Good book to learn the empirical aspects of linguistics, and language families in particular

Question

I wanted to know about how empirical works are done in Linguistics. I have very little idea about this field, but I am interested to know.

I know many empirical "facts" about languages and their families, like my first language, Assamese is an Indo-Aryan language, part of the Indo-European family. My hometown has a lot of people whose first language is Mising, which comes under the Tani languages, of the Sino-Tibetan family. So my language is linguistically closer to German, a language native to a place thousands of miles away, than to Mising, spoken by people living next door. And the classification makes sense: I can understand a little bit of Gujarati or Punjabi spoken 1000+ miles away, because of similarities between Assamese and Gujarati/ Punjabi (both Indo-Aryan languages), but I can't understand my neighbour languages like Mising, Bodo, Mizo etc. at all. This fascinates me a lot!

What is the empirical methodology to construct language-trees like that?

And as the title of the question suggests, while the fascination with language families is one of the primary reasons I want to learn this, firstly, I understand that directly jumping to language-trees may not be possible, and secondly, a general overview of the methods would also be helpful. I have many other questions that interest me, for example, how are two languages considered different (why are Assamese and Bengali different languages, how do we know that Cachar is a dialect of Bengali and not Assamese etc.), how are languages classified as endangered etc.

So maybe a general introduction to statistical methodology in linguistics would help? Maybe an introductory textbook and then an advanced one would help? I have a fair amount of knowledge of statistics, econometrics and psychometrics. I would love a rigorous mathematical treatment, along with say software implementation etc. If there is no comprehensive textbook, a series of textbooks/ articles etc. would also help.

Answer 1

The most important thing to understand is that you have to start with the first step, not the fifth step. The (chronologically) first step is simply saying what the properties of a particular language are. To make this clear, let's introduce a convention that you cannot study your own language, therefore if we (linguists) want to study Assamese, we have to find a person who speaks that language, then devise a program of interviews that will culminate in a higher level conceptual understanding of the properties of Assamese.

It happens, in our study, that we learn that there are multiple speakers of Assamese, so we find a second person who claims to speak the language and discover that his responses sometimes don't match what the first person said. Eventually we discover, having found a hundred people who claim to speak Assamese that they are similar but different, and that there is a significant correlation between what people say, and where they were born. Fortunately, because we have an unbiased theory of how to describe any arbitrary human language, we are able to record these facts. We fare better than a competitor who rejects all of these speakers except certain ones from the eastern area, because the competitor thinks "Those other people don't know Assamese".

Our next step is to understand how it is possible that people speaking a language can be "same, yet different". An intellectual miracle happens, and we discover that languages change over time, therefore the way a language is now is not necessarily how it was 100 or 1000 years ago. We realize that what looks like "just dialects of one language (Assamese)" is a situation that must have existed myriad times in history – we have discovered the idea of language change and differentiation.

Since we can devise many individual descriptions of Assamese (different dialects), we can also try to organize that information into clusters so that we can see what is true of all dialects, versus true of only one dialect, or only two dialects, etc. We may find words pronounced with [x] in most dialects, which are pronounced with [s] in a specific group of dialects. These shared similarities vs. differences (known as "isoglosses") are the foundation of the genetic trees that you seek.

At some point, we expand our study and try to describe Sinhalese, Dogri, Kurdish and Russian. As we progress through that list, we find it harder and harder to discern significant similarities. The problem that we face is that there have been thousands of years of changes to the earlier languages, so that Assamese just simply does not resemble Russian. The next miracle is realizing that we can attempt to reconstruct an earlier form of Assamese, using the modern dialects and we can set forth a conjecture about what the language must have been like, before it split up into dialects. We have discovered the comparative method, where we follow a certain logic to discovering what the earlier state of the language was like. We can do that not just with modern languages, we can do that with the reconstructed languages, and then we see that the languages (now known as Indo-Aryan) originally derived from a more unified language, which split into subgroups. Russian is related to Assamese at a very remote historical level. At a certain point, it is just impossible to reconstruct the earlier state of languages, therefore it is possible that Miso and Assamese are ultimately related many 10's of thousands of years ago, but there is no evidence for that, and we just say that the languages are not related (meaning, there is no proven relation).

The foundation of this enterprise is the ability to neutrally describe the properties of an individual language. If you can do that, then you have some hope of being able to make valid conjectures about the earlier state of some language.