Syntactic Fieldwork for a Field Methods Course

In this blog post, I will lay out some general guidelines on how to elicit syntactic information in a Field Methods course. 

We will use the term target language for the language being investigated, and meta-language, for the language of communication with the consultant (if different from the target language). For example, in a recent Field Methods course at NYU (Fall 2025), the target language was Oshiwambo and the meta-language was English.

While the guidelines below are written specifically for a Field Methods course (for both undergraduate and graduate students), in most cases they extend naturally to fieldwork done in the field outside of an academic setting.

Divide and Conquer

When working on syntactic elicitation, try to work things out slowly from the bottom up. Solve simpler problems before complicated problems. At each point, you should break down the problem into its component parts, and work out each part individually. 

It is tempting when starting to look at syntax to just dive into complex sentences right away to see how they work, such as “When I was walking home, I saw those big cats on the road.” But to understand this particular sentence it is necessary to have a basic understanding of the first-person singular pronoun, demonstratives, attributive adjectives, plurals, and the past tense. Also, the language may have a complicated system of agreement (as in the Bantu languages), so the agreement paradigms must also be carefully worked out. After working out the basics, you can finally return to the syntax of temporal adjunct clauses such as “When I was walking home…”.

The lesson is that at the beginning of the Field Methods course, you should focus on basic topics that will form the basis of more complicated investigations later. 

Set Reasonable Goals

A big problem that beginning students have is that they plan way too many topics in their elicitation sessions. They usually include about three times as many questions as they will be able to address. For example, if your goal is to look at the subject pronominal paradigm (already a tricky task), then you should not also plan to elicit reflexive and reciprocal pronouns in the same session as well. 

Time is limited, and you will never get to all those topics. If you overplan, you risk rushing your session to try to collect all the information specified.

Rather, you should think carefully about the topic that you plan to work on, thinking about different kinds of complications that could come up, and how you will address them. For the subject pronominal paradigm, consultants will often give the wrong pronoun. For example, for the question “How do you say ‘You are working’ in your language?” The consultant might answer “I am working.” To address this kind of confusion, you need to carefully set up contexts that make the pronominal reference clear. 

Your Consultant is not Your Colleague

Don’t treat your consultant like you would treat your linguistic colleagues. 

Do not have the consultant do the linguistic analysis for you. Do not ask questions like: Is this a noun? Is this plural? Is this present tense? Does the language have a gender system? Do these two sounds form a syllable? Instead of these questions, create tests that get to information that you want to obtain. For example, instead of saying “Is X plural?”, see if X can be used to describe more than one object (by first giving a background context).

A consultant will often have linguistics insights into his/her own language. It is important to write these down, whenever they are offered. But refrain from making the consultant do linguistic analysis for you.

Check and Recheck

When working with one consultant, as we do in a field methods course, all data obtained should be confirmed at a later date. Such a confirmation will allow you to see if the consultant really understood the nature of the task the first time around. In fact, the data that you elicit should be checked and rechecked many times in order to ensure it is accurate. It is better to obtain a small amount of solid data, than large amounts of data that will not hold up later on.

Minimal Pairs

If you are trying to establish some syntactic fact, you want to reduce the variation in your sentences as much as possible. Here is an example:

(1) a. I saw those cows.

b. *All that cows is really interesting.

In example (1a), we have a plural noun and a plural demonstrative. In (1b), that does not agree with cows, and the sentence is bad. However, there are so many differences between (1a) and (1b) that it is impossible to know exactly what is going on. A better pair of sentences would be:

(2) a. I saw those cows.

b. *I saw that cows.

There is precisely one difference between these sentences, and so it is plausible to attribute the unacceptability of (2b) to the non-agreeing distal demonstrative. Similar remarks hold for the construction of whole paradigms (lists of several sentences illustrating some point about syntax).

Translation Task

In the translation task, a sentence (the prompt) is formulated in the meta-language (e.g., English), and the consultant is asked to translate it into the target language. The translation task is a rich source for syntactic data. In fact, for some of my elderly Khoisan consultants, who are not able to give syntactic acceptability judgments, the translation task is one of my main sources of data.

For the purposes of syntax, the translation task yields a grammatical target language sentence. On the basis of this sentence, you can start to formulate the syntactic rules of the language (e.g., the position of negation, the pronominal paradigms, etc.). But you must be aware of the fact that establishing the exact meaning of a sentence is quite difficult and involved (see Mathewson 2004). So the translations obtained must be viewed as first approximations.

In general, except for very simple sentences, you should always provide a context with the sentence to translate. The context may be provided in the meta-language (in some cases accompanied by a visual presentation) and it should be given before the sentence to translate is presented. The context should always be included in the write-up of the elicitation session (the Elicitation Report).

When asking for translations, ask for whole sentence translations. This will help you to see what is going on in the phrase you are translating. This advice is particularly useful for people working on DP structure, where there are lots of phrases of the type “three men”, “three big men”, those three men”, “the people that I saw”, etc. All of these should be elicited in complete sentences, such as “I see three men.”

The sentences you elicit should be as simple and as natural as possible. The simpler the sentence is, the easier it will be to understand it, and to obtain minimal pairs. If for any reason, the consultant hesitates at the naturalness of the sentence that you elicit from them, then try to find a more natural sentence. Try to find sentences that are simple, natural and easy to understand.

Often, you can increase the chance of getting a good translation by adding some linguistic context to the sentence itself. For example, consider the following question posed to a consultant:

(3) How do you say “John runs” in your language?

The problem with this question is that “John runs” in English is usually used for the habitual (“John runs every day”). However, many consultants will translate this with a sentence in their language that has a present progressive interpretation, since they did not understand that the English sentence was habitual. In this case, an easy way to elicit the proper information is just to add an adverb (which is a kind of linguistic context):

(4) How do you say “John runs every day” in your language.

The adverb in (4) forces the present habitual interpretation. Other adverbs can force other interpretations.

Even with the controls above, a disadvantage of the translation task is that it crucially implicates the meta-language, leaving open the concern that properties of the meta-language are influencing the way that the consultant uses the target language. For example, if you are investigating word order in the target language, and you produce an English sentence to be translated, it is possible that the translation is mimicking the word order of English.

Back Translation Task

In the back translation task, you have a sentence of the target language and you ask the consultant to translate it back into the meta-language (English).

The back translation task is often used in conjunction with the translation task, as a kind of check. For example, the back translation task can also be used to check data. Suppose on September 1, 2013, you elicit S in the target language. On October 1, 2013, you might ask the consultant to translate S back into English to see how accurate the original translation into the target language was.

The back translation task can be useful as a check on the acceptability judgment task (see below). Suppose you have altered a sentence S in some way, and you then ask the subject to judge the altered form S’ for acceptability. If the consultant says S’ is acceptable, you can then ask him/her to translate S’ back into English.

Acceptability Judgment Task

In the acceptability judgment task, the target language sentence is presented to a consultant, who is asked to judge it for acceptability. In generative linguistics, we have a number of different ratings for sentences: OK, ?, ??, ?*, *, **, etc. For work with consultants, it is best to keep it simple:

OK: Sounds completely natural and it is something I would say;

?: Sounds kind of odd, but I wouldn’t be surprised to hear someone else say it;

*: Sounds completely wrong and no one would say this.

Whatever system you use, you should explain it to the consultant ahead of time, and give some examples of clearly unacceptable sentences from English to illustrate. Also, sometimes the consultant will give remarks about the sentence such as: “It is fine, but people do not use it much.” You should write down these consultant remarks. They contain lots of information that may be useful down the road.

The acceptability judgment task requires a high level of meta-linguistic awareness. You are asking a consultant to think about their language as a linguist does, and this is not a customary activity for most non-linguists. In fact, I have many consultants, especially older ones, who can do the translation task just fine, but cannot do the acceptability judgment task. 

One issue that often arises with the acceptability judgment task is that the consultant actually judges a different sentence from the one presented to him/her. If you pronounce a sentence X in their language, and X is unacceptable in some subtle way, consultants will often just ignore the unacceptable part, and respond to the sentence as if it were acceptable. One way to control for this is to have the consultant repeat the sentence. If they can repeat the unacceptable sentence X, you know that they are judging exactly the sentence you asked them to judge. Once the consultant becomes familiar with the tasks you are performing, it may be possible to drop this repetition.

Just as in the translation task, it is best to give a context to clarify exactly what the sentence is supposed to mean. This is particularly important for longer or more difficult sentences, but you should get into the habit of providing meta-language contexts for all your sentences.

It is best not to create sentences from scratch to be judged. There are all kinds of ways a particular sentence may be unacceptable or unnatural, and as linguists investigating the language, you may not be aware of all these factors. Try to obtain a sentence first using the translation task (or from oral texts), and then to modify it in some simple way when asking for acceptability judgment. If you need to create a sentence from scratch in the target language, be sensitive to remarks that the consultant makes. It may be possible to change it in some simple way to make it more natural.

Another piece of advice is to not put words in the consultant’s mouth. In other words, don’t ask leading questions. I have heard students ask questions like this many times:

(5) This sentence X is also unacceptable, right?

The form of the question asked, and the use of “right” prompts the consultant to agree that X is unacceptable. We are interested in what the consultant says, not what the linguist wants him/her to say.

When getting acceptability judgments, start with simple sentences and build up (a case of divide and conquer). If you need a sentence like, “John left because his dog left”, start with eliciting the basic vocabulary (e.g., “leave”, “dog”), then move up to larger phrases and finally go onto the whole sentence. This will make transcriptions easier and put less of a load on the consultant (who would have to repeat the long sentence many times before you are able to transcribe it accurately).

Try to process one sentence at a time. I have seen students write 4 or more sentences on the board, then make small subtle changes to each, and ask the consultant to go over all four sentences one right after the other, judging them. Some consultants can do this, but for most this is a herculean linguistic task. 

At all points, simplify the task and the materials to reduce the cognitive burden you are placing on the consultant.

Visual Materials

A useful task in some situations is to ask the consultant to describe some situation or object presented visually. For example, this is the most efficient way to get the full range of locative relations from a consultant (draw a picture of a dog sitting in front of a house, behind a house, next to a house, on top of a house, under a house, etc.). Many such visual materials are now being developed to explore a wide range of topics, including quantifier scope. 

The advantage of collecting data in this way is that there is no influence of the meta-language. The disadvantage is that such a method only allows for the elicitation of a narrow range of data. 

A particular kind of visual material is the storyboard. This is a series of pictures forming a story that the consultant needs to translate, using their own words. Storyboards are a way of obtaining interesting syntactic and semantic information about a language, without relying on the translation task. Many different kinds of storyboards are being developed nowadays.

Oral Texts

When you start to work on a language that is understudied, it is useful to gather a set of oral texts.  These texts give you a feel for the range of constructions that appear in the language. Without such texts, it is sometimes difficult to imagine what kinds of structures are present and need to be investigated. Such texts also provide you with natural sentences that can form the basis of future work using the acceptability judgment task. Such texts can be folktales, anecdotes, life histories, conversations, instructions on how to do something, etc.

In a Field Methods course, it is best to start with a small two- or three-minute oral text. You can arrange a meeting with the consultant to decide on a short topic to record. Nowadays, the recording can be made with an iPhone, and then uploaded to ELAN, where it can be transcribed and translated.

Processing even a small oral text can be quite time consuming. Translating and transcribing a small three-minute text could take several hours of time. But there are significant rewards to this exercise in terms of learning new and interesting things about the target language.

One important point: texts only include positive data (sentences that occur in the texts), and do not include negative data (information about unacceptable sentences). So it is not possible to form a systematic understanding of a language based on texts alone.

Conclusion

There is no single best way to obtain syntactic data. Rather there are a range of techniques each with its own strengths and weaknesses. Part of learning how to do fieldwork is learning the full range of methods. Be open, and try to explore different methodologies.