These three approach­es to ana­lyz­ing lan­guage span both word-order depen­dent (Coh-Metrix) and bag-of-word (LIWC, LDA) approach­es. 

Word-order depen­dent approach­es pre­serve the rela­tion­ship between words in a text (sen­tence, para­graph, doc­u­ment, etc.) and can be under­stood in the con­text of syn­tax parse trees. Click below for exam­ple.

The 2011 arti­cle on Coh-Metrix pro­vides a descrip­tion of the fea­tures gen­er­at­ed by this soft­ware. Coh-Metrix was devel­oped at The Uni­ver­si­ty of Mem­phis in the Insti­tute for Intel­li­gent Sys­tems by Art Graess­er, Max Louw­erse, and Danielle McNa­ma­ra. There are five high-lev­el aggre­gate fea­tures gen­er­at­ed by a prin­ci­pal com­po­nents analy­sis (PCA) — these include syn­tax sim­plic­i­ty, word con­crete­ness, nar­ra­tiv­i­ty, deep cohe­sion, and shal­low cohe­sion. Syn­tac­tic fea­tures have been used in many dis­ci­plines to under­stand the con­tex­tu­al aspects of lan­guage such as more con­crete or abstract word choic­es and the use of more or less com­plex phras­es.

LIWC (Lin­guis­tic Inquiry and Word Count) is a dic­tio­nary-based pro­gram devel­oped at UT Austin by James Pen­nebak­er. This pro­gram is a quick and pow­er­ful approach to ana­lyz­ing lan­guage by count­ing the num­ber of words in a text, pri­mar­i­ly closed-class words such as pro­nouns, con­junc­tions, arti­cles, and prepo­si­tions. 

LDA is an open-source process which can be ana­lyzed in many ways, includ­ing R, Python, and many off-the-shelf pro­grams. LDA assigns words into cat­e­gories based on their prox­im­i­ty in the cor­pus, and these terms form “top­ics”. Researchers qual­i­ta­tive­ly assign names to the words in a top­ic, based on their rela­tion­ship and exper­tise in the sub­ject mat­ter.