I'd suggest defining two input objects, the (optionally named) character vector as it is now, and a data.frame with a named text column. This conforms to the Text Interchange Format and is nearly costless to implement.

similar to #6, reliabal should be reliable

While each of these is a breaking change, it's also confusing when code doesn't work because you spelled the word correctly.

Should be classification.

Hi there! Thank you so much for implementing this wrapper! It works so nicely on R. I was just wondering if there was anything we could do to improve its accuracy especially with languages that are very close to each other (for instance Galician -gl- and Spanish -es-)

I ran the detect_language function on around 2,000 titles and it "detected" around 60 as Galician and they were all in Spanish. Here's a sample of those misidentified titles:

1. Los albores de la sinología occidental: la contribución de Fernández de Navarrete
2. Herencia y cuidado: transiciones en la obligación filial
3. Análisis de la tendencia e impacto de la mortalidad por causas externas: México, 2000-2013
4. Significado, relevancia y elementos de género asociados al cuidado: metasíntesis cualitativa
5. Algunas reflexiones metodológicas al abordar experiencias reproductivas de los varones desde las políticas públicas
6. Bienestar financiero, una reflexión desde la ficción neoliberal en un contexto local
7. Orientaciones para las políticas públicas de juventud: una revisión documental
8. Muros y migración México-Estados Unidos
9. Brecha salarial por género en México: Desde un enfoque regional, según su exposición a la apertura comercial 2005-2015
10. Las Salvaguardias y su impacto en sector comercial de Ecuador

I was curious if you might know how the algorithm works and if there was any way to improve it :)

So it could work with character vector instead of single string.

Hi pycld2 team!

Just wanted to let you know that @quinnanya used the pycld2 binding to detect language (see her csv) on the same Conference Titles for which I used your Rstats wrapper, and funnily enough we got different results, any idea why?
Using the Rstats wrapper I get more "french" tags:
32 of those "french" tags where marked as NA with Python, 8 w/un, 3 as en.
I manually checked those 3 pycld2 English tags that had a "fr" tag with the Rstats wrapper and one is indeed French, the other is Frenglish, and the last one is English

I'm not sure if this is a problem with the R package or the underlying library.

For some URLs, detect_language (and detect_language_multi) consistently cause a segfault.

> detect_language( url( "http://thehill.com" ), plain_text=FALSE )

 *** caught segfault ***
address 0x7f4dc86b66a4, cause 'invalid permissions'

Traceback:
 1: detect_language_cc(as_string(text), plain_text, lang_code)
 2: detect_language(url("http://thehill.com"), plain_text = FALSE)

I also get this error on http://qq.com.

Platform details:
R version 3.5.0 (2018-04-23) -- "Joy in Playing"
Copyright (C) 2018 The R Foundation for Statistical Computing
Platform: x86_64-pc-linux-gnu (64-bit)

Running under Arch Linux, and the most recent cld2 installed via devtools.