With Prof. Comrie.

Pre-Meeting

Questions

1. We are looking for more nuanced ways to refer to languages that are syntactically similar to English, and those that aren't. Does Prof. Comrie have any suggestions?

   Possible terms for "English-similar":

   • prestige language variety
   • indo-european languages (is this true for all the languages we categorize as "English-similar": English, German, French, Italian, Spanish, Russian)

2. We do not know the dialects, or even the specific countries of origin, of the text in each language. Could this potentially affect the conclusions we draw about them, e.g. should we state some sort of uncertainty because of this? Or are most semantic differences between these languages independent of dialect?

3. One of the areas we have been looking into is inherent tokenization, since this affects the ability of words to easily be captured in word embeddings. Do any of our languages have varying degrees of tokenization? Specifically, how do they differ?

4. Since our hypothesis is that languages model the best when they are the closest to English, how does he recommend we define "similarity" to English? Is language family an accurate partition for this basis, or does this miss language-specific details?

5. We are looking for a measure of lexical similarity by which to define English-similarity. Is there anything he can recommend? (Something in the realm of lexicostatistics). By what factor of lexical similarity (historical etymology, diffusion / language borrowing, random variation) is this measure?

6. Terms we want clearer definitions of (from a linguistic context):

   • polysynthetic languages (and do any of our languages qualify as this)?
   • typology
   • morpheme
   • morphology
   • semantic
   • etymology

Additional notes

• Languages used: English, German, French, Spanish, Italian, Russian, Chinese, Japanese
• We do not know the dialects of the languages, and there may be different dialects for each language; we also do not know the countries of origin for each of the languages, but can conjecture based on the locations of Reuters offices: the Netherlands, Taiwan, Portugal, Spain, Denmark, Norway, Sweden, Latin America (sic; Argentina?), France, Italy, Germany, and Japan.