Using feature set m*_gram_dif

Introduction

The m*_gram_dif features quantify grammatical differences between the Morpheus analyses and the N1904-TF annotations, which in this context is regarded the reference. Each feature in this group annotates a wordnode with a 9-character string indicating whether Morpheus’s analysis differs from the reference. Differences are indicated using a letter while matches are indicated usint a dot (.).

Related features

Feature(s)	Description
md{ind}_gram_dif	grammatical difference against N1904-TF, with num in the range (1-24), inclusive.
ms{ind}_gram_dif	grammatical difference against N1904-TF, with num in range (1-12), inclusive.
mm_gram_dif	grammatical difference against N1904-TF

Data Format

Each value for a feature m*_gram_dif produces a fixed-length string of 9 characters. Characters are drawn from:

• A dot (.) when Morpheus and N1904-TF agree (including when both values are None).
• A letter (one of l p c n g t m v d) when Morpheus differs, indicating which category is mismatched.
• The character positions correspond to the canonical order of morphological features shown in the following table.

Position	Field Key	Grammatical Property
0	l	Lemma
1	p	Person
2	c	Case
3	n	Number
4	g	Gender
5	t	Tense
6	m	Mood
7	v	Voice
8	d	Degree

Example usage

Some notes on interpretation

The compare will mark as a difference any grammatical property that can be interpreted differently. In the following case the Morpheus parsing does return the tag from N1904-TF, but also another one. This is due to underspecification (the ending ος can be either masculine or feminine).

In case the morphological tag differs between N1904-TF and the Morpheus derived data, this feature can still report ........., in case the tag difference is due to e.g., dialect. See the following example:

The situation may be even more tricky. For the following wordform the N1904-TF has morphologicaly tagged Isaak as N-PRI, just as it was done on the Morpheus based dataset. However, the is a mismatch reported on the grammatical case. This occurs because all TF features in the *_gram_dif group are computed by comparing each grammatical property in a Morpheus analytic block with its counterpart in the N1904-TF dataset. As is clear from the Morpheus analytic block in Figure 20, no information about grammatical case is provided. As a result, the compare function flags this as a difference, since such information is present in the N1904-TF dataset. This can be easily seen when comparing the N1904-TF data with the raw data from Morpheus.