Michael Campbell, 28 Nov. 2022, 10:00 (IST)
Using Syntactic Structures
Using Syntactic Structures to Aid Speaking/Listening Acquisition of the World's Languages on Glossika
Language: English
It's possible to tag sentence structures universally across languages to real world events via first order logic.
Grammatical labels differ from language to language (e.g. perfect/perfective/aorist), even when semantics is the same, due partly from how much data is encapsulated in a token and also from forcing the subject-object paradigms of modern languages to express real-world ergative logic. The use of differing grammatical labels leads to frustration and difficulty for students acquiring foreign languages. We reduce these differences to a common denominator by using 90 syntactico-semantic labels taking into account telicity, valencies, causation, light verbs, aspect, etc.
By sorting this uniform logic across languages in any language family, the grammatical patterns of each language emerge and align, which present easily trainable patterns for students, thus resulting in easier acquisition.
Glossika has developed a proprietary tag set for our language learning solutions, and we see it as an enormous improvement over antiquated industry standards like the over 30-year-old Penn State Treebank and other competing treebanks. Moving forward, the machine learning community will also benefit from using these training datasets to expand their services beyond the bottlenecks that eurocentric part-of-speech tagged datasets cause. This in turn leads to less training time and fewer cross-linguistic mistakes and misunderstandings, even among European languages.
Michael Campbell is founder of Glossika, and for three decades has researched diachronic dialectology (Sinitic, Indo-European, Austronesian), semanto-syntactic mapping to first order logic, vocabulary acquisition across language families, long-range etymologies, human migration and how language families arose.
It's possible to tag sentence structures universally across languages to real world events via first order logic.
Grammatical labels differ from language to language (e.g. perfect/perfective/aorist), even when semantics is the same, due partly from how much data is encapsulated in a token and also from forcing the subject-object paradigms of modern languages to express real-world ergative logic. The use of differing grammatical labels leads to frustration and difficulty for students acquiring foreign languages. We reduce these differences to a common denominator by using 90 syntactico-semantic labels taking into account telicity, valencies, causation, light verbs, aspect, etc.
By sorting this uniform logic across languages in any language family, the grammatical patterns of each language emerge and align, which present easily trainable patterns for students, thus resulting in easier acquisition.
Glossika has developed a proprietary tag set for our language learning solutions, and we see it as an enormous improvement over antiquated industry standards like the over 30-year-old Penn State Treebank and other competing treebanks. Moving forward, the machine learning community will also benefit from using these training datasets to expand their services beyond the bottlenecks that eurocentric part-of-speech tagged datasets cause. This in turn leads to less training time and fewer cross-linguistic mistakes and misunderstandings, even among European languages.
Michael Campbell is founder of Glossika, and for three decades has researched diachronic dialectology (Sinitic, Indo-European, Austronesian), semanto-syntactic mapping to first order logic, vocabulary acquisition across language families, long-range etymologies, human migration and how language families arose.
