{ "paper_id": "2022", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T13:12:05.602960Z" }, "title": "Developing a Part-Of-Speech tagger for te reo M\u0101ori", "authors": [ { "first": "Aoife", "middle": [], "last": "Finn", "suffix": "", "affiliation": { "laboratory": "", "institution": "Te Hiku Media", "location": {} }, "email": "" }, { "first": "Peter-Lucas", "middle": [], "last": "Jones", "suffix": "", "affiliation": {}, "email": "peterlucas@tehiku.co.nz" }, { "first": "Keoni", "middle": [], "last": "Mahelona", "suffix": "", "affiliation": {}, "email": "" }, { "first": "Suzanne", "middle": [], "last": "Duncan", "suffix": "", "affiliation": {}, "email": "suzanne@tehiku.co.nz" }, { "first": "Gianna", "middle": [], "last": "Leoni", "suffix": "", "affiliation": {}, "email": "keoni@tehiku.co.nz" } ], "year": "", "venue": null, "identifiers": {}, "abstract": "This abstract discusses the development of a Part-of-Speech tagger for te reo M\u0101ori, which is the Indigenous language of Aotearoa-New Zealand. It mostly focuses on the creation of a tagset that is appropriate for M\u0101ori. This is in consideration of the fact that some tagsets have existing tags that are not suitable for some M\u0101ori word classes. Alternatively, the existing tagsets might lack entirely a suitable tag for some M\u0101ori word classes. And finally, some existing tagsets do not adequately reflect a M\u0101ori worldview. Emphasis is put on the importance of capturing the language according to the conceptualization of its speakers, and not imposing \"traditional\" grammatical categories where it is not appropriate. The solution involved changing how some existing tags are used and in some cases creating entirely new tags that are appropriate for Te reo M\u0101ori. The Part-of-Speech tagger was successfully built by a M\u0101ori Indigenous organisation and is being used as the foundation for other applications.", "pdf_parse": { "paper_id": "2022", "_pdf_hash": "", "abstract": [ { "text": "This abstract discusses the development of a Part-of-Speech tagger for te reo M\u0101ori, which is the Indigenous language of Aotearoa-New Zealand. It mostly focuses on the creation of a tagset that is appropriate for M\u0101ori. This is in consideration of the fact that some tagsets have existing tags that are not suitable for some M\u0101ori word classes. Alternatively, the existing tagsets might lack entirely a suitable tag for some M\u0101ori word classes. And finally, some existing tagsets do not adequately reflect a M\u0101ori worldview. Emphasis is put on the importance of capturing the language according to the conceptualization of its speakers, and not imposing \"traditional\" grammatical categories where it is not appropriate. The solution involved changing how some existing tags are used and in some cases creating entirely new tags that are appropriate for Te reo M\u0101ori. The Part-of-Speech tagger was successfully built by a M\u0101ori Indigenous organisation and is being used as the foundation for other applications.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Abstract", "sec_num": null } ], "body_text": [ { "text": "This paper discusses the development of a Part-of-Speech tagger for M\u0101ori called Whakairo Kupu, meaning to carve or sculpt words. It specifically focuses on the creation of a tagset that was appropriate for M\u0101ori. Our current precision and recall scores are approximately 93%. Hereinafter, M\u0101ori will be referred to as te reo M\u0101ori or alternatively just M\u0101ori, and Universal Dependencies will be abbreviated to UD. Part-of-speech will be abbreviated as POS.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1." }, { "text": "Furthermore in this paper, linguistic examples will consist of four to five lines. The first line will include a morpheme by morpheme te reo M\u0101ori phrase or sentence. The second line will see each morpheme with a linguistic gloss that gives information about the syntactic properties or meaning of the morpheme. A third line will show the POS tags that our tagger would assign to the morpheme. The fourth and typically final line will show the English translation. However, in the uncommon instance that a literal translation is needed, it will be shown on the fifth line. For an example of this, please see (1).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1." }, { "text": "Glosses Haere mai te reo go DIR linguistic VERB MOD POS \"Welcome!\" translation Lit: \"Go hither\" literal translation", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "1) Example", "sec_num": null }, { "text": "Moving on, te reo M\u0101ori is the Indigenous language of Aotearoa, also known as New Zealand, (Morrison, 2011) . It is a member of the Eastern Polynesian branch of the Austronesian language family which itself has approximately 1200 members, (Harlow, 2007) . M\u0101ori is related to other members of the Polynesian branch of Austronesian such as Rapanui, Rarotongan, Tahitian, Tuamotuan, Marquesan, Hawai'ian and Mangarevan, (Du Feu, 1996) . Te reo M\u0101ori is a head-first and dependent-marking language, it is analytical with a high degree of polysemy.", "cite_spans": [ { "start": 91, "end": 107, "text": "(Morrison, 2011)", "ref_id": "BIBREF5" }, { "start": 239, "end": 253, "text": "(Harlow, 2007)", "ref_id": null }, { "start": 348, "end": 432, "text": "Rarotongan, Tahitian, Tuamotuan, Marquesan, Hawai'ian and Mangarevan, (Du Feu, 1996)", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "1) Example", "sec_num": null }, { "text": "Prior to the development of this tagger, there was no POS tagger for M\u0101ori from Aotearoa. POS taggers tag words according to their syntactic or grammatical category. However, many traditional syntactic categories, and by consequence POS labels, do not \"work for\" M\u0101ori, see (2). By this we mean for some of the traditional categories: We wanted a tagset that is usable with industry-wide tools, but we also needed a tagset that would meet the needs of te reo M\u0101ori. After researching various tagsets, we decided to base our tagset and guidelines on the UD tagset and tagging conventions. However, the categorization of words has been significantly altered to be appropriate for M\u0101ori. This is because at the time of development of our POS tagger, the UD conventions had still not been used to tag a Polynesian language such as te reo M\u0101ori, nor did it provide any guidelines about how to tag them.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "1) Example", "sec_num": null }, { "text": "Therefore the question arose as to how do we tag these words. Of course, we looked at how languages, other than the \"big languages\" such as English, were tagged. Yet, what works for other languages does not necessarily work for M\u0101ori. Furthermore, it would be a mistake to presume that the tagging solution for one Indigenous language should be applied to all Indigenous languages. As part of the re-Indigenization and decolonization, we do not homogenise Indigenous languages.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "1) Example", "sec_num": null }, { "text": "At this point, it is fitting to take a moment to digress and remind ourselves that at Te Hiku media our vision statement is He reo tuku iho, he reo ora which means A living language transmitted intergenerationally. This vision statement informs every decision that we make at every level. That means that it is of the utmost importance that we faithfully and accurately capture te reo M\u0101ori, as the language that has been passed down intergenerationally. In the same vein, we do not want to impose grammatical categories that are not correct or applicable.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "1) Example", "sec_num": null }, { "text": "To that end, we worked with highly-proficient, specially-selected M\u0101ori speakers and linguists who are specialists in M\u0101ori. This has ensured that our POS labels and guidelines conventions faithfully reflect a M\u0101ori speaker's conceptualization of their language.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "1) Example", "sec_num": null }, { "text": "We achieved this by simply asking speakers. We elicited answers without using questions that were influenced by academic theories of language or pedagogical methods of language teaching. The speakers reviewed our guidelines on a regular and consistent basis, they also partook in a survey to target special areas of interest. Furthermore, our guidelines are evergreen, meaning that they can and do change based on speaker feedback. This does not impact negatively on our tagged corpora as we have an automation system in place to retag words when necessary. We now briefly explore each point above in (2) seriatim.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "1) Example", "sec_num": null }, { "text": "As mentioned above, some existing definitions and their guidelines for both syntactic categories and POS labels are not suitable for M\u0101ori.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Existing categories are not suitable for te reo M\u0101ori.", "sec_num": "2." }, { "text": "The UD conventions follow a lexical approach, that is one-word equals one-tag. However, as mentioned previously, M\u0101ori is a highly analytic language in the sense that there are many words with multiple grammatical functions, as opposed to inflection. Sometimes a single concept is represented by many lexical words, see 3. Therefore we worked with our speakers to see when and where single or multiple labels were appropriate. b) Made up of many separate words, \"i\", \"te\" and \"r\u0101\" and \"nei\" c) Other, please elaborate", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Existing categories are not suitable for te reo M\u0101ori.", "sec_num": "2." }, { "text": "We achieved this by asking non-leading questions. For example, in order to establish if the words of i te r\u0101 nei, meaning today, should receive a single or separate tags, we asked questions such as that in (4). If speakers had answered (a), then we could infer that i te r\u0101 nei should receive a single tag. On the other hand, if our speakers had answered (b), then the words should be tagged separately. We also left a blank space in (c) to allow our speakers to provide any other suggestions. As it happened, for time phrase adverbials with many lexical words such as i te r\u0101 nei, our speakers overwhelmingly chose to tag each word separately.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Existing categories are not suitable for te reo M\u0101ori.", "sec_num": "2." }, { "text": "Crucially though, this was not the case for all concepts that were represented by many lexical words, as our speakers indicated that certain types should be tagged with a single word. As such, by working with our speakers we avoided making a blanket judgement and were able to give single or separate tags when and where appropriate, all according to the conceptualization of te reo M\u0101ori by speakers. Some developers of tagging guidelines for other languages choose a blanket approach for this type of problem. For example in the POS tagging of Griko, all apostrophes between words are treated as a single token, (Anastasopoulos et al, 2018). However this was not the right approach for us or te reo M\u0101ori, as evidenced by the fact that our speakers chose both single word and separate word tagging. Moving on, in example (5) tense is marked on the verb hoko with kua. The token hoko is given the POS tag VERB, and the separate tense-marker token kua is given the POS tag AUX. However, tense and aspect can also be marked on numbers in M\u0101ori, Harlow (2015: 256) . This is the case in example (6) wherein wh\u0101, or four, is also marked with the perfect aspect marker kua. This is in the same way that verbs, such as hoko in (5) are tense-marked. This is not limited to te reo M\u0101\u014dri, numbers that behave like verbs are also found in Choctaw and Jarawara (Dixon, 2012) .", "cite_spans": [ { "start": 1044, "end": 1062, "text": "Harlow (2015: 256)", "ref_id": null }, { "start": 1351, "end": 1364, "text": "(Dixon, 2012)", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "Existing categories are not suitable for te reo M\u0101ori.", "sec_num": "2." }, { "text": "Whilst acknowledging that a number can be an \"determiner, adjective or pronoun\". The UD guidelines do not provide for numerals that behave like verbs. Yet, they state that verbs are often associated with \"tense, mood\" and \"aspect\". Therefore, under UD tagset guidelines, these numbers would likely be labelled as VERB.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Existing categories are not suitable for te reo M\u0101ori.", "sec_num": "2." }, { "text": "Notwithstanding, tagging in this way would not be an accurate representation of te reo M\u0101ori. So as the POS gloss in (6) shows, we do not adhere to this. The tense-marked number token wh\u0101 is tagged as numeral/NUM. Whilst, the separate tense-marker token kua is tagged as AUX .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Existing categories are not suitable for te reo M\u0101ori.", "sec_num": "2." }, { "text": "As stated above, UD conventions sometimes do not have a suitable existing category for certain classes of M\u0101ori words. Ergo, we have added POS labels that faithfully capture M\u0101ori, both the grammatical categories and the M\u0101ori view of te reo.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Categories for certain word classes of M\u0101ori does not exist.", "sec_num": "3." }, { "text": "M\u0101ori has a word class commonly known as \"particles\" in linguistic literature, Harlow (2007: 24) . These particles are small words such as an\u014d, iho, noa, p\u016b, tonu etc. Each particle can have meaning and many grammatical functions. Following our own analysis of over ninety particles, we found that grammatically they served many purposes, that their syntactic behaviour is wide, varied and commonplace. As such they do not fall under the remit of any \"traditional\" grammatical categories For example, the \"particle\" rawa can modify nouns, pronouns, verbs, adjectives, numerals and negatives, (Harlow, 2015) . We show a selection of these below. In example (7), rawa modifies the pronoun koutou. Rawa can also modify verbs like hangaia in (8), confirmation that verbal modification is taking place can be gleaned from the passive agreement that takes place on rawatia. The adjective wera is modified by rawa in (9). Whereas (10) and (11), show rawa modifying a negative and question word, i.e. k\u0101ore and aha, respectively. Of course, it is fair to ask why we did not use the UD POS tag \"Particle\", hereafter PART, for te reo M\u0101ori \"particles\". As per the UD guidelines, PART is said to often encode grammatical categories such as \"negation, mood, tense\", see UD guidelines, (References section below). However, crucially the \"particles\" of te reo M\u0101ori do not encode any of these categories. The UD PART tag is also a landing spot for words \"that do not satisfy definitions of other universal parts of speech\". For Indigenous or non-European languages, such as M\u0101ori, this in particular feels unsatisfactory. Rather than providing an accurate tag, anything that is deemed to fall outside of \"universal\" grammar is cast-off into the ambiguous PART category. Therefore, we chose to create a POS tag that would be fitting for this part of te reo M\u0101ori grammar. In a wider context, this fits with our vision statement mentioned above.", "cite_spans": [ { "start": 79, "end": 96, "text": "Harlow (2007: 24)", "ref_id": null }, { "start": 592, "end": 606, "text": "(Harlow, 2015)", "ref_id": "BIBREF4" } ], "ref_spans": [], "eq_spans": [], "section": "Categories for certain word classes of M\u0101ori does not exist.", "sec_num": "3." }, { "text": "It should be noted however, that when and where the UD PART tag was applicable it was used and does appear in our tagset. This is the case for all the UD tags, we did not create new tags just for the sake of it. An example of the PART tag being used in our data is with te reo M\u0101ori words of negation, such as k\u0101ore in (10).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Categories for certain word classes of M\u0101ori does not exist.", "sec_num": "3." }, { "text": "12) K\u0101ore au i haere NEG 1SG PST go PART PRON AUX VERB \"I did not go\"", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Categories for certain word classes of M\u0101ori does not exist.", "sec_num": "3." }, { "text": "There is another class of words for which there is no suitable traditional label. When first-person singular, second-person singular and third-person singular pronouns, i.e. ahau, koe and ia, combine with certain adpositions, i.e. t\u0101, \u0101, t\u014d, \u014d, m\u0101 and m\u014d they combine into a single word, (Bauer, 1997) . These new combinations are concurrently both pronouns and adpositions. This can be seen in example 14wherein t\u014d and ahau have combined into t\u014dku. By contrast, t\u014d does not combine with koutou in (13). These are not very common, but do occur in other languages, such as Irish, where they are commonly called prepositional pronouns. A UD Tagset that was developed for Irish simply tags these as preposition/PREP. Yet, this representation is not as accurate as it could be, they are at once both prepositions and pronouns in the grammar of Irish. Furthermore, the UD guidelines do not provide for such a word class.", "cite_spans": [ { "start": 288, "end": 301, "text": "(Bauer, 1997)", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "Categories for certain word classes of M\u0101ori does not exist.", "sec_num": "3." }, { "text": "With this in mind, we worked with our M\u0101ori speakers and linguists to faithfully capture and represent the equivalent te reo M\u0101ori word class. From working with our M\u0101ori speakers and linguists, it became clear that UD conventions do not have a suitable label for either \"particles\" or \"adposition-pronouns\". As such we created two new M\u0101ori specific labels for our tagset, i.e. modifier/MOD and adposition-pronoun/ADPRON.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Categories for certain word classes of M\u0101ori does not exist.", "sec_num": "3." }, { "text": "As has been said above, some UD conventions do not reflect a M\u0101ori worldview of the M\u0101ori language. For instance, the term M\u0101ori indicates Indigenous to Aotearoa. By contrast, P\u0101keh\u0101 means of European origin, and te reo P\u0101keh\u0101 is the M\u0101ori term for the English language. In our corpus, there are some instances of code-switching between M\u0101ori and English, and also between M\u0101ori and other Polynesian languages.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Categories do not reflect a M\u0101ori worldview of the M\u0101ori language.", "sec_num": "4." }, { "text": "The UD guidelines recommend that foreign words receive the POS label \"X\", however this is problematic for us. Although the English language is not Indigenous to Aotearoa, to label English language words as \"X\" fails to capture the complex bi-cultural reality of modern-day Aotearoa. And to label other Polynesian languages as foreign disregards the historical, linguistic, cultural and genealogical ties among Pacific peoples. If we were to use \"X\" to tag all words that are not in te reo M\u0101ori, then English and other Polynesian languages would be conglomerated, or homogenised, into one group. Furthermore, it also limits the usefulness of our tagger for future applications where these languages are often mixed.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Categories do not reflect a M\u0101ori worldview of the M\u0101ori language.", "sec_num": "4." }, { "text": "This resulted in the creation of two further M\u0101ori specific labels, P\u0101keh\u0101/PAKEHA for English language words, and MOANANUI for the cousin-languages of M\u0101ori. The creation of these P\u0101keh\u0101/PAKEHA and MOANANUI labels, allow us to distinguish other languages from te reo M\u0101ori, without disregarding the connections between the speakers of te reo M\u0101ori and other Polynesian languages.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Categories do not reflect a M\u0101ori worldview of the M\u0101ori language.", "sec_num": "4." }, { "text": "The UD guidelines and tagsets have been used to tag languages where there is code-switching such as Turkish-German and Frisian-Dutch. It is our understanding that in such cases both languages are given UD tags. This approach would not work for us for two reasons. Firstly, as a small M\u0101ori Indigenous organisation, POS tagging English would not be a worthwhile use of our resources. Secondly, while we need to differentiate the other Polynesian languages from te reo M\u0101ori in our data, we would not create a tagset, nor presume to tag them without permission from the speakers of those languages.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Categories do not reflect a M\u0101ori worldview of the M\u0101ori language.", "sec_num": "4." }, { "text": "In summation, the words in our M\u0101ori corpora have been categorised and labelled to reflect M\u0101ori in the minds of its speakers. At present, this same M\u0101ori lead approach is being expanded to include a feature layer that would include features relevant to M\u0101ori such as kupu mino and te reo \u0101-k\u0101inga which are similar but different to loanwords and dialect respectively. Even at the most surface level of our tagging conventions, we do not use terms like dialect, when they are not appropriate to Ma\u014dri society.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Categories do not reflect a M\u0101ori worldview of the M\u0101ori language.", "sec_num": "4." }, { "text": "Our tagset uses a total of 21 POS labels. They have been used to annotate our datasets, which contain over 40,000 tokens. The datasets cover many genres and are being constantly expanded. We have used our tagset and annotated datasets to build Whakairo Kupu, our POS tagger for te reo M\u0101ori. In our most recent Whakairo Kupu model, the precision was 92.5%, and the recall was 93.1%. These increased from 86.3% and 48.3% respectively in the very first model. With regard to sharing our data, or allowing the use of Whakairo Kupu, Te Hiku Media operates under its Kaitiakitanga Licence. This quotation in (15) from our Papa Reo website best explains it. For more about the Kaitiakitanga Licence see our Papa Reo website (References section below).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion", "sec_num": "5." }, { "text": "Kaitiakitanga licence, which states that data is not owned but as cared for\u2026 Te Hiku Media are merely caretakers of the data and seek to ensure that all decisions made about the use of that data respect it's mana and that of the people from whom it descends\u2026M\u0101ori data will not be openly released, but requests for access to the data, or for the use of the tools developed under the platform, will be managed using tikanga M\u0101ori.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "15) Te Hiku Media have developed a", "sec_num": null }, { "text": "In terms of applications for Whakairo Kupu, as it stands, not only does it POS tag te reo, but it has been used to build a grammar checker. It is also being used as a foundation for building a Named Entity Recognition tagger for te reo M\u0101ori. ", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "15) Te Hiku Media have developed a", "sec_num": null } ], "back_matter": [], "bib_entries": { "BIBREF0": { "ref_id": "b0", "title": "International Conference on Computational Linguistics", "authors": [], "year": null, "venue": "", "volume": "", "issue": "", "pages": "2529--2539", "other_ids": {}, "num": null, "urls": [], "raw_text": "International Conference on Computational Linguistics, pages 2529-2539 Santa Fe, New Mexico, USA.", "links": null }, "BIBREF1": { "ref_id": "b1", "title": "The Reed Reference Grammar of M\u0101ori", "authors": [], "year": 1997, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Bauer, Winifred. William Parker Te Kareongawai Evans, Te Aroha Noti Teepa. 1997. The Reed Reference Grammar of M\u0101ori. Auckland: Reed Books.", "links": null }, "BIBREF2": { "ref_id": "b2", "title": "Basic Linguistic Theory 3", "authors": [ { "first": "R", "middle": [ "M W" ], "last": "Dixon", "suffix": "" } ], "year": null, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Dixon, R.M.W. Basic Linguistic Theory 3. Oxford: Oxford University Press.", "links": null }, "BIBREF3": { "ref_id": "b3", "title": "Rapanui. London: Routledge Harlow, Ray", "authors": [ { "first": "V", "middle": [], "last": "Du Feu", "suffix": "" } ], "year": 1996, "venue": "M\u0101ori A Linguistic Introduction", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Du Feu, V. 1996. Rapanui. London: Routledge Harlow, Ray. 2007. M\u0101ori A Linguistic Introduction. Cambridge: Cambridge University Press.", "links": null }, "BIBREF4": { "ref_id": "b4", "title": "A M\u0101ori Reference Grammar", "authors": [ { "first": "Ray", "middle": [], "last": "Harlow", "suffix": "" } ], "year": 2015, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Harlow, Ray. 2015. A M\u0101ori Reference Grammar. Wellington: Huia Publishers.", "links": null }, "BIBREF5": { "ref_id": "b5", "title": "The Raup\u014d Phrasebook of Modern M\u0101ori", "authors": [ { "first": "Scotty", "middle": [], "last": "Morrison", "suffix": "" } ], "year": 2011, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Morrison, Scotty. 2011. The Raup\u014d Phrasebook of Modern M\u0101ori. Auckland: Penguin Group NZ", "links": null } }, "ref_entries": { "FIGREF0": { "type_str": "figure", "uris": null, "num": null, "text": "definition of, or guidelines for, an existing category is not suitable for M\u0101ori. b) They do not have an existing category for certain word classes of M\u0101ori. c) They do not reflect a M\u0101ori worldview of the M\u0101ori language." } } } }