Natural language processing
) is a subfield of artificial intelligence
and computational linguistics
. It studies the problems of automated generation and understanding of natural human languages
Natural-language-generation systems convert information from computer databases into normal-sounding human language. Natural-language-understanding systems convert samples of human language into more formal representations that are easier for computer programs to manipulate.
Tasks and limitations
In theory, natural-language processing is a very attractive method of human-computer interaction
. Early systems such as SHRDLU
, working in restricted "blocks worlds
" with restricted vocabularies, worked extremely well, leading researchers to excessive optimism, which was soon lost when the systems were extended to more realistic situations with real-world ambiguity
Natural-language understanding is sometimes referred to as an AI-complete problem, because natural-language recognition seems to require extensive knowledge about the outside world and the ability to manipulate it. The definition of "understanding" is one of the major problems in natural-language processing.
Some examples of the problems faced by natural-language-understanding systems:
- The sentences We gave the monkeys the bananas because they were hungry and We gave the monkeys the bananas because they were over-ripe have the same surface grammatical structure. However, the pronoun they refers to monkeys in one sentence and bananas in the other, and it is impossible to tell which without a knowledge of the properties of monkeys and bananas.
- A string of words may be interpreted in different ways. For example, the string Time flies like an arrow may be interpreted in a variety of ways:
- The common simile: time moves quickly just like an arrow does;
- measure the speed of flies like you would measure that of an arrow (thus interpreted as an imperative) - i.e. (You should) time flies as you would (time) an arrow.;
- measure the speed of flies like an arrow would - i.e. Time flies in the same way that an arrow would (time them).;
- measure the speed of flies that are like arrows - i.e. Time those flies that are like arrows;
- all of a type of flying insect, "time-flies," collectively enjoys a single arrow (compare Fruit flies like a banana);
- each of a type of flying insect, "time-flies," individually enjoys a different arrow (similar comparison applies);
- A concrete object, for example the magazine, Time, travels through the air in an arrow-like manner.
English is particularly challenging in this regard because it has little inflectional morphology to distinguish between parts of speech.
- English and several other languages don't specify which word an adjective applies to. For example, in the string "pretty little girls' school".
- Does the school look little?
- Do the girls look little?
- Do the girls look pretty?
- Does the school look pretty?
- We will often imply additional information in spoken language by the way we place stress on words. The sentence "I never said she stole my money" demonstrates the importance stress can play in a sentence, and thus the inherent difficulty a natural language processor can have in parsing it. Depending on which word the speaker places the stress, this sentence could have several distinct meanings:
- "I never said she stole my money" - Someone else said it, but I didn't.
- "I never said she stole my money" - I simply didn't ever say it.
- "I never said she stole my money" - I might have implied it in some way, but I never explicitly said it.
- "I never said she stole my money" - I said someone took it; I didn't say it was she.
- "I never said she stole my money" - I just said she probably borrowed it.
- "I never said she stole my money" - I said she stole someone else's money.
- "I never said she stole my money" - I said she stole something, but not my money.
Subproblems Speech segmentation
: In most spoken languages, the sounds representing successive letters blend into each other, so the conversion of the analog signal to discrete characters can be a very difficult process. Also, in natural speech
there are hardly any pauses between successive words; the location of those boundaries usually must take into account grammatical
constraints, as well as the context
. Text segmentation
: Some written languages like Chinese
do not have single-word boundaries either, so any significant text parsing
usually requires the identification of word boundaries, which is often a non-trivial task. Word sense disambiguation
: Many words have more than one meaning
; we have to select the meaning which makes the most sense in context. Syntactic ambiguity
: The grammar
for natural languages
, i.e. there are often multiple possible parse trees
for a given sentence. Choosing the most appropriate one usually requires semantic
and contextual information. Specific problem components of syntactic ambiguity include sentence boundary disambiguation
. Imperfect or irregular input : Foreign or regional accents and vocal impediments in speech; typing or grammatical errors, OCR
errors in texts. Speech acts
and plans: A sentence can often be considered an action by the speaker. The sentence structure, alone, may not contain enough information to define this action. For instance, a question is actually the speaker requesting some sort of response from the listener. The desired response may be verbal, physical, or some combination. For example, "Can you pass the class?" is a request for a simple yes-or-no answer, while "Can you pass the salt?" is requesting a physical action to be performed. It is not appropriate to respond with "Yes, I can pass the salt," without the accompanying action (although "No" or "I can't reach the salt" would explain a lack of action).
Statistical natural-language processing uses stochastic
methods to resolve some of the difficulties discussed above, especially those which arise because longer sentences are highly ambiguous when processed with realistic grammars, yielding thousands or millions of possible analyses. Methods for disambiguation often involve the use of corpora
and Markov models
. Statistical NLP comprises all quantitative approaches to automated language processing, including probabilistic modeling, information theory
, and linear algebra
technology for statistical NLP comes mainly from machine learning
and data mining
, both of which are fields of artificial intelligence
that involve learning from data.
Major tasks in NLP
Evaluation of natural language processing
The goal of NLP evaluation is to measure one or more qualities
of an algorithm or a system, in order to determine if (or to what extent) the system answers the goals of its designers, or the needs of its users. Research in NLP evaluation has received considerable attention, because the definition of proper evaluation criteria is one way to specify precisely an NLP problem, going thus beyond the vagueness of tasks defined only as language understanding
or language generation
. A precise set of evaluation criteria, which includes mainly evaluation data and evaluation metrics, enables several teams to compare their solutions to a given NLP problem.
Short history of evaluation in NLP
The first evaluation campaign on written texts seems to be a campaign dedicated to message understanding in 1987 (Pallet 1998). Then, the Parseval/GEIG project compared phrase-structure grammars (Black 1991). A series of campaigns within Tipster project were realized on tasks like summarization, translation and searching (Hirshman 1998). In 1994, in Germany, the Morpholympics compared German taggers. Then, the Senseval and Romanseval campaigns were conducted with the objectives of semantic disambiguation. In 1996, the Sparkle campaign compared syntactic parsers in four different languages (English, French, German and Italian). In France, the Grace project compared a set of 21 taggers for French in 1997 (Adda 1999). In 2004, during the Technolangue/Easy
project, 13 parsers for French were compared. Large-scale evaluation of dependency parsers were performed in the context of the CoNLL shared tasks in 2006 and 2007. In Italy, the evalita campaign was conducted in 2007 to compare various tools for Italian evalita web site
In France, within the ANR-Passage project (end of 2007), 10 parsers for French were compared passage web site
Adda G., Mariani J., Paroubek P., Rajman M. 1999 L'action GRACE d'évaluation de l'assignation des parties du discours pour le français. Langues vol-2
Black E., Abney S., Flickinger D., Gdaniec C., Grishman R., Harrison P., Hindle D., Ingria R., Jelinek F., Klavans J., Liberman M., Marcus M., Reukos S., Santoni B., Strzalkowski T. 1991 A procedure for quantitatively comparing the syntactic coverage of English grammars. DARPA Speech and Natural Language Workshop
Hirshman L. 1998 Language understanding evaluation: lessons learned from MUC and ATIS. LREC Granada
Pallet D.S. 1998 The NIST role in automatic speech recognition benchmark tests. LREC Granada
Different types of evaluation
Depending on the evaluation procedures, a number of distinctions are traditionally made in NLP evaluation.
- Intrinsic vs. extrinsic evaluation
Intrinsic evaluation considers an isolated NLP system and characterizes its performance mainly with respect to a gold standard result, pre-defined by the evaluators. Extrinsic evaluation, also called evaluation in use considers the NLP system in a more complex setting, either as an embedded system or serving a precise function for a human user. The extrinsic performance of the system is then characterized in terms of its utility with respect to the overall task of the complex system or the human user.
- Black-box vs. glass-box evaluation
Black-box evaluation requires one to run an NLP system on a given data set and to measure a number of parameters related to the quality of the process (speed, reliability, resource consumption) and, most importantly, to the quality of the result (e.g. the accuracy of data annotation or the fidelity of a translation). Glass-box evaluation looks at the design of the system, the algorithms that are implemented, the linguistic resources it uses (e.g. vocabulary size), etc. Given the complexity of NLP problems, it is often difficult to predict performance only on the basis of glass-box evaluation, but this type of evaluation is more informative with respect to error analysis or future developments of a system.
- Automatic vs. manual evaluation
In many cases, automatic procedures can be defined to evaluate an NLP system by comparing its output with the gold standard (or desired) one. Although the cost of producing the gold standard can be quite high, automatic evaluation can be repeated as often as needed without much additional costs (on the same input data). However, for many NLP problems, the definition of a gold standard is a complex task, and can prove impossible when inter-annotator agreement is insufficient. Manual evaluation is performed by human judges, which are instructed to estimate the quality of a system, or most often of a sample of its output, based on a number of criteria. Although, thanks to their linguistic competence, human judges can be considered as the reference for a number of language processing tasks, there is also considerable variation across their ratings. This is why automatic evaluation is sometimes referred to as objective evaluation, while the human kind appears to be more subjective.
Shared tasks (Campaigns)
Standardization in NLP
An ISO sub-committee is working in order to ease interoperability between Lexical resources
and NLP programs. The sub-committee is part of ISO/TC37
and is called ISO/TC37/SC4. Some ISO standards are already published but most of them are under construction, mainly on lexicon representation (see LMF
), annotation and data category registry.
Organizations and conferences
Related academic articles
- Bates, M. (1995). Models of natural language understanding. Proceedings of the National Academy of Sciences of the United States of America, Vol. 92, No. 22 (Oct. 24, 1995), pp. 9977-9982.