56.17 (UZH) 39.36 (UZH)

Table 12: Best accuracies (in %-points) and the for all tracks, settings and languages in task 2. The best performing system is given in parentheses. “CPH” refers to “COPENHAGEN”, “NBL” to the neural baseline system and “CBL” to the “lemma copying” baseline system. Note, that there are no results for French with regard to plausible forms because this gold standard data set was not annotated for plausible forms (see subsection 3.2).

6 Future Directions

In the case of inflection an interesting future topic could involve departing from orthographic representation and using more IPA-like representations, i.e. transductions over pronunciations. Different languages, in particular those with idiosyncratic orthographies, may offer new challenges in this respect.12

Neither task this year included unannotated monolingual corpora. Using such data is well-motivated from an L1-learning point of view, and may affect the performance of low-resource data settings, especially for the cloze task. In the inflection task, some results from last year Zhou and Neubig (2017) did not see significant gains by using extra data.

Only one team tried to learn inflection in a multilingual setting—i.e. to use all training data to train one model. Such transfer learning is an interesting avenue of future research, but evaluation could be difficult. Whether any cross-language transfer is actually being learned vs. whether having more data better biases the networks to copy strings is an evaluation step to disentangle.13

Creating new data sets that accurately reflect learner exposure (whether L1 or L2) is also an important consideration in the design of future shared tasks.

The results for task 2 show that evaluation against the original test form versus against set of plausible forms results in a very similar ranking of systems, justifying the use of the former, much simpler, method for future shared tasks. No manual annotation would then be required for the creation of test sets, allowing the inclusion of a wider variety of languages.

In track 2 of task 2, it turned out to be difficult to achieve clear improvements over the neural baseline system. This may be a consequence of the limited amount of training data. Increasing the amount of training data is an obvious solution, but encouraging the use of external datasets for semi-supervised learning could also be an interesting direction to pursue. Such semi-supervised methods could take the form of pretrained embeddings from monolingual corpora or more expressive models dedicated to improving morphological inflection, e.g., Wolf-Sonkin et al. (2018).

7 Conclusion

The CoNLL–SIGMORPHON 2018 shared task introduced a new cloze-test task with data sets for 7 languages, as well as extended the existing inflection task to include 103 languages. In task 1 (inflection) 27 systems were submitted, while 6 systems were submitted in task 2 (cloze test). Neural network models prevailed in both, although significant modifications to standard architectures were required to beat a simple baseline in the low data settings in both tasks.

As in previous years, we compared inflection system performance to oracle ensembles, showing that systems possessed complementary strengths. We released the training, development, and test sets for each task, and expect these to be useful for future endeavors in morphological learning, both in sentential context and in the case of isolated word inflection.

Acknowledgements

The first author would like to acknowledge the support of an NDSEG fellowship. MS was supported by a grant from the Society of Swedish Literature in Finland (SLS). Several authors (CK, DY, JSG, MH) were supported in part by the Defense Advanced Research Projects Agency (DARPA) in the program Low Resource Languages for Emergent Incidents (LORELEI) under contract No. HR0011-15-C-0113. Any opinions, findings and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the Defense Advanced Research Projects Agency (DARPA). NVIDIA Corp. donated a Titan Xp GPU used for this research.

Appendix A Detailed Task 1 Results

This section contains detailed results for each submitted system on each language. Systems are ordered by average per-form accuracy for each sub-task and data condition. Three metrics are presented for each system/language combination.

  1. Per-Form Accuracy: Percentage of test forms inflected correctly.

  2. Levenshtein Distance: Average Levenshtein distance of system-predicted form from gold inflected form.

Scores in bold include the highest scoring non-oracle system for each language as well as any other systems that did not differ significantly in terms of per-form accuracy according to a sign test (). Scores marked with a indicate submissions that were significantly better than the feature combination oracle (), showing per-feature generalization. Scores marked with did not differ significantly from the ensemble oracle, suggesting minimal complementary information across systems.

{adjustbox}

width= oracle-fc oracle-e uzh-01 uzh-02 bme-02 iitbhu-iiith-01 iitbhu-iiith-02 bme-03 bme-01 msu-04 iit-varanasi-01 waseda-01 Adyghe 100.00/* 100.00/* 100.00/0.00 100.00/0.00 97.90/0.02 99.90/0.00 99.90/0.00 99.40/0.01 99.20/0.01 99.30/0.01 99.80/0.00 95.00/0.07 Albanian 100.00/* 99.60/* 97.70/0.08 96.50/0.12 98.90/0.02 98.00/0.04 98.00/0.04 97.50/0.05 97.50/0.05 95.00/0.14 96.50/0.05 88.80/0.41 Arabic 100.00/* 97.40/* 93.70/0.28 93.50/0.28 92.20/0.30 93.30/0.26 93.30/0.26 90.30/0.32 90.60/0.37 0.00/9.16 84.40/0.45 90.20/0.34 Armenian 100.00/* 99.60/* 96.40/0.07 96.80/0.05 96.90/0.05 96.10/0.07 96.10/0.07 94.70/0.09 94.70/0.09 94.90/0.12 93.70/0.10 94.30/0.10 Asturian 99.20/* 99.80/* 98.70/0.02 98.50/0.03 98.00/0.04 98.60/0.03 98.60/0.03 97.80/0.04 97.80/0.04 98.40/0.03 98.50/0.03 98.40/0.03 Azeri 100.00/* 100.00/* 98.00/0.02 98.00/0.02 97.00/0.05 99.00/0.01 99.00/0.01 98.00/0.02 98.00/0.02 98.00/0.02 99.00/0.02 93.00/0.10 Bashkir 100.00/* 100.00/* 99.90/0.00 99.90/0.00 99.80/0.00 99.80/0.00 99.80/0.00 99.80/0.00 99.80/0.00 99.80/0.00 99.70/0.01 93.20/0.11 Basque 99.00/* 99.70/* 98.90/0.02 98.70/0.03 98.90/0.02 98.60/0.03 98.60/0.03 98.30/0.03 98.10/0.07 95.10/0.09 98.00/0.04 97.20/0.06 Belarusian 100.00/* 98.90/* 94.90/0.09 94.70/0.09 93.10/0.14 92.10/0.14 92.10/0.14 92.90/0.12 92.60/0.12 92.70/0.14 88.40/0.20 85.60/0.31 Bengali 100.00/* 99.00/* 99.00/0.03 99.00/0.03 99.00/0.05 99.00/0.02 99.00/0.02 99.00/0.05 99.00/0.05 98.00/0.06 99.00/0.05 98.00/0.06 Breton 94.00/* 100.00/* 97.00/0.05 98.00/0.06 92.00/0.22 99.00/0.03 99.00/0.03 93.00/0.17 93.00/0.17 96.00/0.08 93.00/0.17 100.00/0.00 Bulgarian 100.00/* 99.40/* 98.10/0.04 98.30/0.04 96.40/0.06 96.80/0.05 96.80/0.05 95.40/0.08 94.90/0.08 96.20/0.06 94.50/0.10 91.90/0.21 Catalan 100.00/* 99.60/* 98.70/0.04 98.90/0.03 98.40/0.04 97.70/0.04 97.70/0.04 98.20/0.05 98.20/0.05 98.40/0.04 98.00/0.05 98.50/0.04 Classical-syriac 100.00/* 100.00/* 100.00/0.00 100.00/0.00 99.00/0.01 99.00/0.01 99.00/0.01 98.00/0.02 97.00/0.04 100.00/0.00 100.00/0.00 98.00/0.02 Crimean-tatar 100.00/* 100.00/* 98.00/0.04 98.00/0.04 99.00/0.03 99.00/0.03 99.00/0.03 99.00/0.03 99.00/0.03 99.00/0.03 100.00/0.00 97.00/0.05 Czech 99.80/* 97.90/* 94.70/0.10 94.50/0.11 93.20/0.12 91.10/0.16 91.10/0.16 92.00/0.14 92.00/0.14 93.10/0.12 88.00/0.23 90.70/0.21 Danish 100.00/* 98.50/* 95.50/0.07 94.60/0.08 90.40/0.16 91.50/0.13 91.50/0.13 89.40/0.17 89.40/0.17 91.80/0.12 91.30/0.13 90.70/0.15 Dutch 100.00/* 99.20/* 97.90/0.03 97.70/0.04 96.80/0.05 95.30/0.09 95.30/0.09 94.60/0.08 94.60/0.08 96.00/0.06 93.10/0.12 95.60/0.07 English 100.00/* 99.20/* 97.00/0.06 97.10/0.06 96.70/0.06 96.30/0.07 96.30/0.07 96.10/0.07 95.80/0.08 94.30/0.11 95.80/0.07 95.40/0.08 Estonian 100.00/* 99.60/* 98.30/0.05 98.40/0.05 97.00/0.07 97.30/0.06 97.30/0.06 96.90/0.07 96.90/0.07 94.40/0.11 95.90/0.09 91.50/0.21 Faroese 100.00/* 97.50/* 85.60/0.29 86.40/0.26 87.10/0.26 83.90/0.33 83.90/0.33 83.80/0.33 85.30/0.27 84.80/0.33 81.10/0.37 81.30/0.37 Finnish 100.00/* 98.90/* 95.40/0.09 94.90/0.09 93.30/0.11 92.30/0.15 92.30/0.15 92.00/0.15 92.00/0.15 89.80/0.19 76.40/0.47 84.40/0.36 French 100.00/* 96.60/* 90.20/0.16 90.40/0.16 88.10/0.19 87.60/0.21 87.60/0.21 86.40/0.22 86.40/0.22 88.50/0.20 81.10/0.32 84.80/0.25 Friulian 100.00/* 99.00/* 99.00/0.03 99.00/0.03 99.00/0.03 99.00/0.03 99.00/0.03 99.00/0.03 99.00/0.03 99.00/0.03 99.00/0.03 99.00/0.03 Galician 100.00/* 99.80/* 99.50/0.01 99.30/0.01 99.30/0.01 98.90/0.01 98.90/0.01 99.20/0.01 99.20/0.01 97.90/0.04 98.70/0.02 98.60/0.03 Georgian 100.00/* 99.60/* 99.10/0.01 98.80/0.01 98.00/0.05 98.10/0.02 98.10/0.02 97.10/0.05 96.60/0.05 97.50/0.04 97.30/0.04 97.80/0.04 German 100.00/* 96.70/* 89.70/0.25 90.20/0.23 88.50/0.27 87.60/0.26 87.60/0.26 86.20/0.29 86.20/0.29 88.30/0.24 83.70/0.33 87.50/0.27 Greek 100.00/* 97.30/* 91.70/0.16 91.00/0.17 89.10/0.24 88.20/0.23 88.20/0.23 86.60/0.26 86.60/0.26 88.60/0.22 80.80/0.38 83.70/0.31 Haida 100.00/* 100.00/* 99.00/0.02 99.00/0.02 99.00/0.02 93.00/0.23 93.00/0.23 100.00/0.00 100.00/0.00 99.00/0.02 100.00/0.00 66.00/0.73 Hebrew 100.00/* 99.70/* 99.50/0.01 99.30/0.01 99.30/0.01 98.80/0.02 98.80/0.02 98.50/0.02 98.50/0.02 98.20/0.02 97.30/0.03 98.40/0.02 Hindi 100.00/* 100.00/* 100.00/0.00 100.00/0.00 100.00/0.00 99.70/0.01 99.70/0.01 99.90/0.00 100.00/0.00 99.90/0.00 99.40/0.01 73.60/0.71 Hungarian 100.00/* 94.30/* 87.20/0.31 86.60/0.31 85.50/0.34 85.90/0.34 85.90/0.34 83.70/0.38 83.70/0.38 84.30/0.36 82.30/0.41 79.60/0.47 Icelandic 100.00/* 96.90/* 91.30/0.18 91.10/0.18 87.00/0.27 85.00/0.29 85.00/0.29 86.00/0.27 86.00/0.27 86.10/0.30 83.90/0.32 81.90/0.34 Irish 100.00/* 96.90/* 91.40/0.28 91.50/0.27 91.10/0.27 87.60/0.39 87.60/0.39 87.70/0.37 88.80/0.33 87.20/0.36 67.20/0.99 77.80/0.72 Italian 100.00/* 98.80/* 98.00/0.04 98.00/0.04 97.40/0.04 97.50/0.06 97.50/0.06 96.30/0.07 96.70/0.06 97.40/0.04 95.70/0.08 97.40/0.04 Kabardian 100.00/* 100.00/* 99.00/0.01 96.00/0.04 99.00/0.01 99.00/0.01 99.00/0.01 98.00/0.02 98.00/0.02 97.00/0.03 99.00/0.01 95.00/0.09 Kannada 100.00/* 100.00/* 100.00/0.00 100.00/0.00 100.00/0.00 98.00/0.02 98.00/0.02 100.00/0.00 99.00/0.01 99.00/0.01 100.00/0.00 99.00/0.02 Khaling 100.00/* 100.00/* 99.70/0.01 99.60/0.01 99.60/0.00 99.50/0.01 99.50/0.01 99.00/0.02 99.60/0.01 93.90/0.08 98.40/0.03 99.40/0.01 Kurmanji 100.00/* 98.90/* 94.60/0.06 94.40/0.07 93.60/0.12 93.80/0.11 93.80/0.11 92.80/0.11 92.80/0.11 93.40/0.08 93.50/0.08 89.60/0.21 Ladin 100.00/* 100.00/* 98.00/0.03 99.00/0.01 97.00/0.06 98.00/0.03 98.00/0.03 97.00/0.07 94.00/0.10 97.00/0.07 98.00/0.03 98.00/0.05 Latin 100.00/* 94.40/* 75.90/0.35 74.60/0.39 78.90/0.32 73.70/0.42 73.70/0.42 77.40/0.37 77.40/0.37 72.40/0.42 61.50/0.60 65.40/0.55 Latvian 100.00/* 99.70/* 97.70/0.03 98.20/0.02 96.00/0.07 96.10/0.07 96.10/0.07 94.00/0.12 94.00/0.12 95.80/0.07 92.80/0.16 94.90/0.09 Lithuanian 100.00/* 98.90/* 95.40/0.07 95.50/0.07 91.60/0.13 91.80/0.15 91.80/0.15 87.50/0.19 90.90/0.14 88.90/0.16 88.10/0.19 88.10/0.23 Livonian 100.00/* 100.00/* 98.00/0.03 100.00/0.00 87.00/0.24 98.00/0.03 98.00/0.03 92.00/0.15 92.00/0.15 92.00/0.12 97.00/0.06 94.00/0.11 Lower-sorbian 100.00/* 99.30/* 97.80/0.05 97.50/0.06 96.60/0.07 96.30/0.07 96.30/0.07 96.40/0.06 96.40/0.06 95.60/0.09 94.30/0.11 95.40/0.10 Macedonian 100.00/* 99.00/* 97.40/0.04 97.10/0.04 95.30/0.07 96.50/0.06 96.50/0.06 94.40/0.10 94.30/0.09 95.30/0.07 94.60/0.08 88.30/0.16 Maltese 99.00/* 98.00/* 96.00/0.09 97.00/0.07 94.00/0.08 95.00/0.10 95.00/0.10 90.00/0.15 88.00/0.15 83.00/0.27 69.00/0.56 95.00/0.08 Middle-french 99.90/* 99.80/* 99.20/0.01 99.30/0.02 99.00/0.02 98.70/0.02 98.70/0.02 98.60/0.03 98.60/0.03 99.00/0.02 96.70/0.08 98.70/0.03 Navajo 100.00/* 97.00/* 88.10/0.29 87.40/0.33 91.00/0.21 83.50/0.44 83.50/0.44 86.70/0.32 86.70/0.32 87.50/0.28 75.30/0.53 83.70/0.39 Neapolitan 99.00/* 99.00/* 99.00/0.02 99.00/0.02 98.00/0.06 97.00/0.06 97.00/0.06 97.00/0.09 97.00/0.09 97.00/0.04 97.00/0.06 95.00/0.13 North-frisian 100.00/* 99.00/* 94.00/0.15 95.00/0.14 95.00/0.10 96.00/0.05 96.00/0.05 94.00/0.15 96.00/0.06 87.00/0.35 96.00/0.12 92.00/0.14 Northern-sami 100.00/* 99.10/* 98.30/0.04 98.10/0.05 96.20/0.07 96.90/0.06 96.90/0.06 95.20/0.10 95.80/0.08 96.10/0.07 92.60/0.12 94.50/0.11 Norwegian-bokmaal 99.90/* 98.60/* 92.00/0.13 92.10/0.13 88.30/0.19 89.00/0.18 89.00/0.18 88.20/0.19 88.20/0.19 88.60/0.18 89.00/0.17 90.50/0.17 Norwegian-nynorsk 100.00/* 98.90/* 94.60/0.09 94.90/0.08 85.30/0.26 84.60/0.26 84.60/0.26 90.80/0.15 90.80/0.15 86.80/0.23 82.60/0.29 83.40/0.28 Occitan 100.00/* 99.00/* 99.00/0.01 99.00/0.01 99.00/0.01 98.00/0.03 98.00/0.03 97.00/0.04 97.00/0.04 99.00/0.01 99.00/0.01 98.00/0.02 Old-armenian 99.80/* 97.50/* 90.40/0.19 90.40/0.19 89.30/0.21 89.10/0.21 89.10/0.21 87.70/0.23 87.70/0.23 88.60/0.20 87.30/0.25 88.80/0.22 Old-church-slavonic 100.00/* 99.00/* 97.00/0.04 97.00/0.04 94.00/0.10 96.00/0.07 96.00/0.07 94.00/0.10 93.00/0.11 93.00/0.12 97.00/0.04 97.00/0.04 Old-english 100.00/* 97.20/* 88.70/0.20 87.90/0.22 88.20/0.23 86.00/0.25 86.00/0.25 87.10/0.23 87.10/0.23 84.50/0.27 83.40/0.30 83.70/0.28 Old-french 99.60/* 98.00/* 92.40/0.13 91.70/0.15 91.40/0.17 89.90/0.20 89.90/0.20 91.50/0.15 91.50/0.16 90.30/0.18 86.50/0.25 88.40/0.22 Old-saxon 100.00/* 99.60/* 98.30/0.03 97.80/0.04 97.20/0.05 97.00/0.06 97.00/0.06 96.10/0.07 96.30/0.06 95.40/0.07 96.30/0.06 95.10/0.08 Pashto 100.00/* 100.00/* 98.00/0.04 97.00/0.05 98.00/0.03 99.00/0.01 99.00/0.01 94.00/0.09 93.00/0.14 97.00/0.07 100.00/0.00 100.00/0.00 Persian 100.00/* 100.00/* 99.80/0.00 99.80/0.00 99.90/0.00 99.70/0.00 99.70/0.00 99.60/0.01 99.50/0.01 98.60/0.02 98.90/0.01 62.90/1.14 Polish 100.00/* 97.80/* 93.00/0.16 93.40/0.14 90.80/0.25 88.70/0.26 88.70/0.26 89.30/0.25 89.30/0.25 87.30/0.25 82.80/0.40 89.30/0.24 Portuguese 100.00/* 99.40/* 98.60/0.02 98.60/0.02 98.20/0.03 98.00/0.04 98.00/0.04 97.90/0.03 97.90/0.03 97.80/0.04 98.50/0.02 97.90/0.04 Quechua 100.00/* 99.90/* 99.90/0.00 99.90/0.00 99.60/0.01 99.50/0.01 99.50/0.01 99.30/0.01 99.30/0.01 99.30/0.01 99.40/0.01 95.10/0.10 Romanian 100.00/* 96.80/* 88.30/0.35 89.00/0.34 87.90/0.39 85.80/0.40 85.80/0.40 84.10/0.50 85.20/0.46 84.60/0.42 79.00/0.58 83.20/0.52 Russian 100.00/* 97.90/* 94.00/0.15 94.40/0.13 92.00/0.28 91.00/0.27 91.00/0.27 90.40/0.27 90.40/0.27 92.00/0.21 85.40/0.35 88.40/0.34 Sanskrit 100.00/* 99.30/* 96.50/0.05 96.30/0.06 93.00/0.12 94.40/0.09 94.40/0.09 93.80/0.10 93.80/0.10 93.50/0.10 92.10/0.12 87.10/0.20 Serbo-croatian 100.00/* 96.80/* 92.10/0.16 92.40/0.15 91.30/0.17 91.30/0.18 91.30/0.18 90.00/0.18 90.00/0.18 87.80/0.25 84.10/0.36 85.50/0.28 Slovak 100.00/* 99.70/* 97.10/0.04 96.90/0.04 94.20/0.09 92.50/0.14 92.50/0.14 93.30/0.10 93.30/0.10 93.10/0.12 92.80/0.13 90.80/0.16 Slovene 100.00/* 99.60/* 97.40/0.05 97.30/0.05 94.90/0.10 94.80/0.08 94.80/0.08 94.10/0.11 94.90/0.10 33.30/0.77 95.30/0.08 95.60/0.07 Sorani 100.00/* 99.30/* 90.60/0.11 90.60/0.11 88.90/0.13 89.50/0.13 89.50/0.13 89.50/0.14 89.50/0.14 89.30/0.13 89.00/0.13 88.30/0.14 Spanish 100.00/* 99.90/* 97.90/0.03 98.10/0.02 97.90/0.03 97.70/0.03 97.70/0.03 96.20/0.06 96.20/0.06 97.70/0.03 93.40/0.13 97.20/0.04 Swahili 100.00/* 100.00/* 99.00/0.01 99.00/0.01 100.00/0.00 99.00/0.01 99.00/0.01 100.00/0.00 99.00/0.01 98.00/0.02 100.00/0.00 98.00/0.02 Swedish 100.00/* 98.50/* 93.30/0.12 93.10/0.12 89.30/0.19 88.10/0.23 88.10/0.23 88.40/0.21 88.40/0.21 89.10/0.20 87.20/0.21 88.10/0.22 Tatar 100.00/* 99.00/* 98.00/0.02 98.00/0.02 98.00/0.02 99.00/0.01 99.00/0.01 99.00/0.01 99.00/0.01 99.00/0.01 99.00/0.01 96.00/0.04 Turkish 100.00/* 99.40/* 98.10/0.03 98.50/0.02 97.80/0.03 97.10/0.07 97.10/0.07 97.30/0.04 97.30/0.04 97.40/0.04 97.30/0.04 89.50/0.21 Ukrainian 100.00/* 98.70/* 96.00/0.06 96.20/0.06 91.40/0.16 92.80/0.14 92.80/0.14 91.60/0.14 91.70/0.16 93.40/0.12 92.50/0.13 89.10/0.18 Urdu 100.00/* 100.00/* 99.50/0.01 99.50/0.01 99.30/0.01 100.00/0.00 100.00/0.00 99.20/0.01 99.40/0.01 99.20/0.01 99.70/0.01 97.30/0.04 Uzbek 100.00/* 100.00/* 100.00/0.00 100.00/0.00 100.00/0.00 100.00/0.00 100.00/0.00 100.00/0.00 100.00/0.00 100.00/0.00 98.00/0.03 93.00/0.28 Venetian 99.60/* 99.50/* 99.00/0.02 99.20/0.01 98.80/0.02 98.80/0.02 98.80/0.02 98.40/0.03 98.70/0.02 99.00/0.02 99.10/0.02 99.00/0.02 Votic 100.00/* 97.00/* 88.00/0.15 90.00/0.13 84.00/0.29 82.00/0.26 82.00/0.26 85.00/0.25 79.00/0.48 84.00/0.23 66.00/0.53 80.00/0.31 Welsh 100.00/* 98.00/* 94.00/0.10 93.00/0.11 94.00/0.10 94.00/0.09 94.00/0.09 95.00/0.10 95.00/0.10 95.00/0.06 94.00/0.09 90.00/0.19 West-frisian 100.00/* 100.00/* 99.00/0.01 98.00/0.02 91.00/0.26 91.00/0.23 91.00/0.23 93.00/0.22 93.00/0.22 89.00/0.23 74.00/0.55 92.00/0.18 Yiddish 100.00/* 100.00/* 100.00/0.00 100.00/0.00 99.00/0.01 100.00/0.00 100.00/0.00 99.00/0.01 98.00/0.02 97.00/0.04 99.00/0.06 98.00/0.04 Zulu 99.90/* 99.90/* 99.80/0.00 99.70/0.01 99.00/0.02 99.30/0.01 99.30/0.01 99.20/0.02 98.10/0.04 96.00/0.08 97.20/0.05 98.90/0.01

Table 13: Task 1 High Condition Part 1.
{adjustbox}

width= msu-03 axsemantics-01 msu-02 racai-01 hamburg-01 baseline axsemantics-02 msu-01 tuebingen-oslo-03 tuebingen-oslo-02 kucst-01 tuebingen-oslo-01 Adyghe 96.70/0.03 99.00/0.01 99.00/0.01 92.00/0.08 93.90/0.07 91.60/0.08 99.60/0.00 95.40/0.05 97.40/0.04 88.20/0.16 98.30/0.02 88.90/0.24 Albanian 92.70/0.17 88.90/0.53 92.40/0.20 95.60/0.11 79.90/0.54 79.60/0.69 40.90/2.66 83.80/0.60 17.90/5.12 10.20/5.87 13.80/3.68 11.60/5.88 Arabic 0.00/9.25 58.20/1.38 0.00/4.70 88.50/0.40 74.70/0.76 47.10/1.49 34.00/2.42 0.00/7.79 79.40/0.67 68.70/1.05 0.00/10.78 0.10/7.55 Armenian 95.40/0.08 90.30/0.17 91.30/0.15 87.60/0.20 86.70/0.25 73.40/0.84 90.10/0.15 52.70/1.85 47.10/2.06 47.60/1.36 33.60/2.38 Asturian 98.60/0.03 95.30/0.11 94.60/0.12 95.40/0.13 96.20/0.07 95.10/0.11 91.10/0.13 97.00/0.07 84.20/0.24 82.70/0.34 88.80/0.26 69.70/0.56 Azeri 98.00/0.06 94.00/0.15 95.00/0.05 90.00/0.19 88.00/0.24 71.00/0.86 100.00/0.00 66.00/0.76 91.00/0.14 88.00/0.24 47.00/1.07 79.00/0.50 Bashkir 98.90/0.03 99.80/0.00 99.10/0.02 99.10/0.01 90.50/0.26 99.80/0.00 96.50/0.11 97.10/0.05 90.90/0.17 62.80/0.72 91.60/0.18 Basque 94.50/0.10 8.20/2.45 13.40/3.17 86.80/0.35 46.90/1.24 7.60/3.37 5.40/2.28 24.30/2.85 6.20/2.85 9.60/2.58 73.20/0.51 8.10/2.58 Belarusian 86.40/0.24 47.30/1.25 81.60/0.62 3.80/2.73 57.20/1.28 40.80/1.56 52.80/1.23 68.20/1.66 69.00/1.22 53.30/1.68 62.60/0.78 57.40/1.61 Bengali 99.00/0.05 96.00/0.12 96.00/0.09 99.00/0.03 93.00/0.14 81.00/0.26 78.00/0.27 90.00/0.23 55.00/0.75 72.00/0.52 82.00/0.48 65.00/0.60 Breton 95.00/0.14 80.00/0.55 79.00/0.68 91.00/0.17 73.00/0.85 85.00/0.22 82.00/0.38 48.00/1.05 70.00/0.66 67.00/0.56 48.00/1.03 Bulgarian 95.80/0.07 88.70/0.21 91.70/0.14 77.80/0.38 79.60/0.31 89.00/0.18 75.60/0.82 95.30/0.08 63.70/1.31 46.90/1.83 59.70/0.80 51.40/1.89 Catalan 98.00/0.04 95.90/0.12 95.00/0.12 89.40/0.14 87.60/0.21 95.60/0.10 92.10/0.14 96.30/0.09 91.60/0.15 76.90/0.43 87.90/0.20 56.00/0.96 Classical-syriac 99.00/0.01 100.00/0.00 99.00/0.01 87.00/0.18 100.00/0.00 97.00/0.03 99.00/0.01 88.00/0.16 76.00/0.30 65.00/0.40 80.00/0.23 94.00/0.06 Crimean-tatar 94.00/0.14 98.00/0.04 99.00/0.03 94.00/0.09 98.00/0.04 95.00/0.08 98.00/0.04 98.00/0.04 97.00/0.06 92.00/0.12 96.00/0.07 95.00/0.09 Czech 92.40/0.15 89.20/0.23 85.90/0.26 90.20/0.20 82.40/0.35 90.50/0.20 83.00/0.69 79.90/0.50 67.40/1.50 55.00/1.55 31.90/2.07 43.60/2.18 Danish 90.40/0.15 92.70/0.13 79.50/0.32 91.80/0.13 76.00/0.36 86.70/0.25 90.40/0.14 66.40/1.04 85.00/0.27 73.30/0.45 64.50/0.81 66.60/0.73 Dutch 94.20/0.11 85.10/0.34 87.20/0.24 92.20/0.13 82.30/0.26 87.70/0.22 88.60/0.22 86.70/0.24 77.00/0.50 59.40/0.90 53.20/1.20 English 96.40/0.06 95.80/0.07 85.30/0.22 93.80/0.11 94.70/0.08 95.90/0.06 96.50/0.06 91.60/0.12 88.50/0.27 88.00/0.26 70.90/0.70 Estonian 93.30/0.14 87.70/0.33 88.60/0.21 93.80/0.12 67.20/0.65 78.40/0.38 64.10/1.24 81.20/0.84 45.80/2.64 33.30/2.82 42.40/1.35 37.00/3.21 Faroese 79.20/0.42 79.60/0.44 72.90/0.54 76.30/0.52 56.30/0.84 75.90/0.49 76.80/0.39 40.10/1.90 56.60/0.87 56.80/0.91 27.60/1.50 48.40/1.18 Finnish 92.40/0.14 77.00/0.65 73.70/0.43 87.20/0.24 43.80/0.88 78.60/0.34 52.00/2.18 83.30/0.44 23.20/4.90 19.30/5.04 1.40/5.87 14.10/5.66 French 87.90/0.22 85.30/0.27 78.00/0.36 84.30/0.26 72.50/0.48 82.80/0.29 79.60/0.40 51.50/1.60 61.80/0.84 62.20/0.87 60.20/0.91 Friulian 97.00/0.05 97.00/0.08 97.00/0.07 78.00/0.24 85.00/0.17 96.00/0.09 99.00/0.03 88.00/0.26 81.00/0.31 89.00/0.15 95.00/0.09 83.00/0.27 Galician 97.60/0.04 96.70/0.08 94.90/0.09 89.90/0.12 93.90/0.10 95.10/0.09 95.00/0.08 97.50/0.04 86.60/0.23 74.30/0.45 91.10/0.13 62.40/0.78 Georgian 97.90/0.03 95.10/0.13 93.50/0.12 97.80/0.04 95.10/0.06 94.10/0.12 95.40/0.15 96.10/0.06 79.00/0.46 77.60/0.53 80.60/0.41 82.30/0.46 German 87.40/0.24 82.30/0.44 78.90/0.46 37.40/1.11 77.10/0.49 81.00/0.58 82.30/0.43 68.20/0.69 60.50/1.06 52.70/1.44 46.50/1.77 Greek 88.10/0.22 78.20/0.82 81.00/0.45 81.10/0.36 58.90/0.95 78.40/0.40 54.80/1.58 81.40/0.48 40.50/2.64 27.30/3.44 20.70/2.39 31.20/3.22 Haida 96.00/0.05 93.00/0.17 95.00/0.09 96.00/0.06 15.00/1.95 66.00/0.73 100.00/0.00 95.00/0.12 90.00/0.22 77.00/0.69 21.00/3.72 60.00/1.60 Hebrew 98.20/0.02 84.30/0.30 86.10/0.20 85.70/0.15 83.70/0.22 53.70/0.57 54.50/0.70 61.20/0.50 23.10/1.57 28.80/1.45 77.80/0.30 30.40/1.32 Hindi 99.70/0.01 100.00/0.00 99.80/0.00 89.40/0.14 98.70/0.02 93.00/0.08 80.00/0.43 98.80/0.03 65.60/1.43 73.40/1.16 83.50/0.83 2.50/3.01 Hungarian 82.10/0.40 76.90/0.54 78.60/0.47 79.50/0.46 59.20/0.79 69.50/0.68 80.90/0.39 77.30/0.53 64.80/0.84 50.20/1.25 16.20/2.93 38.40/1.99 Icelandic 81.10/0.36 80.90/0.41 72.80/0.54 80.60/0.37 55.20/0.78 77.10/0.46 79.30/0.36 50.90/1.16 63.60/0.69 43.60/1.26 37.00/1.44 46.80/1.24 Irish 89.10/0.33 67.20/1.16 71.20/0.79 81.80/0.48 56.30/1.17 53.00/1.13 34.10/2.88 59.00/1.50 16.90/4.39 14.70/5.28 14.10/3.00 8.50/6.09 Italian 97.30/0.05 94.20/0.16 94.20/0.12 90.00/0.15 88.80/0.17 77.50/0.69 63.70/1.30 96.00/0.07 63.10/0.91 52.30/1.63 32.60/1.73 58.80/1.50 Kabardian 94.00/0.06 99.00/0.01 98.00/0.02 92.00/0.12 100.00/0.00 86.00/0.14 99.00/0.01 89.00/0.13 82.00/0.26 82.00/0.28 96.00/0.05 94.00/0.11 Kannada 96.00/0.08 90.00/0.36 74.00/0.92 99.00/0.01 91.00/0.20 66.00/0.75 97.00/0.03 52.00/1.85 38.00/2.21 36.00/1.77 50.00/2.02 Khaling 93.00/0.11 72.00/0.89 73.50/0.36 44.70/0.82 77.30/0.37 53.70/0.87 17.10/1.80 51.70/1.16 8.50/3.46 15.60/2.53 87.50/0.20 18.40/2.57 Kurmanji 94.50/0.06 92.60/0.12 94.60/0.07 90.40/0.17 94.00/0.08 93.00/0.08 87.80/0.36 93.90/0.07 70.00/0.98 69.40/1.10 66.40/0.99 59.30/1.47 Ladin 96.00/0.08 93.00/0.17 94.00/0.12 86.00/0.23 79.00/0.29 92.00/0.18 87.00/0.18 90.00/0.17 88.00/0.29 84.00/0.28 93.00/0.16 74.00/0.53 Latin 69.80/0.46 46.20/1.28 56.70/0.67 16.20/1.97 18.10/2.05 48.00/0.81 37.20/1.43 51.30/1.06 55.20/0.95 32.20/1.52 9.10/3.15 36.10/1.57 Latvian 96.10/0.06 93.20/0.21 90.70/0.15 92.90/0.11 91.10/0.15 92.80/0.17 90.20/0.29 93.40/0.11 71.50/0.74 66.10/0.85 50.20/1.30 56.50/1.26 Lithuanian 84.80/0.21 70.60/0.68 80.90/0.28 65.20/0.41 41.70/1.11 64.10/0.48 52.00/0.85 79.00/0.41 68.90/0.59 48.10/1.17 36.20/1.68 41.20/1.41 Livonian 92.00/0.16 82.00/0.40 77.00/0.59 87.00/0.31 68.00/0.62 67.00/0.75 76.00/0.60 74.00/0.66 60.00/1.43 50.00/1.56 56.00/0.95 40.00/2.16 Lower-sorbian 94.00/0.11 94.20/0.13 91.30/0.17 93.70/0.13 83.50/0.30 88.30/0.22 95.50/0.09 72.60/0.59 72.70/0.48 70.60/0.50 74.20/0.40 61.20/0.71 Macedonian 95.60/0.07 92.70/0.12 89.10/0.17 91.90/0.10 88.90/0.16 91.20/0.15 94.20/0.10 85.80/0.22 75.80/0.56 65.30/0.70 50.20/0.97 60.50/1.07 Maltese 85.00/0.26 63.00/0.72 66.00/0.61 4.00/3.00 89.00/0.20 16.00/1.85 28.00/1.43 48.00/0.92 21.00/2.13 12.00/2.33 41.00/1.32 8.00/2.62 Middle-french 98.90/0.02 97.00/0.07 96.50/0.07 96.70/0.05 93.50/0.09 95.10/0.10 95.40/0.16 98.50/0.02 90.30/0.26 80.20/0.47 80.90/0.42 79.60/0.49 Navajo 86.50/0.29 43.60/2.19 49.30/1.78 79.00/0.54 25.80/2.25 38.60/2.06 6.80/3.25 37.10/1.93 12.40/3.11 10.70/3.31 27.20/1.80 9.20/3.75 Neapolitan 96.00/0.06 94.00/0.24 95.00/0.14 52.00/0.66 95.00/0.13 95.00/0.09 72.00/0.72 79.00/0.33 73.00/0.41 54.00/1.72 76.00/0.63 North-frisian 90.00/0.21 80.00/0.41 59.00/1.44 15.00/4.44 83.00/0.40 36.00/2.70 33.00/2.23 45.00/3.17 17.00/3.91 16.00/3.64 7.00/4.53 Northern-sami 96.30/0.08 61.70/1.17 74.30/0.44 90.70/0.15 47.50/1.16 62.70/0.70 75.50/0.35 69.50/1.05 67.10/0.60 51.70/1.00 47.20/1.29 46.40/1.27 Norwegian-bokmaal 85.90/0.20 90.80/0.15 79.20/0.31 88.90/0.19 81.70/0.28 90.50/0.17 87.20/0.20 77.80/0.34 72.20/0.57 70.90/0.53 49.80/1.13 66.50/0.68 Norwegian-nynorsk 82.40/0.28 82.80/0.30 70.40/0.55 79.40/0.34 56.60/0.71 74.70/0.42 88.00/0.20 57.00/0.81 76.60/0.43 50.00/1.03 39.00/1.35 45.10/0.97 Occitan 97.00/0.05 94.00/0.15 95.00/0.07 83.00/0.27 83.00/0.17 96.00/0.07 92.00/0.09 94.00/0.17 86.00/0.21 85.00/0.25 96.00/0.06 75.00/0.63 Old-armenian 88.00/0.22 84.90/0.33 82.80/0.35 80.40/0.36 68.00/0.62 78.90/0.46 82.20/0.36 58.40/1.10 57.80/0.91 64.70/0.79 62.60/0.68 58.90/0.98 Old-church-slavonic 82.00/0.35 92.00/0.15 93.00/0.15 9.00/2.10 93.00/0.10 81.00/0.45 88.00/0.16 52.00/1.20 67.00/0.58 66.00/0.59 33.00/0.97 71.00/0.47 Old-english 84.20/0.27 69.30/0.59 65.40/0.55 28.20/1.18 50.90/0.84 40.60/0.92 34.30/1.30 63.90/0.77 58.30/0.86 56.60/0.94 43.80/1.23 Old-french 89.80/0.18 80.80/0.48 82.40/0.37 61.80/0.78 80.80/0.40 82.00/0.39 55.10/1.26 75.90/0.51 62.90/0.81 54.10/0.88 57.00/0.87 Old-saxon 94.50/0.10 87.30/0.28 77.70/0.39 54.50/0.66 72.30/0.46 59.90/0.67 54.00/0.67 64.30/0.94 72.80/0.49 68.70/0.53 76.70/0.33 42.90/1.24 Pashto 89.00/0.19 92.00/0.11 82.00/0.47 84.00/0.25 87.00/0.16 71.00/0.63 78.00/0.42 56.00/1.41 31.00/1.63 37.00/1.46 82.00/0.29 30.00/1.81 Persian 98.20/0.03 63.70/1.50 90.80/0.20 95.60/0.08 90.80/0.17 80.70/0.53 62.60/1.38 98.20/0.03 56.60/2.63 50.50/2.48 29.30/3.33 3.70/4.30 Polish 88.60/0.23 87.60/0.33 82.40/0.37 87.80/0.26 76.40/0.45 87.00/0.26 82.90/0.56 80.40/0.62 61.60/1.33 44.60/1.82 49.00/1.78 Portuguese 96.90/0.05 97.30/0.06 95.10/0.08 84.00/0.23 87.80/0.14 96.60/0.06 94.60/0.08 98.50/0.02 88.30/0.19 77.40/0.41 76.40/0.42 55.50/0.96 Quechua 99.00/0.02 99.80/0.00 99.00/0.02 96.80/0.05 81.00/0.32 95.10/0.10 98.80/0.04 99.30/0.02 91.30/0.20 9.70/3.79 43.60/12.02 78.40/0.56 Romanian 81.20/0.49 82.60/0.53 77.70/0.58 82.00/0.49 69.20/0.75 79.70/0.54 62.40/1.33 56.40/1.87 44.90/2.34 41.40/2.65 40.40/1.83 33.80/2.76 Russian 91.90/0.20 88.00/0.33 82.20/0.45 86.80/0.40 85.20/0.35 86.50/0.44 76.10/0.70 86.70/0.62 63.80/1.38 53.20/1.64 46.00/2.20 Sanskrit 89.70/0.16 92.80/0.12 87.50/0.22 76.60/0.38 79.20/0.36 80.40/0.35 93.70/0.09 50.40/1.43 70.70/0.65 71.70/0.51 53.80/1.31 74.40/0.48 Serbo-croatian 87.60/0.24 87.40/0.29 84.50/0.32 88.00/0.27 75.20/0.46 83.10/0.36 69.10/1.13 61.90/1.50 35.50/2.94 44.70/2.54 26.20/2.35 33.20/2.84 Slovak 92.70/0.12 91.50/0.13 82.90/0.29 88.80/0.17 73.90/0.44 83.30/0.28 91.10/0.14 51.40/1.25 67.60/0.54 66.00/0.55 61.60/0.67 59.20/0.68 Slovene 33.00/0.79 7.00/1.61 17.80/1.18 93.40/0.10 84.80/0.24 84.90/0.24 90.90/0.16 40.80/0.74 88.30/0.21 72.20/0.49 0.70/8.23 33.90/1.66 Sorani 87.90/0.15 76.00/0.53 80.90/0.27 86.80/0.18 59.30/0.60 63.60/0.68 27.60/2.22 53.20/1.49 12.60/3.06 16.40/2.83 28.50/1.81 12.70/2.95 Spanish 96.80/0.05 94.40/0.17 93.10/0.13 72.70/0.34 93.60/0.11 92.40/0.20 81.30/0.64 95.60/0.07 72.80/0.93 64.90/1.27 59.30/0.85 46.80/1.44 Swahili 98.00/0.02 98.00/0.06 99.00/0.01 98.00/0.02 98.00/0.02 71.00/0.44 1.00/3.15 94.00/0.09 5.00/3.22 5.00/2.74 88.00/0.19 4.00/2.73 Swedish 89.50/0.18 88.00/0.22 59.90/0.80 85.20/0.27 75.60/0.42 85.00/0.27 88.50/0.23 84.50/0.27 72.70/0.61 60.70/0.94 33.10/2.01 49.50/1.34 Tatar 97.00/0.03 99.00/0.01 99.00/0.01 95.00/0.06 97.00/0.03 96.00/0.04 97.00/0.03 99.00/0.01 97.00/0.03 86.00/0.17 45.00/1.73 93.00/0.08 Turkish 97.00/0.04 87.90/0.26 93.90/0.09 79.30/0.44 75.40/0.53 73.20/0.70 95.90/0.08 97.30/0.04 77.20/0.54 59.50/1.14 23.30/2.88 51.50/1.62 Ukrainian 91.60/0.15 93.10/0.15 83.70/0.30 90.50/0.20 76.10/0.41 86.20/0.29 87.20/0.25 82.50/0.48 76.70/0.44 62.80/0.75 36.10/1.36 65.30/0.71 Urdu 98.40/0.03 99.30/0.01 99.10/0.02 43.50/0.78 99.00/0.01 95.90/0.05 83.70/0.38 95.10/0.21 74.40/1.11 72.70/1.10 97.00/0.04 5.90/3.60 Uzbek 96.00/0.10 100.00/0.00 100.00/0.00 100.00/0.00 96.00/0.04 99.00/0.01 44.00/2.20 61.00/0.66 61.00/0.81 8.00/2.87 76.00/0.51 Venetian 97.90/0.03 98.50/0.03 97.60/0.05 98.50/0.03 98.20/0.03 93.10/0.11 97.60/0.04 98.40/0.03 89.20/0.18 87.10/0.20 87.60/0.46 77.80/0.43 Votic 77.00/0.37 37.00/1.37 52.00/0.85 41.00/0.83 37.00/1.35 66.00/0.47 12.00/2.46 48.00/0.88 49.00/0.90 23.00/1.53 34.00/1.58 Welsh 92.00/0.16 82.00/0.47 79.00/0.43 92.00/0.13 72.00/0.47 72.00/0.54 88.00/0.16 82.00/0.44 75.00/0.41 85.00/0.29 90.00/0.15 52.00/0.96 West-frisian 84.00/0.27 82.00/0.51 72.00/0.70 92.00/0.10 67.00/0.78 67.00/0.56 47.00/1.17 54.00/1.02 45.00/1.07 17.00/1.73 33.00/1.52 Yiddish 95.00/0.05 97.00/0.10 98.00/0.07 92.00/0.19 96.00/0.07 94.00/0.14 99.00/0.01 94.00/0.13 94.00/0.06 84.00/0.31 80.00/0.29 86.00/0.29 Zulu 96.90/0.06 93.80/0.16 95.60/0.11 73.30/0.40 98.50/0.02 68.30/0.64 2.50/3.03 87.60/0.28 2.70/2.88 2.50/3.63 72.30/0.62 2.40/3.55

Table 14: Task 1 High Condition Part 2.
{adjustbox}

width=0.98 oracle-fc oracle-e uzh-01 uzh-02 iitbhu-iiith-02 iitbhu-iiith-01 waseda-01 msu-04 msu-03 hamburg-01 iit-varanasi-01 msu-02 Adyghe 100.00/* 97.20/* 94.40/0.06 94.20/0.06 93.40/0.07 93.40/0.07 87.80/0.16 80.60/0.46 93.50/0.07 92.10/0.08 92.90/0.08 94.00/0.06 Albanian 100.00/* 95.00/* 88.10/0.30 87.20/0.33 88.80/0.24 87.30/0.27 72.70/1.29 78.80/0.54 77.20/0.58 79.00/0.59 39.20/1.63 72.80/1.01 Arabic 98.40/* 86.90/* 79.40/0.65 78.60/0.68 78.30/0.71 74.40/0.81 66.70/1.02 0.00/10.73 0.00/9.77 63.80/1.03 38.80/1.90 0.00/4.70 Armenian 97.00/* 97.10/* 92.80/0.13 92.40/0.13 90.80/0.16 88.70/0.19 85.30/0.26 86.80/0.23 86.80/0.24 88.20/0.19 67.40/0.80 80.50/0.47 Asturian 97.10/* 96.20/* 91.60/0.17 92.00/0.17 92.40/0.17 91.80/0.18 91.80/0.18 90.30/0.19 89.40/0.22 91.70/0.19 90.00/0.20 88.40/0.27 Azeri 99.00/* 99.00/* 95.00/0.11 94.00/0.15 96.00/0.08 92.00/0.14 84.00/0.34 96.00/0.11 92.00/0.16 89.00/0.19 94.00/0.14 80.00/0.37 Bashkir 100.00/* 99.30/* 96.70/0.08 97.30/0.06 96.50/0.08 95.00/0.11 87.20/0.24 92.00/0.22 92.40/0.24 94.10/0.13 95.40/0.09 88.30/0.25 Basque 43.90/* 95.50/* 88.00/0.23 86.00/0.27 88.10/0.25 86.50/0.28 80.50/0.45 59.50/0.91 51.60/1.11 33.10/1.66 69.20/0.58 2.70/6.01 Belarusian 100.00/* 84.50/* 70.40/0.97 69.50/0.97 64.30/1.23 63.20/1.23 53.60/1.44 61.30/1.47 56.40/1.50 54.00/1.45 45.80/1.71 59.50/1.41 Bengali 100.00/* 99.00/* 99.00/0.05 99.00/0.05 97.00/0.11 96.00/0.12 96.00/0.14 88.00/0.29 95.00/0.15 92.00/0.22 97.00/0.10 84.00/0.31 Breton 93.00/* 97.00/* 95.00/0.11 96.00/0.10 92.00/0.25 90.00/0.29 93.00/0.17 93.00/0.18 91.00/0.14 88.00/0.22 94.00/0.20 77.00/0.80 Bulgarian 100.00/* 92.70/* 83.30/0.26 83.80/0.25 81.00/0.29 78.00/0.33 80.80/0.31 76.90/0.37 77.80/0.33 78.80/0.35 35.00/1.36 72.20/0.48 Catalan 100.00/* 96.80/* 92.10/0.15 92.20/0.15 92.10/0.16 91.80/0.17 92.80/0.17 90.00/0.21 90.50/0.19 84.40/0.27 83.80/0.28 85.40/0.34 Classical-syriac 100.00/* 100.00/* 100.00/0.00 100.00/0.00 99.00/0.01 99.00/0.01 100.00/0.00 99.00/0.01 96.00/0.04 97.00/0.03 99.00/0.01 97.00/0.04 Cornish 92.00/* 88.00/* 70.00/0.74 66.00/0.76 70.00/0.76 70.00/0.74 46.00/2.44 58.00/0.86 36.00/1.62 52.00/1.00 66.00/0.70 10.00/2.78 Crimean-tatar 100.00/* 99.00/* 98.00/0.04 98.00/0.04 97.00/0.06 97.00/0.06 92.00/0.11 91.00/0.21 93.00/0.12 96.00/0.06 98.00/0.04 95.00/0.07 Czech 97.20/* 93.80/* 87.20/0.26 86.60/0.28 80.40/0.37 79.00/0.39 82.30/0.46 79.40/0.42 78.60/0.47 82.00/0.38 61.10/0.94 76.30/0.53 Danish 100.00/* 93.60/* 80.40/0.30 80.00/0.31 79.20/0.33 78.60/0.34 79.60/0.32 78.80/0.34 76.10/0.37 79.00/0.32 75.70/0.36 68.80/0.50 Dutch 100.00/* 96.20/* 85.70/0.22 85.60/0.21 82.00/0.28 81.90/0.28 77.10/0.37 80.40/0.31 84.00/0.25 77.80/0.38 71.00/0.49 65.90/0.52 English 100.00/* 97.20/* 94.50/0.10 94.20/0.10 94.20/0.10 93.50/0.11 91.50/0.14 92.70/0.13 94.40/0.09 90.20/0.14 90.90/0.15 86.40/0.25 Estonian 100.00/* 90.80/* 81.60/0.30 81.50/0.31 78.20/0.34 76.20/0.37 72.20/0.50 71.30/0.51 65.60/0.63 63.10/0.77 46.50/1.20 67.50/0.54 Faroese 100.00/* 86.60/* 72.60/0.59 71.70/0.60 69.10/0.61 68.00/0.62 68.20/0.72 68.00/0.68 66.20/0.71 56.10/0.84 48.90/1.02 60.10/0.78 Finnish 97.10/* 91.50/* 82.80/0.27 82.30/0.29 73.80/0.43 71.10/0.47 65.10/0.64 58.10/0.83 57.00/0.84 35.90/1.22 22.00/2.55 41.90/1.98 French 100.00/* 89.40/* 80.20/0.33 80.90/0.32 78.60/0.35 78.20/0.35 73.10/0.47 75.70/0.42 76.50/0.41 71.80/0.50 69.30/0.60 73.90/0.46 Friulian 100.00/* 99.00/* 96.00/0.06 96.00/0.06 97.00/0.05 97.00/0.05 96.00/0.06 95.00/0.07 93.00/0.09 81.00/0.24 92.00/0.14 92.00/0.12 Galician 99.90/* 95.70/* 90.80/0.16 90.40/0.17 88.90/0.18 88.40/0.18 88.90/0.18 85.80/0.27 82.90/0.29 86.00/0.23 82.50/0.29 82.80/0.33 Georgian 96.20/* 96.00/* 93.90/0.14 94.00/0.14 93.50/0.17 93.40/0.17 93.20/0.17 91.20/0.21 92.30/0.20 92.60/0.17 91.20/0.22 84.40/0.29 German 100.00/* 89.20/* 80.10/0.48 79.50/0.48 76.80/0.56 76.70/0.56 74.00/0.64 74.10/0.60 74.40/0.56 70.90/0.65 65.50/0.71 68.00/0.72 Greek 97.40/* 84.40/* 75.30/0.53 75.50/0.55 71.50/0.56 69.10/0.61 60.50/1.73 67.00/0.71 62.00/0.83 58.40/0.98 29.90/1.99 62.00/1.65 Greenlandic 100.00/* 100.00/* 98.00/0.02 98.00/0.02 88.00/0.20 86.00/0.22 74.00/2.40 68.00/0.38 80.00/0.22 68.00/0.40 84.00/0.22 70.00/0.38 Haida 100.00/* 96.00/* 94.00/0.12 94.00/0.12 87.00/0.32 85.00/0.36 62.00/1.98 91.00/0.19 92.00/0.20 16.00/2.15 90.00/0.19 81.00/0.38 Hebrew 100.00/* 96.10/* 85.40/0.19 84.50/0.20 84.50/0.21 83.10/0.23 76.90/0.32 81.70/0.25 80.10/0.26 77.90/0.30 64.70/0.52 64.40/0.47 Hindi 99.70/* 98.90/* 97.50/0.03 97.60/0.03 96.20/0.07 96.00/0.08 95.60/0.05 94.80/0.07 95.00/0.07 94.20/0.07 87.70/0.32 90.90/0.16 Hungarian 100.00/* 86.50/* 73.70/0.53 73.40/0.55 74.50/0.51 72.00/0.56 53.80/0.96 65.90/0.68 68.80/0.65 51.60/0.91 57.10/0.85 64.50/0.65 Icelandic 100.00/* 88.00/* 73.80/0.52 73.60/0.52 65.90/0.67 63.50/0.70 62.40/0.78 64.60/0.69 59.20/0.78 54.60/0.79 45.70/1.03 55.70/0.87 Ingrian 100.00/* 96.00/* 92.00/0.14 92.00/0.12 90.00/0.24 90.00/0.18 68.00/1.56 88.00/0.20 86.00/0.22 58.00/0.66 88.00/0.18 56.00/0.68 Irish 99.10/* 87.30/* 77.10/0.67 75.60/0.70 68.80/0.89 64.70/0.97 54.30/1.22 60.80/1.12 57.10/1.19 59.60/1.13 18.80/3.40 52.20/1.43 Italian 100.00/* 97.60/* 94.90/0.09 95.10/0.09 93.20/0.13 91.70/0.15 91.90/0.16 88.90/0.19 92.50/0.16 85.90/0.25 68.90/0.78 85.80/0.32 Kabardian 100.00/* 100.00/* 97.00/0.03 98.00/0.02 98.00/0.02 98.00/0.02 92.00/0.11 86.00/0.27 92.00/0.10 95.00/0.06 97.00/0.03 97.00/0.03 Kannada 100.00/* 99.00/* 94.00/0.12 94.00/0.12 92.00/0.14 91.00/0.15 93.00/0.11 87.00/0.26 87.00/0.23 86.00/0.24 85.00/0.26 64.00/1.19 Karelian 100.00/* 100.00/* 100.00/0.00 100.00/0.00 96.00/0.08 94.00/0.08 78.00/1.92 92.00/0.08 92.00/0.12 82.00/0.28 96.00/0.06 76.00/0.28 Kashubian 100.00/* 96.00/* 88.00/0.20 88.00/0.26 86.00/0.22 82.00/0.26 66.00/1.96 84.00/0.26 78.00/0.38 82.00/0.34 84.00/0.24 72.00/0.48 Kazakh 100.00/* 96.00/* 84.00/0.16 82.00/0.18 88.00/0.12 84.00/0.16 70.00/1.76 76.00/0.26 74.00/0.28 86.00/0.16 50.00/0.50 82.00/0.20 Khakas 100.00/* 98.00/* 98.00/0.04 98.00/0.04 98.00/0.04 98.00/0.04 64.00/1.54 98.00/0.04 98.00/0.02 96.00/0.06 98.00/0.04 98.00/0.04 Khaling 90.90/* 94.40/* 83.20/0.29 82.40/0.30 85.60/0.23 86.00/0.23 85.70/0.25 41.40/1.18 43.40/1.23 68.30/0.65 59.40/0.80 20.10/2.61 Kurmanji 97.20/* 97.30/* 93.20/0.10 92.30/0.11 91.10/0.18 90.60/0.19 88.30/0.32 74.20/0.55 87.20/0.21 90.00/0.13 86.10/0.34 91.00/0.22 Ladin 97.00/* 98.00/* 93.00/0.16 95.00/0.11 92.00/0.13 93.00/0.11 93.00/0.13 80.00/0.27 87.00/0.21 76.00/0.26 91.00/0.10 89.00/0.29 Latin 100.00/* 74.30/* 53.30/0.75 51.70/0.80 46.20/0.89 44.80/0.94 37.90/1.16 38.20/1.17 40.80/1.10 21.40/1.94 31.30/1.37 37.50/1.26 Latvian 99.60/* 94.90/* 90.60/0.15 89.80/0.19 88.00/0.24 86.50/0.27 88.20/0.26 85.30/0.31 84.40/0.31 86.30/0.26 73.70/0.55 86.50/0.25 Lithuanian 98.90/* 78.80/* 63.90/0.52 63.00/0.55 55.60/0.75 53.40/0.78 52.00/0.70 45.90/0.88 48.10/0.87 40.80/1.13 46.80/0.96 55.00/0.77 Livonian 99.00/* 94.00/* 82.00/0.35 79.00/0.42 75.00/0.45 77.00/0.40 74.00/0.50 67.00/0.61 63.00/0.68 70.00/0.59 62.00/0.75 54.00/1.14 Lower-sorbian 100.00/* 95.00/* 85.10/0.26 84.20/0.29 84.00/0.28 82.30/0.30 81.40/0.34 78.20/0.39 80.80/0.35 78.60/0.40 69.80/0.50 78.90/0.39 Macedonian 98.40/* 97.20/* 91.60/0.11 91.50/0.12 90.10/0.13 88.50/0.15 88.10/0.16 86.40/0.19 84.30/0.21 87.40/0.17 75.10/0.38 84.20/0.24 Maltese 99.00/* 98.00/* 95.00/0.10 94.00/0.11 91.00/0.20 90.00/0.19 89.00/0.22 77.00/0.35 84.00/0.25 82.00/0.38 85.00/0.24 59.00/0.73 Mapudungun 100.00/* 100.00/* 98.00/0.04 100.00/0.00 96.00/0.04 96.00/0.04 76.00/1.46 90.00/0.12 88.00/0.14 100.00/0.00 98.00/0.04 92.00/0.08 Middle-french 99.80/* 97.50/* 94.30/0.11 94.50/0.12 93.30/0.14 92.60/0.16 93.20/0.13 92.90/0.15 92.70/0.15 89.00/0.19 89.40/0.23 90.80/0.19 Middle-high-german 100.00/* 100.00/* 100.00/0.00 100.00/0.00 96.00/0.08 96.00/0.08 80.00/1.20 92.00/0.12 92.00/0.14 96.00/0.10 92.00/0.14 90.00/0.14 Middle-low-german 100.00/* 100.00/* 98.00/0.02 98.00/0.02 98.00/0.02 100.00/0.00 72.00/1.50 94.00/0.06 82.00/0.36 96.00/0.04 92.00/0.16 68.00/0.78 Murrinhpatha 100.00/* 100.00/* 96.00/0.04 96.00/0.04 90.00/0.20 86.00/0.22 70.00/1.34 76.00/0.52 78.00/0.50 88.00/0.22 96.00/0.10 44.00/1.26 Navajo 99.50/* 69.60/* 54.30/1.20 54.00/1.27 44.60/1.60 43.00/1.66 40.80/1.81 42.40/1.63 42.00/1.61 24.20/2.74 19.50/2.68 33.80/2.65 Neapolitan 99.00/* 99.00/* 99.00/0.03 99.00/0.03 98.00/0.05 98.00/0.05 95.00/0.08 96.00/0.06 93.00/0.11 49.00/0.71 98.00/0.06 94.00/0.20 Norman 100.00/* 94.00/* 84.00/0.38 86.00/0.34 88.00/0.24 88.00/0.24 66.00/2.08 38.00/1.10 34.00/1.28 58.00/0.94 26.00/1.62 46.00/2.04 North-frisian 100.00/* 94.00/* 91.00/0.24 89.00/0.27 88.00/0.29 89.00/0.27 87.00/0.30 78.00/0.67 70.00/0.74 75.00/0.61 85.00/0.39 49.00/1.98 Northern-sami 100.00/* 90.30/* 76.10/0.42 75.20/0.44 70.20/0.52 66.10/0.58 60.30/0.78 56.70/0.85 57.80/0.81 31.30/1.66 39.40/1.26 34.60/1.30 Norwegian-bokmaal 99.30/* 93.60/* 84.10/0.24 84.00/0.24 83.50/0.27 81.50/0.30 81.30/0.30 80.40/0.31 82.70/0.28 79.00/0.34 81.50/0.29 66.60/0.47 Norwegian-nynorsk 99.80/* 90.00/* 67.10/0.55 65.90/0.57 64.20/0.60 62.90/0.63 60.50/0.65 61.10/0.65 61.00/0.63 53.90/0.77 57.50/0.72 49.30/0.88 Occitan 100.00/* 97.00/* 94.00/0.10 95.00/0.11 94.00/0.10 94.00/0.10 96.00/0.10 92.00/0.12 89.00/0.22 81.00/0.22 93.00/0.12 89.00/0.23 Old-armenian 99.00/* 91.80/* 80.20/0.39 79.20/0.41 78.10/0.44 77.10/0.46 72.70/0.58 66.00/0.67 69.70/0.61 65.90/0.66 57.30/0.82 71.40/0.63 Old-church-slavonic 100.00/* 99.00/* 93.00/0.11 93.00/0.11 87.00/0.24 87.00/0.23 90.00/0.16 83.00/0.29 78.00/0.34 89.00/0.18 87.00/0.19 80.00/0.35 Old-english 100.00/* 83.90/* 65.60/0.58 65.00/0.58 62.70/0.64 60.30/0.69 56.30/0.72 52.90/0.77 58.30/0.70 45.70/0.93 42.40/1.13 33.50/1.20 Old-french 95.90/* 89.30/* 79.30/0.41 77.90/0.45 74.10/0.51 72.20/0.54 71.30/0.58 68.10/0.59 68.50/0.62 62.60/0.72 60.60/0.74 66.60/0.73 Old-irish 84.00/* 50.00/* 40.00/1.66 34.00/2.14 26.00/2.58 22.00/2.78 16.00/3.40 12.00/2.62 26.00/2.42 6.00/3.78 24.00/2.52 16.00/3.88 Old-saxon 97.60/* 92.40/* 80.70/0.34 80.90/0.33 74.90/0.39 74.10/0.41 70.30/0.49 70.90/0.49 71.10/0.47 67.40/0.55 64.70/0.60 47.60/1.04 Pashto 100.00/* 97.00/* 85.00/0.31 83.00/0.28 85.00/0.23 81.00/0.28 77.00/0.50 73.00/0.55 75.00/0.49 80.00/0.31 73.00/0.36 68.00/0.72 Persian 100.00/* 97.20/* 93.40/0.10 93.40/0.10 91.70/0.13 90.70/0.14 87.00/0.24 84.50/0.24 86.20/0.23 82.60/0.37 70.90/0.71 77.30/0.63 Polish 98.20/* 89.50/* 82.20/0.40 82.40/0.38 76.10/0.54 73.40/0.58 75.90/0.55 72.30/0.58 74.50/0.56 75.10/0.58 55.10/1.09 74.10/0.61 Portuguese 100.00/* 97.20/* 94.70/0.08 94.80/0.08 92.00/0.12 92.00/0.12 93.80/0.10 90.40/0.14 91.30/0.13 84.90/0.21 37.30/1.41 90.40/0.18 Quechua 85.90/* 99.60/* 98.90/0.02 98.70/0.03 98.60/0.03 98.20/0.03 70.90/1.49 96.70/0.06 95.30/0.11 78.40/0.40 77.30/0.55 73.30/1.45 Romanian 100.00/* 88.20/* 77.60/0.57 77.10/0.58 73.00/0.77 72.20/0.83 71.50/0.89 68.30/0.73 68.40/0.79 68.70/0.73 48.50/1.51 68.80/0.73 Russian 99.70/* 92.70/* 86.90/0.30 86.80/0.28 80.10/0.57 78.20/0.60 79.50/0.56 76.40/0.50 76.20/0.50 77.30/0.47 59.90/0.97 72.20/0.77 Sanskrit 99.30/* 93.20/* 85.90/0.23 85.90/0.23 85.50/0.25 84.70/0.26 79.10/0.36 78.50/0.36 77.70/0.39 79.80/0.33 68.80/0.54 79.60/0.36 Scottish-gaelic 100.00/* 96.00/* 92.00/0.10 92.00/0.14 94.00/0.12 94.00/0.08 72.00/1.66 84.00/0.28 72.00/0.50 88.00/0.18 86.00/0.18 88.00/0.24 Serbo-croatian 95.10/* 93.00/* 86.10/0.27 85.20/0.27 82.50/0.34 81.90/0.36 76.30/0.45 75.50/0.47 73.30/0.54 72.30/0.58 43.40/1.49 66.80/0.83 Slovak 100.00/* 92.40/* 78.60/0.37 77.90/0.39 73.90/0.45 72.30/0.47 73.90/0.47 71.00/0.46 68.90/0.50 70.30/0.51 58.70/0.67 68.40/0.54 Slovene 99.80/* 94.30/* 86.20/0.22 85.80/0.23 85.00/0.25 83.60/0.28 84.10/0.24 40.90/0.79 27.10/0.99 84.70/0.26 65.60/0.52 7.60/1.51 Sorani 97.40/* 93.00/* 79.60/0.32 79.20/0.33 80.20/0.30 77.80/0.34 72.00/0.49 64.20/0.61 64.40/0.61 49.30/0.81 55.90/0.90 56.40/1.19 Spanish 100.00/* 95.60/* 91.40/0.14 91.10/0.15 92.00/0.14 91.20/0.16 91.50/0.15 88.60/0.19 89.30/0.18 88.30/0.23 75.70/0.64 85.10/0.40 Swahili 100.00/* 99.00/* 99.00/0.01 99.00/0.01 97.00/0.05 95.00/0.07 92.00/0.14 85.00/0.22 84.00/0.28 95.00/0.09 87.00/0.29 92.00/0.13 Swedish 99.40/* 93.10/* 79.80/0.35 79.50/0.35 77.80/0.40 78.30/0.40 76.80/0.41 77.80/0.37 74.30/0.43 71.90/0.48 31.40/1.41 65.40/0.57 Tatar 100.00/* 99.00/* 97.00/0.03 98.00/0.02 95.00/0.05 95.00/0.05 92.00/0.11 94.00/0.06 95.00/0.07 93.00/0.09 96.00/0.04 94.00/0.07 Tibetan 100.00/* 72.00/* 52.00/0.78 56.00/0.64 50.00/0.90 40.00/1.00 38.00/1.36 44.00/0.86 44.00/0.94 46.00/0.74 22.00/1.36 52.00/0.80 Turkish 94.10/* 95.90/* 90.70/0.17 90.30/0.17 90.00/0.27 88.20/0.31 69.20/0.78 82.50/0.39 81.20/0.45 68.60/0.70 79.20/0.47 65.40/1.21 Turkmen 100.00/* 98.00/* 94.00/0.08 94.00/0.08 98.00/0.02 98.00/0.02 74.00/1.72 94.00/0.06 94.00/0.06 90.00/0.18 96.00/0.04 98.00/0.02 Ukrainian 99.30/* 93.50/* 81.40/0.35 80.40/0.38 77.60/0.42 76.00/0.43 74.00/0.49 73.00/0.46 71.70/0.47 77.50/0.41 46.60/0.85 69.80/0.50 Urdu 98.70/* 98.70/* 96.70/0.04 96.80/0.04 96.70/0.04 96.40/0.05 93.90/0.10 94.10/0.10 92.20/0.14 94.70/0.07 87.50/0.25 89.40/0.20 Uzbek 100.00/* 100.00/* 100.00/0.00 100.00/0.00 100.00/0.00 100.00/0.00 96.00/0.04 100.00/0.00 94.00/0.10 100.00/0.00 100.00/0.00 100.00/0.00 Venetian 99.30/* 98.10/* 94.40/0.08 95.10/0.08 92.80/0.10 92.30/0.11 91.80/0.10 93.10/0.10 91.60/0.11 92.10/0.11 91.90/0.12 88.50/0.17 Votic 100.00/* 96.00/* 86.00/0.16 88.00/0.14 86.00/0.18 83.00/0.22 74.00/0.43 67.00/0.53 71.00/0.44 42.00/0.82 78.00/0.32 49.00/0.90 Welsh 100.00/* 90.00/* 84.00/0.29 83.00/0.30 82.00/0.30 82.00/0.30 81.00/0.33 77.00/0.39 71.00/0.48 67.00/0.63 81.00/0.34 67.00/0.61 West-frisian 100.00/* 100.00/* 97.00/0.05 98.00/0.04 89.00/0.25 86.00/0.30 96.00/0.14 90.00/0.11 74.00/0.43 91.00/0.21 94.00/0.07 73.00/0.68 Yiddish 100.00/* 99.00/* 92.00/0.10 94.00/0.10 89.00/0.16 89.00/0.16 91.00/0.18 85.00/0.19 89.00/0.15 92.00/0.13 88.00/0.23 89.00/0.18 Zulu 96.90/* 94.60/* 87.20/0.28 87.30/0.27 85.20/0.35 82.80/0.40 82.80/0.38 70.40/0.68 68.40/0.70 81.80/0.39 52.60/1.14 71.60/0.70

Table 15: Task 1 Medium Condition Part 1.
{adjustbox}

width= bme-01 bme-03 bme-02 msu-01 baseline axsemantics-02 axsemantics-01 kucst-01 tuebingen-oslo-03 tuebingen-oslo-02 tuebingen-oslo-01 Adyghe 90.80/0.12 90.50/0.13 93.10/0.11 88.90/0.13 84.80/0.15 91.90/0.08 66.10/0.86 55.00/1.28 34.60/1.96 47.10/1.70 Albanian 46.20/1.41 46.20/1.41 46.30/1.38 65.50/1.01 60.70/1.44 26.50/3.21 10.50/3.78 3.30/6.25 3.40/6.60 2.00/7.59 Arabic 47.00/1.63 47.00/1.63 37.20/2.76 0.00/7.80 39.50/1.83 22.60/3.23 0.00/10.78 5.10/3.63 22.30/2.85 0.00/8.80 Armenian 58.70/1.00 58.70/1.00 58.00/1.13 77.40/0.53 71.00/0.55 63.40/1.06 12.30/3.18 9.80/3.67 8.30/4.11 8.00/4.29 Asturian 87.80/0.22 88.60/0.22 86.40/0.27 89.60/0.21 89.10/0.25 82.50/0.35 49.70/1.01 40.10/1.26 51.80/0.89 26.20/1.68 Azeri 56.00/0.89 56.00/0.89 65.00/0.83 81.00/0.40 50.00/1.91 89.00/0.20 11.00/2.36 54.00/1.10 46.00/1.34 34.00/1.74 Bashkir 92.40/0.14 91.50/0.17 93.20/0.21 93.70/0.13 72.30/0.66 93.90/0.12 65.40/0.84 59.40/1.03 51.40/1.17 45.20/1.53 Basque 37.70/1.38 37.70/1.38 34.20/1.60 10.40/3.73 1.80/5.63 0.70/3.17 49.40/1.17 1.40/4.21 2.70/3.82 2.40/4.11 Belarusian 52.40/1.41 52.40/1.41 45.40/1.69 41.60/2.03 21.70/2.09 16.60/2.82 13.00/2.84 26.90/2.59 20.60/2.82 15.40/3.25 Bengali 87.00/0.32 87.00/0.32 85.00/0.39 83.00/0.37 76.00/0.33 72.00/0.37 66.00/0.82 43.00/1.04 45.00/1.09 46.00/1.04 Breton 88.00/0.33 90.00/0.26 93.00/0.20 70.00/0.69 67.00/1.09 81.00/0.33 66.00/0.67 42.00/1.21 55.00/0.71 30.00/1.59 Bulgarian 58.00/0.71 58.00/0.71 59.70/0.86 75.80/0.36 70.60/0.49 52.60/1.27 32.10/2.05 22.90/2.80 17.10/3.20 20.60/3.11 Catalan 84.50/0.29 84.20/0.33 88.10/0.21 86.10/0.27 85.70/0.31 79.50/0.37 23.30/1.99 37.30/1.47 43.40/1.28 14.10/2.36 Classical-syriac 98.00/0.03 98.00/0.03 96.00/0.05 91.00/0.09 99.00/0.01 100.00/0.00 81.00/0.26 67.00/0.45 71.00/0.30 90.00/0.11 Cornish 20.00/2.20 28.00/1.96 32.00/1.80 12.00/2.94 32.00/1.54 2.00/3.54 34.00/1.70 20.00/1.76 26.00/1.70 10.00/3.46 Crimean-tatar 94.00/0.11 92.00/0.11 94.00/0.19 90.00/0.19 78.00/0.31 97.00/0.05 61.00/0.93 50.00/1.31 52.00/1.17 58.00/1.11 Czech 35.50/1.47 35.70/1.73 32.90/1.69 69.00/0.70 79.70/0.48 61.10/1.09 7.20/4.19 10.50/3.61 17.40/3.13 8.60/4.20 Danish 72.10/0.46 71.90/0.45 71.50/0.56 70.90/0.47 77.60/0.37 72.00/0.41 31.30/1.97 24.00/2.65 30.80/2.39 27.90/2.21 Dutch 66.40/0.53 66.40/0.53 66.10/0.72 73.90/0.42 72.70/0.45 61.20/0.69 16.30/2.73 25.60/2.21 14.30/2.85 English 90.40/0.15 90.40/0.15 89.60/0.16 92.60/0.11 90.50/0.15 89.00/0.15 29.30/2.05 28.10/2.32 26.10/2.14 Estonian 31.70/1.57 31.70/1.57 27.00/1.98 59.90/0.81 62.70/0.77 36.10/1.82 8.30/3.48 14.60/3.86 3.40/4.84 3.70/5.09 Faroese 48.90/1.04 46.80/1.11 50.70/1.05 45.80/1.28 65.30/0.77 47.00/1.01 12.50/2.67 13.10/2.72 19.80/2.38 13.80/2.64 Finnish 31.70/1.88 31.70/1.88 26.20/2.34 51.20/1.04 44.20/1.53 21.40/3.01 0.00/8.23 0.20/7.52 1.20/7.45 1.10/8.10 French 73.40/0.47 73.40/0.47 74.60/0.45 74.60/0.46 73.10/0.47 66.90/0.66 14.50/2.58 29.80/2.04 22.30/2.39 Friulian 91.00/0.11 91.00/0.11 91.00/0.12 56.00/0.93 92.00/0.11 94.00/0.11 74.00/0.47 62.00/0.58 66.00/0.50 48.00/1.01 Galician 84.40/0.27 81.40/0.30 81.50/0.33 77.50/0.41 82.80/0.34 78.90/0.43 40.60/1.26 46.20/1.01 43.50/1.11 26.70/1.77 Georgian 90.20/0.31 90.20/0.31 91.50/0.33 91.10/0.22 92.10/0.21 91.40/0.30 35.40/1.72 30.50/2.09 43.90/1.53 28.10/2.33 German 67.50/0.79 67.50/0.79 66.00/0.95 73.50/0.57 71.60/0.71 67.60/0.80 12.10/3.72 14.10/3.58 11.90/3.64 Greek 14.90/2.65 15.10/2.34 16.40/3.21 52.90/1.16 59.30/1.03 34.40/2.27 4.20/4.51 5.90/4.82 4.50/5.21 4.90/5.45 Greenlandic 80.00/0.22 74.00/0.40 78.00/0.24 70.00/0.42 82.00/0.24 54.00/0.90 22.00/1.50 70.00/0.42 42.00/0.96 22.00/1.86 Haida 80.00/1.02 84.00/0.91 71.00/1.47 75.00/1.08 61.00/1.02 81.00/0.46 21.00/4.58 78.00/0.69 58.00/1.58 28.00/2.49 Hebrew 76.60/0.30 76.60/0.30 79.00/0.28 67.90/0.44 38.10/0.98 33.00/1.22 22.20/1.45 11.90/1.97 8.10/2.07 12.10/2.10 Hindi 87.40/0.30 87.40/0.30 86.40/0.47 94.80/0.08 87.20/0.18 74.40/0.53 71.80/1.02 47.30/1.68 52.20/1.54 0.80/3.28 Hungarian 58.60/0.83 58.60/0.83 51.10/1.35 65.30/0.77 44.40/1.42 56.80/0.82 1.50/4.83 5.90/4.34 7.80/3.95 2.40/4.92 Icelandic 51.70/0.94 51.70/0.94 49.90/1.06 44.20/1.15 58.80/0.83 44.90/1.04 10.80/2.79 11.60/3.04 16.30/2.64 9.70/2.99 Ingrian 78.00/0.36 88.00/0.30 84.00/0.30 46.00/0.86 80.00/0.30 44.00/1.28 68.00/0.54 46.00/0.88 36.00/1.00 24.00/1.40 Irish 45.40/1.75 45.40/1.75 43.30/1.92 47.90/1.68 44.10/1.57 19.50/3.79 3.00/5.48 1.70/6.88 3.70/6.84 4.20/7.16 Italian 77.50/0.44 78.30/0.43 79.10/0.45 87.30/0.23 72.50/0.81 50.60/1.56 10.90/3.02 13.20/3.08 1.50/5.54 11.40/3.74 Kabardian 99.00/0.01 99.00/0.01 100.00/0.00 42.00/1.31 83.00/0.17 97.00/0.03 46.00/0.74 81.00/0.27 79.00/0.29 75.00/0.41 Kannada 73.00/0.64 74.00/0.81 78.00/0.61 59.00/1.26 55.00/1.30 81.00/0.30 20.00/2.64 19.00/2.22 20.00/2.90 Karelian 96.00/0.12 96.00/0.12 96.00/0.10 42.00/1.00 98.00/0.02 62.00/0.90 60.00/1.24 58.00/0.80 50.00/0.82 26.00/1.42 Kashubian 78.00/0.30 78.00/0.30 88.00/0.20 68.00/0.58 60.00/0.64 76.00/0.38 6.00/2.12 74.00/0.38 56.00/0.70 32.00/1.36 Kazakh 24.00/1.42 24.00/1.42 4.00/5.68 48.00/1.14 86.00/0.14 44.00/1.08 16.00/1.74 0.00/4.34 44.00/0.66 38.00/0.88 Khakas 98.00/0.02 98.00/0.02 98.00/0.04 84.00/0.36 96.00/0.06 96.00/0.06 32.00/1.56 62.00/0.60 60.00/0.50 66.00/0.54 Khaling 62.20/0.97 53.80/1.33 56.20/1.28 15.30/2.59 18.00/2.01 5.30/2.56 27.20/2.09 5.50/3.65 6.80/3.42 5.40/3.86 Kurmanji 80.30/0.76 80.30/0.76 83.30/0.89 88.70/0.21 85.10/0.28 86.70/0.43 26.30/2.38 20.30/2.71 32.40/2.40 15.70/3.12 Ladin 92.00/0.14 88.00/0.25 90.00/0.17 79.00/0.33 86.00/0.35 77.00/0.37 62.00/0.57 67.00/0.53 63.00/0.63 24.00/1.77 Latin 22.60/2.05 22.60/2.05 22.60/2.32 27.90/1.63 37.90/1.16 11.20/2.31 2.60/3.91 2.90/4.00 5.20/3.40 3.20/4.18 Latvian 72.60/0.65 72.60/0.65 64.40/0.98 82.40/0.38 85.50/0.26 79.90/0.54 9.70/3.25 19.50/2.77 23.30/2.41 6.40/3.68 Lithuanian 26.70/1.77 26.50/1.64 23.60/2.13 39.50/1.17 52.00/0.70 25.20/1.59 3.60/3.36 18.20/2.29 12.20/2.56 3.40/3.54 Livonian 46.00/1.14 46.00/1.14 44.00/1.25 59.00/1.04 51.00/1.36 52.00/1.23 23.00/2.13 42.00/1.68 5.00/4.25 13.00/3.21 Lower-sorbian 69.30/0.54 69.30/0.54 65.70/0.64 57.90/0.88 68.90/0.60 68.60/0.52 15.50/2.06 27.30/1.61 29.70/1.38 20.30/1.85 Macedonian 71.50/0.47 71.50/0.47 71.50/0.50 73.80/0.37 82.60/0.32 79.50/0.32 14.20/2.38 18.00/2.38 20.40/2.43 21.20/2.37 Maltese 83.00/0.33 83.00/0.33 90.00/0.15 57.00/0.90 21.00/1.71 24.00/1.61 57.00/0.88 18.00/2.27 17.00/2.29 4.00/3.16 Mapudungun 98.00/0.02 98.00/0.02 98.00/0.02 82.00/0.20 98.00/0.02 90.00/0.24 94.00/0.12 52.00/0.86 54.00/0.78 32.00/1.38 Middle-french 85.40/0.32 85.40/0.32 90.20/0.22 93.10/0.14 90.20/0.18 86.00/0.32 54.70/1.09 61.30/0.97 19.90/2.51 31.40/1.74 Middle-high-german 66.00/0.72 66.00/0.72 96.00/0.04 54.00/0.66 52.00/0.88 84.00/0.26 78.00/0.34 40.00/1.00 48.00/0.84 30.00/1.40 Middle-low-german 86.00/0.30 86.00/0.30 98.00/0.04 38.00/1.44 30.00/1.38 76.00/0.64 58.00/1.10 60.00/1.06 36.00/1.92 Murrinhpatha 84.00/0.38 90.00/0.26 90.00/0.26 22.00/2.04 4.00/1.86 54.00/1.14 62.00/0.94 6.00/2.88 4.00/3.10 8.00/2.68 Navajo 30.30/2.13 30.30/2.13 33.60/2.26 29.60/2.10 30.40/2.49 4.30/5.01 9.10/3.40 3.00/4.36 3.10/4.41 0.80/5.44 Neapolitan 98.00/0.06 98.00/0.06 98.00/0.06 93.00/0.14 94.00/0.17 96.00/0.09 76.00/0.65 78.00/0.38 65.00/0.56 39.00/1.18 Norman 32.00/1.98 32.00/1.98 44.00/1.70 46.00/2.02 34.00/1.56 28.00/2.62 14.00/2.58 30.00/1.80 24.00/1.86 6.00/2.88 North-frisian 72.00/0.59 72.00/0.59 69.00/0.72 42.00/2.73 33.00/2.85 29.00/2.52 14.00/4.07 20.00/3.61 3.00/5.09 Northern-sami 44.60/1.22 44.60/1.22 47.50/1.42 33.00/2.02 34.90/1.38 23.10/1.72 8.30/3.63 13.10/2.61 8.20/2.70 2.60/3.85 Norwegian-bokmaal 77.00/0.38 77.00/0.38 77.30/0.43 81.40/0.29 80.50/0.31 77.30/0.34 26.70/2.13 28.50/2.33 28.30/2.36 29.40/1.84 Norwegian-nynorsk 56.90/0.70 56.90/0.70 58.60/0.70 57.20/0.71 60.50/0.65 54.40/0.77 15.20/2.40 18.00/2.54 19.80/2.64 20.60/2.08 Occitan 82.00/0.35 82.00/0.35 69.00/0.61 79.00/0.44 92.00/0.16 84.00/0.24 41.00/1.26 62.00/0.69 59.00/0.60 14.00/2.14 Old-armenian 42.10/1.32 42.20/1.23 43.30/1.27 41.00/1.41 67.30/0.71 55.00/0.93 13.00/2.34 14.00/2.31 24.40/1.95 17.20/2.40 Old-church-slavonic 87.00/0.22 87.00/0.22 86.00/0.22 63.00/0.85 77.00/0.51 83.00/0.22 39.00/0.94 54.00/0.67 49.00/0.67 47.00/0.82 Old-english 37.70/1.27 35.30/1.63 37.30/1.52 46.10/1.04 27.70/1.29 18.00/1.88 23.60/2.06 24.00/1.98 10.70/2.57 Old-french 60.50/0.83 58.60/0.77 62.60/0.78 56.00/0.98 63.00/0.74 55.80/0.89 20.20/2.02 27.40/1.79 27.10/1.80 18.40/2.17 Old-irish 6.00/3.26 6.00/3.26 6.00/3.46 16.00/3.80 8.00/2.94 8.00/2.82 6.00/3.26 8.00/3.22 6.00/3.50 4.00/3.96 Old-saxon 61.50/0.65 61.50/0.65 48.80/1.06 38.30/1.27 39.10/1.02 31.60/1.25 19.40/1.88 21.50/1.78 18.80/1.87 11.10/2.31 Pashto 65.00/0.67 65.00/0.67 70.00/0.59 46.00/1.51 68.00/0.71 57.00/0.77 43.00/1.02 18.00/1.82 7.00/3.29 18.00/2.46 Persian 72.60/0.73 72.60/0.73 76.30/0.52 80.40/0.35 66.50/1.01 49.90/1.80 48.80/1.63 37.50/2.83 29.40/3.38 1.30/5.04 Polish 49.70/1.21 49.70/1.21 49.40/1.20 62.60/0.90 73.50/0.64 60.80/0.99 9.50/3.61 7.50/3.92 9.60/3.73 Portuguese 89.30/0.17 88.60/0.19 80.80/0.31 89.10/0.18 92.40/0.13 84.30/0.24 26.30/1.78 35.60/1.50 11.90/2.76 22.90/1.98 Quechua 80.80/0.53 80.80/0.53 79.70/0.60 69.30/1.13 70.90/1.49 95.90/0.09 21.00/3.08 48.60/1.33 1.90/6.31 21.70/2.92 Romanian 61.40/1.03 61.40/1.03 59.70/1.04 56.60/1.14 69.30/0.75 47.20/1.72 12.70/3.34 13.70/3.68 15.90/3.51 5.60/4.20 Russian 51.70/1.16 51.70/1.16 51.80/1.47 66.70/0.72 76.40/0.52 56.70/1.09 8.50/4.04 1.80/5.44 9.60/4.14 Sanskrit 64.80/0.70 64.80/0.70 65.30/0.71 36.50/2.33 60.10/0.76 76.30/0.37 14.10/2.48 21.70/2.05 21.50/1.76 26.50/1.98 Scottish-gaelic 90.00/0.16 90.00/0.16 88.00/0.22 50.00/0.74 50.00/1.30 70.00/0.74 56.00/0.74 48.00/0.96 52.00/0.88 36.00/1.66 Serbo-croatian 38.60/1.26 38.60/1.26 30.30/1.87 58.10/1.01 68.20/0.77 47.20/1.58 7.70/4.18 10.90/4.20 10.80/4.16 5.00/4.55 Slovak 56.60/0.73 56.60/0.73 58.00/0.85 53.50/0.95 71.30/0.53 62.30/0.64 10.50/2.12 21.80/1.72 26.90/1.46 17.60/1.79 Slovene 68.70/0.56 68.70/0.56 55.00/1.01 43.60/0.76 72.30/0.46 73.30/0.47 0.20/8.25 29.90/1.86 34.40/1.60 8.30/3.26 Sorani 40.30/1.35 40.10/1.44 46.10/1.19 28.40/2.35 52.10/1.05 15.50/2.66 30.70/1.80 6.40/3.58 6.60/3.70 1.60/4.42 Spanish 82.30/0.38 82.30/0.38 79.40/0.51 85.70/0.26 86.50/0.35 73.10/0.85 28.10/2.04 25.20/2.52 21.90/2.95 20.10/2.63 Swahili 84.00/0.34 86.00/0.27 85.00/0.35 77.00/0.42 73.00/0.37 1.00/3.21 72.00/0.59 1.00/3.35 3.00/3.30 1.00/4.16 Swedish 68.70/0.53 69.10/0.53 70.00/0.54 76.20/0.36 76.40/0.41 69.50/0.50 11.10/3.05 13.90/3.48 2.70/4.93 19.00/2.99 Tatar 92.00/0.10 92.00/0.10 90.00/0.17 95.00/0.05 90.00/0.14 92.00/0.08 53.00/1.30 62.00/1.12 57.00/1.11 52.00/1.34 Tibetan 26.00/1.30 26.00/1.30 30.00/1.40 36.00/1.00 42.00/1.00 44.00/0.88 24.00/1.32 26.00/1.24 38.00/1.04 42.00/0.94 Turkish 38.30/2.06 38.30/1.78 36.50/2.31 80.50/0.55 32.20/2.95 73.70/0.51 9.30/4.03 20.70/2.92 18.90/3.18 12.60/4.37 Turkmen 96.00/0.04 94.00/0.06 96.00/0.06 68.00/0.64 94.00/0.06 96.00/0.04 48.00/0.68 70.00/0.36 82.00/0.22 66.00/0.62 Ukrainian 66.80/0.62 66.80/0.62 61.10/0.72 69.20/0.61 74.00/0.49 62.70/0.66 14.50/2.20 19.10/2.10 25.20/1.94 20.40/2.23 Urdu 85.90/0.42 82.90/0.53 83.90/0.54 87.20/0.46 87.00/0.23 78.20/0.46 80.70/0.80 53.60/1.55 45.50/1.58 2.70/3.38 Uzbek 100.00/0.00 100.00/0.00 100.00/0.00 89.00/0.18 96.00/0.04 100.00/0.00 14.00/3.03 63.00/0.62 59.00/0.73 66.00/0.93 Venetian 86.40/0.18 81.10/0.31 82.20/0.28 70.60/0.40 88.90/0.16 89.80/0.17 45.40/1.27 56.00/0.74 58.40/0.69 20.10/1.81 Votic 78.00/0.37 79.00/0.38 79.00/0.37 19.00/2.05 37.00/1.39 64.00/0.48 59.00/0.78 45.00/0.94 39.00/1.03 21.00/1.62 Welsh 81.00/0.38 79.00/0.36 85.00/0.29 56.00/0.98 58.00/1.01 62.00/0.68 23.00/1.80 41.00/1.06 46.00/1.03 9.00/2.20 West-frisian 91.00/0.27 91.00/0.25 92.00/0.26 60.00/0.79 65.00/0.81 67.00/0.53 76.00/0.44 48.00/1.13 53.00/0.96 28.00/1.56 Yiddish 79.00/0.37 80.00/0.30 84.00/0.28 80.00/0.31 87.00/0.22 78.00/0.39 62.00/0.87 56.00/1.01 63.00/0.96 39.00/1.51 Zulu 60.00/0.93 60.00/0.93 54.00/1.05 67.40/0.78 53.60/0.98 1.70/3.34 15.80/2.42 1.20/3.73 1.50/4.41 0.80/5.04

Table 16: Task 1 Medium Condition Part 2.
{adjustbox}

width=0.94 oracle-fc oracle-e axsemantics-01 uzh-02 uzh-01 ua-08 iitbhu-iiith-02 ua-05 iitbhu-iiith-01 ua-06 ua-03 waseda-01 Adyghe 98.30/* 98.40/* 90.30/0.10 90.00/0.10 90.60/0.14 89.10/0.17 82.10/0.22 85.50/0.21 88.90/0.17 81.00/0.23 73.10/0.51 Albanian 54.80/* 54.90/* 35.70/1.89 36.40/1.87 20.50/4.00 29.70/2.00 21.80/4.35 26.40/2.20 20.30/4.02 18.90/3.03 24.40/4.82 Arabic 54.20/* 57.40/* 44.30/1.80 45.20/1.77 35.80/2.24 35.20/2.04 19.80/5.28 31.90/2.19 35.50/2.23 0.10/6.27 27.60/3.33 Armenian 55.30/* 76.20/* 64.60/0.82 64.90/0.77 49.10/1.73 54.40/0.92 50.50/1.51 51.60/1.02 43.30/1.91 49.90/1.53 37.00/2.18 Asturian 65.20/* 82.80/* 74.60/0.47 72.50/0.52 58.00/0.95 70.30/0.55 67.90/0.83 66.30/0.62 56.80/1.00 57.00/1.05 58.60/1.06 Azeri 71.00/* 79.00/* 62.00/0.82 62.00/0.86 46.00/1.31 65.00/0.76 48.00/1.17 63.00/0.81 24.00/2.62 36.00/1.63 39.00/1.35 Bashkir 98.00/* 94.90/* 70.10/0.51 67.20/0.53 77.50/0.48 77.20/0.44 42.90/1.10 77.80/0.44 61.90/1.42 42.10/1.11 39.40/1.84 Basque 5.60/* 31.20/* 12.70/3.05 13.30/2.98 11.40/3.29 11.80/3.02 6.70/3.88 9.60/3.13 10.50/3.34 3.30/3.96 6.50/3.70 Belarusian 86.30/* 54.80/* 30.20/2.18 30.00/2.16 33.40/1.97 22.90/2.35 16.20/2.52 21.20/2.45 31.30/2.04 16.20/2.52 10.30/2.72 Bengali 83.00/* 91.00/* 72.00/0.49 71.00/0.53 68.00/1.00 58.00/0.83 62.00/0.63 55.00/0.88 64.00/1.16 59.00/0.90 52.00/0.76 Breton 74.00/* 86.00/* 71.00/0.76 72.00/0.78 69.00/0.92 61.00/0.95 69.00/0.85 61.00/0.93 67.00/0.99 66.00/0.90 61.00/0.99 Bulgarian 66.10/* 80.90/* 58.50/0.70 58.20/0.70 62.90/1.02 54.00/0.81 50.40/0.97 51.50/0.89 62.40/1.02 49.30/0.98 50.00/0.90 Catalan 86.90/* 87.40/* 69.90/0.57 68.50/0.60 72.50/0.65 64.40/0.65 62.80/0.79 61.80/0.70 71.60/0.68 60.50/0.82 60.80/0.86 Classical-syriac 95.00/* 98.00/* 96.00/0.04 96.00/0.04 94.00/0.06 92.00/0.09 95.00/0.06 91.00/0.10 93.00/0.07 94.00/0.06 94.00/0.06 Cornish 68.00/* 58.00/* 24.00/1.82 28.00/1.56 38.00/1.40 30.00/1.68 40.00/1.56 22.00/1.82 6.00/2.80 14.00/2.90 12.00/3.62 Crimean-tatar 98.00/* 98.00/* 89.00/0.15 89.00/0.15 89.00/0.17 91.00/0.14 81.00/0.25 89.00/0.17 85.00/0.21 71.00/0.39 67.00/0.43 Czech 56.70/* 58.30/* 46.50/1.97 46.50/1.80 40.00/2.09 37.00/1.51 42.30/1.80 33.80/1.68 40.00/2.09 39.80/1.83 39.30/2.14 Danish 96.20/* 91.10/* 69.90/0.51 70.00/0.50 87.20/0.24 65.20/0.58 68.20/0.49 65.80/0.58 87.70/0.24 68.10/0.48 64.70/0.67 Dutch 95.20/* 80.80/* 55.20/0.71 56.00/0.70 68.90/0.60 58.60/0.62 56.00/0.70 57.60/0.65 69.30/0.62 55.30/0.74 53.70/0.86 English 100.00/* 92.90/* 90.30/0.14 90.30/0.14 91.80/0.16 86.50/0.20 89.80/0.15 81.30/0.28 91.80/0.16 89.90/0.15 80.80/0.26 Estonian 70.30/* 62.20/* 33.60/1.49 35.20/1.36 33.30/2.17 33.00/1.59 28.70/1.73 29.00/1.72 31.80/2.22 29.00/1.74 30.80/1.79 Faroese 85.70/* 68.90/* 43.40/1.18 45.90/1.12 49.80/1.20 33.00/1.37 43.10/1.14 27.60/1.51 49.30/1.22 39.20/1.25 39.70/1.50 Finnish 58.10/* 39.80/* 24.90/2.05 25.70/2.01 20.00/3.29 19.90/2.31 21.30/2.90 19.00/2.37 19.30/3.31 20.80/2.91 19.50/2.59 French 85.50/* 78.50/* 66.60/0.61 65.20/0.61 60.80/1.03 55.00/0.83 59.00/0.96 53.10/0.90 58.90/1.08 55.60/0.76 58.90/0.97 Friulian 89.00/* 85.00/* 79.00/0.39 79.00/0.36 70.00/0.41 72.00/0.64 75.00/0.59 70.00/0.63 69.00/0.52 68.00/0.64 70.00/0.52 Galician 73.00/* 72.40/* 61.10/0.72 60.80/0.70 47.60/1.17 48.40/0.94 53.70/1.21 43.50/1.09 41.80/1.39 50.30/1.24 53.00/1.22 Georgian 93.80/* 92.40/* 84.50/0.35 84.00/0.34 88.20/0.30 80.40/0.41 83.00/0.38 77.70/0.46 87.50/0.33 83.40/0.38 70.60/0.58 German 79.60/* 85.60/* 62.00/0.78 62.40/0.76 55.00/1.14 57.00/0.93 59.30/0.83 53.50/1.00 55.00/1.14 67.10/0.69 52.00/1.04 Greek 57.70/* 49.00/* 32.20/1.93 32.30/1.83 23.90/2.89 27.10/1.96 28.90/2.32 26.10/2.01 23.40/2.90 26.90/2.36 25.10/2.71 Greenlandic 100.00/* 96.00/* 74.00/0.48 74.00/0.42 68.00/0.48 72.00/0.36 68.00/0.42 80.00/0.30 66.00/0.42 64.00/0.50 52.00/2.74 Haida 45.00/* 77.00/* 63.00/1.42 60.00/1.77 20.00/4.47 57.00/1.45 49.00/2.07 48.00/1.86 20.00/4.47 10.00/4.94 29.00/5.64 Hebrew 82.40/* 71.80/* 39.90/1.02 39.10/1.02 56.70/0.91 34.30/1.12 35.40/1.04 32.40/1.19 56.40/0.92 37.20/1.03 26.00/1.46 Hindi 38.80/* 85.80/* 78.00/0.79 78.00/0.68 45.80/1.57 75.90/0.89 70.80/1.17 75.10/0.90 39.60/1.66 49.40/1.22 59.60/1.19 Hungarian 78.90/* 65.40/* 41.30/1.19 40.30/1.18 48.20/1.41 39.20/1.35 33.10/1.37 33.50/1.54 47.90/1.42 28.00/1.58 25.90/1.83 Icelandic 92.20/* 71.80/* 48.90/1.04 48.20/1.01 56.20/0.94 31.70/1.39 42.60/1.16 29.30/1.49 55.50/0.95 40.60/1.19 38.90/1.49 Ingrian 100.00/* 66.00/* 36.00/1.58 36.00/1.34 20.00/1.78 46.00/1.10 32.00/1.14 34.00/1.32 38.00/0.92 26.00/1.36 4.00/2.88 Irish 82.70/* 50.40/* 37.60/2.18 37.70/2.09 35.30/2.40 26.00/2.58 34.80/2.11 22.30/2.74 35.50/2.38 35.10/2.13 34.10/3.03 Italian 82.80/* 73.80/* 57.40/1.02 57.00/1.17 47.80/1.64 47.30/1.07 42.20/2.06 45.40/1.09 46.50/1.71 38.40/1.97 42.20/2.05 Kabardian 99.00/* 100.00/* 90.00/0.10 92.00/0.08 88.00/0.21 87.00/0.13 86.00/0.18 84.00/0.16 86.00/0.25 84.00/0.20 75.00/0.32 Kannada 74.00/* 78.00/* 60.00/0.80 61.00/0.79 38.00/2.14 57.00/0.89 57.00/1.10 55.00/0.92 37.00/2.14 31.00/2.40 36.00/2.10 Karelian 88.00/* 98.00/* 88.00/0.20 88.00/0.16 70.00/0.60 92.00/0.14 94.00/0.10 90.00/0.18 56.00/0.78 64.00/0.70 46.00/2.54 Kashubian 100.00/* 82.00/* 64.00/0.72 66.00/0.72 44.00/1.16 58.00/0.88 68.00/0.70 52.00/1.14 58.00/0.76 54.00/1.08 56.00/0.90 Kazakh 100.00/* 92.00/* 86.00/0.14 80.00/0.20 62.00/0.50 78.00/0.22 76.00/0.30 84.00/0.16 34.00/1.48 42.00/0.94 38.00/2.18 Khakas 100.00/* 92.00/* 78.00/0.26 78.00/0.28 70.00/0.64 86.00/0.16 56.00/0.54 80.00/0.24 42.00/0.92 54.00/0.56 36.00/1.96 Khaling 22.00/* 53.20/* 22.30/2.13 23.00/2.04 33.80/2.32 24.50/1.83 15.50/2.14 19.20/2.01 30.50/2.44 15.50/2.14 12.10/2.89 Kurmanji 90.20/* 89.20/* 87.40/0.64 87.10/0.34 84.00/0.44 80.80/0.55 87.20/0.35 75.20/0.64 84.00/0.44 86.70/0.35 86.20/0.43 Ladin 77.00/* 83.00/* 71.00/0.55 72.00/0.52 59.00/0.80 68.00/0.62 67.00/1.03 64.00/0.68 58.00/0.80 55.00/0.89 58.00/0.96 Latin 52.30/* 42.00/* 17.10/2.42 17.70/2.32 33.00/1.84 15.30/2.48 16.90/2.46 14.00/2.52 33.10/1.84 17.40/2.44 16.10/2.84 Latvian 80.10/* 70.90/* 54.50/1.00 54.50/1.16 55.90/0.97 35.90/1.43 56.00/1.41 36.40/1.43 57.30/0.96 52.00/1.04 52.80/0.87 Lithuanian 65.40/* 44.30/* 20.70/1.84 20.40/1.78 32.40/1.97 16.90/1.91 23.20/1.87 14.50/2.03 32.60/1.97 19.20/2.15 23.00/1.89 Livonian 73.00/* 53.00/* 33.00/1.69 32.00/1.69 35.00/1.60 33.00/1.86 30.00/1.67 32.00/1.88 34.00/1.73 29.00/1.68 29.00/2.21 Lower-sorbian 75.90/* 77.00/* 45.60/1.04 46.20/1.02 54.20/1.08 40.90/1.15 48.70/0.98 37.50/1.23 54.30/1.10 50.50/0.97 36.30/1.38 Macedonian 79.20/* 85.90/* 59.80/0.61 60.20/0.60 67.60/0.63 57.40/0.60 56.50/0.68 53.70/0.65 68.80/0.62 54.20/0.76 49.80/0.89 Maltese 99.00/* 69.00/* 40.00/1.16 39.00/1.12 48.00/1.02 32.00/1.31 18.00/1.64 24.00/1.43 49.00/0.96 18.00/1.64 20.00/1.96 Mapudungun 88.00/* 100.00/* 78.00/0.34 80.00/0.36 84.00/0.38 84.00/0.28 86.00/0.24 84.00/0.28 70.00/0.60 76.00/0.46 68.00/1.64 Middle-french 86.70/* 93.40/* 84.50/0.32 83.10/0.34 74.50/0.66 83.10/0.34 81.40/0.36 82.60/0.39 68.80/0.67 74.20/0.55 76.90/0.49 Middle-high-german 94.00/* 90.00/* 84.00/0.30 82.00/0.34 58.00/0.80 70.00/0.62 36.00/2.46 70.00/0.62 32.00/0.98 38.00/0.94 52.00/1.76 Middle-low-german 92.00/* 64.00/* 50.00/1.34 54.00/1.28 24.00/2.26 42.00/1.44 34.00/1.32 36.00/1.50 30.00/1.54 34.00/1.32 18.00/2.84 Murrinhpatha 98.00/* 64.00/* 36.00/1.48 36.00/1.36 38.00/1.68 34.00/1.66 38.00/1.58 36.00/1.66 26.00/1.72 38.00/1.58 34.00/2.24 Navajo 88.90/* 30.00/* 19.80/3.42 20.80/2.96 12.10/3.75 10.30/3.46 8.00/6.97 8.80/3.58 12.00/3.75 13.70/3.99 16.10/3.75 Neapolitan 90.00/* 90.00/* 89.00/0.28 86.00/0.31 80.00/0.59 80.00/0.41 83.00/0.32 76.00/0.50 80.00/0.66 81.00/0.59 79.00/0.59 Norman 88.00/* 84.00/* 52.00/0.92 54.00/0.88 58.00/1.16 64.00/0.70 66.00/0.66 64.00/1.00 32.00/1.62 40.00/1.72 44.00/2.34 North-frisian 85.00/* 61.00/* 42.00/2.23 40.00/2.29 31.00/2.39 45.00/2.11 31.00/2.69 38.00/2.12 20.00/4.53 27.00/2.80 36.00/2.25 Northern-sami 69.10/* 50.70/* 20.70/2.02 21.10/1.90 35.80/2.29 15.40/2.33 16.80/2.35 12.30/2.48 34.00/2.34 14.80/2.54 16.40/2.35 Norwegian-bokmaal 99.30/* 94.10/* 79.10/0.32 78.70/0.33 89.30/0.21 69.60/0.47 78.30/0.35 68.40/0.51 90.10/0.20 76.80/0.37 73.30/0.40 Norwegian-nynorsk 98.30/* 89.30/* 56.70/0.73 56.80/0.74 83.60/0.39 54.80/0.79 57.30/0.70 52.30/0.85 82.90/0.42 56.50/0.72 54.60/0.81 Occitan 91.00/* 82.00/* 77.00/0.49 76.00/0.52 68.00/0.72 72.00/0.55 71.00/1.24 69.00/0.61 66.00/0.84 68.00/0.76 72.00/0.82 Old-armenian 47.40/* 59.70/* 42.00/1.32 41.80/1.30 32.90/1.50 29.70/1.58 36.10/1.70 24.30/1.77 31.10/1.53 31.20/1.46 30.50/1.79 Old-church-slavonic 97.00/* 76.00/* 48.00/1.06 46.00/1.08 50.00/0.88 53.00/0.78 42.00/1.11 50.00/0.87 50.00/0.92 34.00/1.20 40.00/1.14 Old-english 81.00/* 64.80/* 24.10/1.58 24.70/1.54 46.50/1.11 23.10/1.62 29.00/1.34 20.70/1.75 46.40/1.11 29.00/1.34 17.60/1.72 Old-french 65.80/* 68.70/* 46.20/1.10 46.10/1.05 36.30/1.48 39.50/1.23 34.90/1.67 36.70/1.32 34.80/1.53 32.90/1.78 36.00/1.31 Old-irish 46.00/* 16.00/* 8.00/3.62 8.00/3.84 0.00/6.94 4.00/3.80 6.00/4.28 4.00/3.84 4.00/3.90 4.00/4.54 8.00/4.34 Old-saxon 68.30/* 64.50/* 30.00/1.32 31.40/1.29 46.50/1.06 25.10/1.52 29.30/1.38 22.20/1.61 46.60/1.06 28.70/1.41 22.80/1.84 Pashto 59.00/* 65.00/* 48.00/1.25 48.00/1.29 36.00/1.59 35.00/1.30 37.00/1.70 33.00/1.34 36.00/1.57 33.00/1.75 31.00/2.26 Persian 54.70/* 81.10/* 67.60/0.59 67.50/0.55 35.30/2.25 61.60/0.81 34.80/2.98 56.20/0.94 25.50/2.87 28.10/2.77 36.60/1.78 Polish 75.90/* 66.20/* 45.30/1.43 44.60/1.47 49.10/1.87 32.50/1.72 44.60/1.49 29.00/1.83 49.40/1.86 43.60/1.50 42.00/1.72 Portuguese 73.70/* 87.00/* 75.80/0.40 73.80/0.43 62.50/0.72 59.20/0.69 63.90/0.80 57.80/0.69 61.70/0.74 56.50/1.01 62.60/0.83 Quechua 21.40/* 88.70/* 70.20/0.90 69.00/0.93 62.00/1.21 61.60/1.02 50.80/1.42 56.50/1.10 36.20/2.98 26.40/2.79 33.50/2.35 Romanian 79.40/* 65.10/* 46.00/1.34 46.20/1.34 45.00/1.94 35.50/1.83 45.80/1.48 32.80/1.94 45.40/1.94 40.40/1.58 44.70/1.52 Russian 80.20/* 71.50/* 53.20/1.09 53.50/1.07 49.90/1.33 42.10/1.36 47.40/1.14 39.20/1.42 50.20/1.34 44.60/1.18 47.00/1.27 Sanskrit 68.90/* 74.20/* 56.80/0.90 58.00/0.93 43.50/1.53 52.20/1.07 46.20/1.23 51.40/1.08 39.20/1.73 46.00/1.33 42.40/1.24 Scottish-gaelic 100.00/* 84.00/* 68.00/0.62 70.00/0.58 58.00/0.98 74.00/0.50 50.00/0.82 60.00/0.78 62.00/0.94 46.00/0.82 38.00/2.50 Serbo-croatian 34.50/* 62.80/* 43.00/1.50 43.50/1.53 29.50/2.38 35.10/1.56 43.80/1.69 34.20/1.68 28.70/2.40 44.80/1.70 29.30/1.88 Slovak 90.00/* 71.00/* 51.80/0.96 51.10/0.97 48.00/1.06 42.30/1.11 51.30/1.00 39.30/1.18 48.00/1.06 49.00/1.02 48.40/1.09 Slovene 70.80/* 79.70/* 58.00/0.73 57.50/0.74 54.10/0.82 48.60/0.88 46.00/0.96 42.90/0.96 53.10/0.84 30.80/1.19 34.10/1.27 Sorani 38.20/* 55.00/* 38.90/1.55 40.10/1.32 28.60/2.16 26.40/1.70 24.80/2.38 22.90/1.83 27.50/2.20 24.20/2.39 19.90/3.35 Spanish 82.70/* 79.10/* 68.90/0.80 67.80/0.66 73.20/0.80 57.20/0.79 68.60/0.76 53.50/0.87 72.50/0.81 64.90/0.81 61.80/1.08 Swahili 39.00/* 80.00/* 58.00/0.70 58.00/0.73 36.00/1.49 72.00/0.51 33.00/1.57 69.00/0.56 33.00/1.59 33.00/1.57 70.00/0.51 Swedish 95.00/* 86.90/* 68.40/0.51 67.90/0.52 79.00/0.40 61.40/0.70 62.00/0.67 58.50/0.75 77.70/0.42 60.30/0.72 61.70/0.62 Tatar 98.00/* 96.00/* 86.00/0.16 88.00/0.14 90.00/0.14 85.00/0.26 72.00/0.35 79.00/0.32 89.00/0.15 66.00/0.44 67.00/0.44 Telugu 86.00/* 98.00/* 72.00/0.96 96.00/0.12 92.00/0.24 96.00/0.12 96.00/0.16 96.00/0.06 94.00/0.22 82.00/0.74 82.00/0.42 54.00/2.90 Tibetan 100.00/* 82.00/* 52.00/0.78 52.00/0.76 36.00/1.10 48.00/0.80 34.00/0.98 58.00/0.62 54.00/0.66 38.00/0.84 34.00/1.02 Turkish 39.60/* 55.10/* 38.10/2.04 39.00/1.89 22.60/2.99 39.50/2.09 27.50/2.53 37.50/2.13 11.80/5.16 12.20/3.67 26.70/2.79 Turkmen 100.00/* 96.00/* 90.00/0.14 86.00/0.20 60.00/1.00 80.00/0.28 86.00/0.20 78.00/0.34 68.00/0.46 72.00/0.44 50.00/2.06 Ukrainian 85.40/* 71.40/* 48.90/0.92 49.40/0.92 56.50/0.92 38.10/1.05 47.70/0.98 30.70/1.22 57.10/0.92 47.50/1.01 46.60/0.99 Urdu 41.30/* 82.80/* 72.50/0.48 71.90/0.48 50.70/1.56 69.60/0.60 66.50/0.74 65.90/0.66 33.90/1.67 44.90/1.07 58.90/0.84 Uzbek 75.00/* 98.00/* 90.00/0.12 92.00/0.10 85.00/0.29 91.00/0.14 77.00/0.31 90.00/0.14 43.00/2.01 46.00/1.13 63.00/0.78 Venetian 88.50/* 88.80/* 78.80/0.35 78.40/0.37 76.80/0.42 75.90/0.38 73.80/0.35 74.10/0.40 76.40/0.43 71.50/0.56 71.80/0.55 Votic 94.00/* 55.00/* 26.00/1.56 26.00/1.56 32.00/1.47 26.00/1.60 21.00/1.55 23.00/1.69 29.00/1.54 19.00/1.57 17.00/1.88 Welsh 88.00/* 75.00/* 55.00/1.08 55.00/1.01 50.00/1.40 50.00/1.12 48.00/1.11 45.00/1.27 51.00/1.35 43.00/1.17 42.00/1.25 West-frisian 100.00/* 72.00/* 53.00/1.04 56.00/1.01 46.00/1.13 51.00/1.05 51.00/1.17 48.00/1.08 47.00/1.12 40.00/1.26 50.00/1.23 Yiddish 100.00/* 96.00/* 82.00/0.31 83.00/0.30 87.00/0.23 71.00/0.39 82.00/0.30 68.00/0.44 82.00/0.29 78.00/0.35 78.00/0.38 Zulu 43.50/* 54.10/* 32.10/1.52 33.00/1.49 29.30/1.62 29.80/1.60 20.10/1.91 27.70/1.73 29.30/1.62 20.10/1.91 31.00/1.53

Table 17: Task 1 Low Condition Part 1.
{adjustbox}

width= msu-02 ua-02 hamburg-01 ua-07 baseline ua-01 msu-04 msu-03 iit-varanasi-01 ua-04 axsemantics-02 tuebingen-oslo-02 Adyghe 82.50/0.23 82.00/0.22 83.20/0.26 59.00/0.46 59.00/0.46 52.00/0.94 3.70/3.21 57.80/0.66 40.00/1.61 13.40/1.47 0.90/5.25 Albanian 25.40/4.57 12.60/7.09 16.10/2.69 13.50/7.10 22.00/4.37 22.00/4.37 5.50/4.88 4.40/5.10 0.80/6.08 2.90/6.88 0.00/12.09 0.00/8.87 Arabic 1.20/4.39 25.50/4.04 23.50/2.52 20.40/2.99 25.60/2.98 25.60/2.98 0.10/25.85 0.00/22.93 0.80/5.30 0.40/6.74 0.00/10.78 0.30/6.16 Armenian 37.70/2.27 47.70/2.28 39.80/1.44 37.80/2.09 37.00/2.18 37.00/2.18 16.60/2.33 18.20/2.48 5.80/3.80 4.10/4.28 10.70/2.87 0.10/6.56 Asturian 58.00/1.08 58.90/0.96 64.50/0.64 45.80/1.03 58.60/1.06 58.60/1.06 51.10/1.02 43.00/1.18 23.90/1.62 22.30/2.39 9.80/1.96 1.80/4.14 Azeri 43.00/1.96 48.00/1.19 37.00/1.40 47.00/1.27 24.00/2.84 24.00/2.84 26.00/2.11 34.00/1.55 28.00/2.10 12.00/2.98 24.00/1.74 1.00/4.77 Bashkir 45.50/1.07 58.20/0.76 75.40/0.50 39.40/1.84 39.40/1.84 27.20/2.08 14.50/3.16 35.80/1.35 24.50/2.08 22.30/1.26 0.20/4.96 Basque 0.10/6.56 0.50/5.48 1.70/4.06 6.20/3.53 0.10/6.52 0.10/6.52 0.90/4.72 0.10/5.43 0.50/4.71 6.70/3.88 0.00/10.18 0.20/5.45 Belarusian 20.60/2.38 6.50/3.47 17.10/2.44 17.80/3.07 6.80/2.44 6.80/2.44 17.20/2.50 12.30/2.61 2.90/3.68 3.40/3.90 0.00/8.99 0.30/5.72 Bengali 51.00/1.22 59.00/0.71 48.00/1.13 45.00/1.27 50.00/1.24 50.00/1.24 26.00/1.76 23.00/1.85 26.00/2.25 27.00/2.47 4.00/2.44 1.00/3.40 Breton 21.00/2.73 62.00/1.02 34.00/1.31 52.00/0.96 20.00/2.58 20.00/2.58 29.00/1.60 27.00/2.04 30.00/1.52 50.00/1.13 1.00/3.27 10.00/2.82 Bulgarian 31.00/1.92 27.50/2.59 38.50/1.24 33.80/1.98 30.90/1.69 30.90/1.69 21.90/1.96 13.30/2.08 9.70/2.85 13.80/2.96 4.70/2.98 0.00/6.02 Catalan 59.20/1.01 55.20/0.83 54.60/0.84 58.10/0.83 60.80/0.86 60.80/0.86 51.00/1.15 29.60/1.33 25.70/1.62 17.10/2.28 0.00/9.15 0.80/5.16 Classical-syriac 93.00/0.08 93.00/0.08 92.00/0.09 55.00/0.70 94.00/0.06 94.00/0.06 86.00/0.17 73.00/0.34 62.00/0.47 55.00/0.75 71.00/0.39 19.00/1.82 Cornish 12.00/4.02 8.00/3.94