Monthly Archives: January 2017

Testing #machine translation now facing new elision problem: Riventosa (fr) = A Riventosa(co) proper noun (fr) = definite article + proper noun (co) it should read: in u paese di A Riventosa (without elision) Elision rules are not trivial: le … Continue reading →

Rule-based translation : adjective accordance : interesting stuff: sur les réseaux japonais et américain (fr) = annantu à e rete sgiappunesa è americana (co) = on the japanese and american networks (en) noun (networks) is plural but adjectives (japanese and american) are … Continue reading →

Now handling gender reversal: – mer (FR, feminine) = sea (EN) = mare (CO, masculine) – saveur (FR, feminine) = flavor (EN) = sapore (CO, masculine) – liqueur (FR, feminine) = liquor (EN) = licore (CO, masculine) ‘c’est une bonne liqueur’ … Continue reading →

Introducing new feature for #MachineTranslation: some verbal locutions: prendre d’assaut = assaltà mettre à sac = sacchighjà prendre au collet = incappià

Now considering the issue of Semantical disambiguation. Some instances For French to Corsican are: – ‘défense’ = sanna/difesa = tusk/defense – ‘vol’ = bulu/furtu = flight/theft – ‘comprend’ = capisce/cumprende = understands/comprises – ‘palais’ = palatu/palazzu = palate/palace – ‘expérience’ = … Continue reading →

Now scoring 1 – 6/134 = 95.52%. Lack of vocabulary ‘passacaile’. It should read: ‘versu u 1678‘, ‘di a so epica‘, ‘di u so tempu‘.

Let us mention the issue of threefold ambiguity: french ‘nouvelle’ can translate into: ‘nutizia‘ (‘piece of news’) or ‘nuvella‘ (short story’) or ‘nova‘ (‘new’) The disambiguation between ‘nutizia‘ (‘piece of news’) or ‘nuvella‘ (short story’) is semantic (hard) while the … Continue reading →

Some further reflections on the definition of ‘above human level’ translation: – the answer may not be based solely on the quantitative side, being of the type: ‘above 96%’, “above 97%’, ‘above 98%’, etc. – it seems the answer should also … Continue reading →

Defining an instance of Feigenbaum test (from wikipedia: generally defined as a variant of the Turing test where a computer software attemps to imitate a human expert in a given field): Translating French into Corsican. We expect 98% accuracy and lack of gross errors in order … Continue reading →

Scoring 1-10/159 = 93.71%. Partitive article successfully handled: ‘participe à de nombreuses batailles’ = participeghja à numerose battaglie ‘fournit des renseignements’ = furnisce i rinsignamenti

scoring 1- 4/100 = 96%. It should read: natu …in Yffiniac 216 vittorie frà e quale 144 … eccituatu i criterium

scoring 1 – 5/146 = 96.57% lack of vocabulary: ‘equine’, ‘tracter’, verdine’

Scoring 1 – 3/119 = 97.47% 2 100% consecutive sentences 1 accordance error: sviluppata should read ‘sviluppatu‘, according with ghjocu (game) and not with avventura (adventure)

scoring 1 – 3/119 = 97.47%: would it qualify as a successful Feigenbaum test? The answer is clearly ‘no’, since 2 errors can be identified as gross errors, i.e.some errors that a human would not do: – ‘sviluppata‘ should read ‘sviluppatu‘, … Continue reading →

In the style of ‘I saw wood with a saw’, from French to Corsican: French ‘vis’ is multi-ambiguous: – ‘vis’ (noun singular) = vita = screw – ‘vis’ (noun plural) = vite = screws – ‘vis’ (present 1rst person) = … Continue reading →

Scoring 1- 11/149 = 92.6%. Errors (9/11) come mainly from disambiguation: Disambiguation engine needs to be entirely re-written. An average 2-3% increase in accuracy is expected.

Testing the #semanticdisambiguation of ‘palais’ (EN palace/palate) French ‘palais’ has fourfold ambiguity: – palazzu (EN palace): noun singular – palatu (EN palate): noun singular – palazzi (EN palaces): noun plural – palati (EN palates): noun plural Le palais du calife … Continue reading →

Scoring 1 – 2/138 = 98,55%. Is it a  successful Feigenbaum test? No, at least due to one gross error: ”fils d’un riche’ should translate ‘figliolu d’un riccu‘ ‘fils’ is threefold ambiguous: figliolu (son), figlioli (sons), fili (threads)