Creating and Loading Holding Data¶

This tutorial will show how to load objects representing judicial holdings into AuthoritySpoke. First we'll see how to create an instance of the Holding class using Python commands in a Python shell or notebook. Then we'll create a second holding in a separate text file in the YAML format, and we'll load that YAML file into AuthoritySpoke. We'll see that when AuthoritySpoke loads holdings from YAML, those holdings can be written in a more succinct format with more abbreviations. Then we'll demonstrate that the two Holdings we created can be used together even though they were created with different techniques.

The case we'll be using for this demonstration is United States v. Mazza-Alaluf, 621 F.3d 205 This is a 2010 case from the Court of Appeals for the Second Circuit. It describes how a violation of a state's licensing requirement for money transmitting businesses can result in a felony conviction under federal law.

We'll start by loading a copy of the judicial opinion from Harvard's Caselaw Access Project (CAP). AuthoritySpoke includes a download client for getting these opinions. To download full cases from CAP, you'll need to register for a CAP API key. Once you have an API key, you can make it available as an environment variable by saving it as a file called .env. This process is described in the python-dotenv documentation.

Next, we'll create a CAPClient object, which is a download client for getting decisions from the Caselaw Access Project. It has methods to "fetch" decisions, which means to get them in standard JSON format, and also methods to "read" decisions, which means to fetch them and then convert them into AuthoritySpoke Decision objects. Since we know the citation of the case we want, we'll use the read_cite method.

Because we used full_case=True, we have the option to view the full text of the majority opinion using the command licensing_case.majority.text.

Creating Holdings with Python¶

Now we'll try creating a Holding from the Mazza-Alaluf case using Python commands. One main idea from this case is that certain violations of state law can establish an element of a federal criminal offense called "conducting a money transmitting business without a license required by state law". To model this concept in AuthoritySpoke we need to make it procedural. In the context of a litigation process, the Holding tells us something about how one factual finding can lead to another.

There's still something missing from the object we're calling new_york_holding. We need an object that represents the statute being interpreted by the court. To get that, we're going to use the Legislice API, so we're going to need a second API client class, called LegisClient. While CAPClient was for getting court opinions, LegisClient is for getting legislation. Once again we need to sign up for an API token, then save that API token in the .env file, and then load the API token using dotenv.

Now we can use the LegisClient to download statutes from the United States Code. We do this by specifying a URL-like path to the statute we want. (These identifiers are used by the US government's United States Legislative Markup format.) Here, we want part of United States Code title 18, section 1960(b)(1). The path to that provision is "/us/usc/t18/s1960/b/1".

We don't have to use the entire text of this statute provision. Instead we can select just the part of the text we want. Using the end parameter, we can indicate that we want everything through the string we've identified as the end, but that we don't want anything past that.

One way to add this Enactment to the Holding is by using the addition operator (the plus sign).

Now we've created a computable model of a judicial holding with Python. But there's a problem: Python is a programming language, not a data storage format (unless we wanted to store our data in Python's pickle format, which is usually not the best choice). We want a way to store data about legal doctrines without writing Python commands or running a Python interpreter. So in the next section, we'll try creating a second holding in YAML, a structured data format based on readable text files.

Creating Holdings as YAML Data¶

AuthoritySpoke's YAML data import workflow is designed for creating summaries of Holdings by hand, and then loading them into AuthoritySpoke to convert them into computable Python objects. Under the hood, the data import script will first convert the YAML file into JSON, and then it will load the data into AuthoritySpoke using a JSON schema. If you're not creating data by hand (for instance, if you're passing JSON data from a web API into AuthoritySpoke or vice versa) then you probably will be working with JSON directly and you won't need to use the YAML data format.

Similar to JSON, a YAML file can be converted into a structure of nested Python dictionaries and Python lists. A YAML file that AuthoritySpoke can understand should start with holdings: on a line by itself followed by an indented list of summaries of holdings. YAML uses whitespace and hyphens to represent the structure of the data fields in the file.

The holdings Field¶

In YAML, a list is indicated by putting a hyphen before every item of the list. The holdings field should contain a list of Holdings, but each Holding itself has multiple fields. Inserting a hyphen before one of those fields indicates where one Holding ends and the next begins. For instance, the YAML file would have this structure if it contained two Holdings, and each Holding had fields named "inputs", "outputs", and "enactments".

holdings:
  - inputs:
    outputs:
    enactments:
  - inputs:
    outputs:
    enactments:

For brevity, our first example YAML file is only going to contain one Holding.

Factors and Entities in AuthoritySpoke YAML¶

The fields inputs, outputs, and despite should each contain a list of Factors. (If any of them is an empty list with no Factors, then it can be omitted.) A Factor can be any of several types, including Fact, Evidence, Exhibit, Pleading, or Allegation. Let's add one Factor to the inputs field of the first Holding.

holdings:
  - inputs:
      - type: fact
        content: "{Mazza-Alaluf} operated {Turismo Costa Brava} without an appropriate money transmitting license in a State where such operation was punishable as a misdemeanor or a felony under State law"
        anchors: we conclude that sufficient evidence supports Mazza-Alaluf's convictions under 18 U.S.C. § 1960(b)(1)(A) for conspiring to operate and operating a money transmitting business without appropriate state licenses.

So this Factor has "type" Fact, it has some content, and it has an "anchors" field. The purpose of the "anchors" field is to indicate what text in the opinion the factor should be linked to (for instance, if the factor is being displayed visually as an annotation to the opinion). In this case, we simply placed the the full text where the anchor should be placed. However, as we'll see later, we also could have used the TextQuoteSelector syntax from the anchorpoint module.

The pairs of curly brackets in the "content" field above also have special meaning. A bracketed phrase in a content field identifies an Entity. Typically an Entity is a person or party, but important objects or concepts can also be labelled as class Entity. If you identify a phrase as an Entity by putting brackets around it, the parser will recognize that phrase as the same Entity every time the phrase appears, even if we don't put brackets around the other instances. So when we choose the name of an Entity, we need to make sure the name is a unique word or phrase that always refers to the same Entity whenever it appears in the file.

If we need to include a bracket at the beginning or end of the text in the "content" field, then we also need to put quote marks around the text so the brackets won't be the first character. If the quote mark is missing and a curly bracket is the first character of the text field, then the parser won't understand that the field is supposed to be text.

Facts can also have truth fields. For instance, because this Fact contains truth: false, its meaning is reversed, so it now means "it is False that Turismo Costa Brava was a domestic financial institution".

  - type: fact
    content: Turismo Costa Brava was a domestic financial institution
    truth: false

Enactments in AuthoritySpoke YAML¶

Here's the enactments field from the main Holding in the Mazza-Alaluf case. It only contains one Enactment:

    enactments:
      - node: /us/usc/t18/s1960/b/1/A
        anchors: state money transmitting licenses, see |18 U.S.C. § 1960(b)(1)(A)|

(In this example, the enactments field isn't prefixed with a hyphen, because it's not the first field of a new Holding. However, the node field is prefixed with a hyphen, because it is the first field of a new Enactment.)

The node field indicates the location of the statute text being cited: USC title 18, section 1960(b)(1)(A). (The AuthoritySpoke API currently only serves citations to the US Constitution or the United States Code.) Like Factors, Enactments can also have anchors. This time, the anchors field contains added "|" characters called pipes, that look like vertical lines. These pipes are part of an optional shorthand syntax for locating a text passage within the Opinion. The purpose of the pipe characters is to split the quotation into a "prefix" to the left of the first pipe, an "exact" text, and a "suffix" to the right of the second pipe. Only the "exact" text is considered to be the anchor for an annotation showing were the Enactment can be found. The reason for also including a prefix and suffix is to make the text selector unambiguous. If the "exact" text of the anchor is the same as text that appears somewhere else in the opinion, then the text anchor can become unique by adding a unique prefix. Because the second pipe in the anchors string has nothing after it, there is no suffix for this text selector.

Instead of using the pipe syntax, enactments can also identify only part of the text of a provision with "prefix", "exact", and "suffix" fields, using the TextQuoteSelector data standard. Here's an example:

    enactments:
      - node: /us/usc/t18/s1960/b/1/A
        anchors: 
          prefix: state money transmitting licenses, see
          exact: 18 U.S.C. § 1960(b)(1)(A)

The suffix field has been omitted because a suffix isn't needed to make the text selector unique.

Even though the enactments field only contains one Enactment, the enactments_despite field contains one additional Enactment. This means that the Holding applies "despite" the Enactment in that field. In other words, the Holding rejects any argument that the Enactment will change the outcome. This Enactment is a provision from Title 31 of the United States Code, while the other Enactment was from Title 18.

    enactments_despite:
      - node: /us/usc/t31/s5312/b/1
        name: domestic institution statute

Holding Anchors¶

The holding also contains an anchors field that isn't nested inside any Factor. This field represents the text anchor for the holding itself. If this field is included, the holding anchor should be the place in the opinion where the court indicates that it's endorsing the legal rule stated in the holding, and accepting it as binding law in the court's jurisdiction.

    anchors:
      - prefix: Accordingly, we conclude that the
        suffix: In any event

This time the anchors field uses another slightly different format. The prefix and suffix for the text quote selector are includes as separate lines in the YAML file, but the exact text of the anchor passage has been omitted. Alternatively, we could have included the prefix and exact fields, but omitted the suffix. We just need to include enough information so the text selector can only possibly refer to one location in the court opinion.

Booleans in YAML¶

Holdings can also contain three true/false fields describing the legal doctrine supported by the holding. All three of these fields default to False, so they only need to be included in the file if they need to be set to True. The fields are:

universal: whether the Holding applies in "all" situations where the inputs are present

mandatory: whether the court "must" impose the results described in the "outputs" field when the Holding applies. (In other words, "mandatory" means "not discretionary")

exclusive: whether the inputs described by the Holding are the only way to achieve the outputs. (For instance, if a Holding describes the elements of a crime, it might also say that committing the elements of the crime is the "exclusive" way for a person to be guilty of the crime.)

Here's the complete holdings field of the YAML file, with all the Factors filled in. Two boolean fields appear at the end.

holdings:
  - inputs:
      - type: fact
        content: "{Mazza-Alaluf} operated {Turismo Costa Brava} without an appropriate money transmitting license in a State where such operation was punishable as a misdemeanor or a felony under State law"
        anchors: we conclude that sufficient evidence supports Mazza-Alaluf's convictions under 18 U.S.C. § 1960(b)(1)(A) for conspiring to operate and operating a money transmitting business without appropriate state licenses.
      - type: fact
        content: Mazza-Alaluf operated Turismo Costa Brava as a business
        anchors: Mazza-Alaluf does not contest that he owned and managed Turismo
      - type: fact
        content: Turismo Costa Brava was a money transmitting business
        anchors: record evidence that Turismo conducted substantial money transmitting business in the three states
    despite:
      - type: fact
        content: Turismo Costa Brava was a domestic financial institution
        truth: False
        anchors: without respect to whether or not Turismo was a "domestic financial institution"
    outputs:
      - type: fact
        content: Mazza-Alaluf committed the offense of conducting a money transmitting business without a license required by state law
        anchors: a crime to operate a money transmitting business without appropriate state licenses,
    enactments:
      - node: /us/usc/t18/s1960/b/1/A
        anchors: state money transmitting licenses, see |18 U.S.C. § 1960(b)(1)(A)|
    enactments_despite:
      - node: /us/usc/t31/s5312/b/1
        anchors:
          - § 5312(b)(1) (defining "domestic financial institution")
    anchors:
      - prefix: Accordingly, we conclude that the
        suffix: In any event
    universal: true
    mandatory: true

Loading Holdings from YAML¶

Let's save the example YAML above to a file, and then load the file with AuthoritySpoke. Let's say the YAML file will be called myfile.yaml, and the path to that file from this notebook will be path/to/myfile.yaml. In order to load not just the Holdings but also the text anchors, we'll load the file using the read_holdings_with_anchors function.

Next, we'll print the holding we loaded to see how AuthoritySpoke interpreted the YAML file.

The Holding that we created in Python and the Holding that we created in YAML are both valid AuthoritySpoke objects. We can demonstrate this by adding the two Holdings together to make a combined Holding that uses information from both of them.

By combining the two Holdings, AuthoritySpoke has inferred that the Fact that a defendant committed the New York offense can substitute for the Fact that the defendant operated "without an appropriate money transmitting license in a State where such operation was punishable as a misdemeanor or a felony under State law". If the former Fact is available, then the offense can be established even if the latter Fact hasn't been found yet.

Assigning Names to Factors and Enactments in YAML¶

The YAML data input format is still verbose, but there's one more feature we can use to shorten it. If a Factor or Enactment is going to be used more than once in the same file, we can add a "name" field to the YAML for that object. Then, anytime the same object needs to be reused, we can just write the object's name rather than rewriting the whole object. Similar to the names of Entities as discussed above, any names assigned to Factors or Enactments need to be unique in the file. They should only appear in the text of a content field if they're intended to refer to the Factor or Enactment that has been given that name.

Here's an example where the Holding we've been calling holding_from_python has been rewritten into the YAML file. Inside that Holding, one Factor and one Enactment have been assigned names. Then, later in the file, the Factor and Enactment are referenced by name instead of being rewritten in full. Here's the entire updated YAML file. (This is the same as the file example_data/holdings/holding_mazza_alaluf.yaml.)

In the YAML above, a Factor is assigned the name "operated without license", and then the second time the Factor is used, it's referenced just by the name "operated without license". In the same way, an Enactment is assigned the name "state money transmitting license provision".

Now when we load a file with this YAML, we'll get both Holdings.

Now that we generated this AnchoredHoldings object containing the data from the YAML file, we can use the posit method to link those Holdings to the judicial Decision we created from the data we downloaded from the CAP API. Then we can verify that those two Holdings are now considered the two holdings of the Decision.